xref: /freebsd/sys/geom/raid/g_raid.c (revision 145992504973bd16cf3518af9ba5ce185fefa82a)
1 /*-
2  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/limits.h>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/bio.h>
38 #include <sys/sbuf.h>
39 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
41 #include <sys/eventhandler.h>
42 #include <vm/uma.h>
43 #include <geom/geom.h>
44 #include <sys/proc.h>
45 #include <sys/kthread.h>
46 #include <sys/sched.h>
47 #include <geom/raid/g_raid.h>
48 #include "g_raid_md_if.h"
49 #include "g_raid_tr_if.h"
50 
51 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
52 
53 SYSCTL_DECL(_kern_geom);
54 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
55 int g_raid_enable = 1;
56 TUNABLE_INT("kern.geom.raid.enable", &g_raid_enable);
57 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RW,
58     &g_raid_enable, 0, "Enable on-disk metadata taste");
59 u_int g_raid_aggressive_spare = 0;
60 TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare);
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW,
62     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug);
65 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0,
66     "Debug level");
67 int g_raid_read_err_thresh = 10;
68 TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh);
69 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW,
70     &g_raid_read_err_thresh, 0,
71     "Number of read errors equated to disk failure");
72 u_int g_raid_start_timeout = 30;
73 TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout);
74 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW,
75     &g_raid_start_timeout, 0,
76     "Time to wait for all array components");
77 static u_int g_raid_clean_time = 5;
78 TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time);
79 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW,
80     &g_raid_clean_time, 0, "Mark volume as clean when idling");
81 static u_int g_raid_disconnect_on_failure = 1;
82 TUNABLE_INT("kern.geom.raid.disconnect_on_failure",
83     &g_raid_disconnect_on_failure);
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW,
85     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
86 static u_int g_raid_name_format = 0;
87 TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format);
88 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW,
89     &g_raid_name_format, 0, "Providers name format.");
90 static u_int g_raid_idle_threshold = 1000000;
91 TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold);
92 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW,
93     &g_raid_idle_threshold, 1000000,
94     "Time in microseconds to consider a volume idle.");
95 
96 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
97 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
98 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
99 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
100 } while (0)
101 
102 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
103     LIST_HEAD_INITIALIZER(g_raid_md_classes);
104 
105 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
106     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
107 
108 LIST_HEAD(, g_raid_volume) g_raid_volumes =
109     LIST_HEAD_INITIALIZER(g_raid_volumes);
110 
111 static eventhandler_tag g_raid_pre_sync = NULL;
112 static int g_raid_started = 0;
113 
114 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
115     struct g_geom *gp);
116 static g_taste_t g_raid_taste;
117 static void g_raid_init(struct g_class *mp);
118 static void g_raid_fini(struct g_class *mp);
119 
120 struct g_class g_raid_class = {
121 	.name = G_RAID_CLASS_NAME,
122 	.version = G_VERSION,
123 	.ctlreq = g_raid_ctl,
124 	.taste = g_raid_taste,
125 	.destroy_geom = g_raid_destroy_geom,
126 	.init = g_raid_init,
127 	.fini = g_raid_fini
128 };
129 
130 static void g_raid_destroy_provider(struct g_raid_volume *vol);
131 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
132 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
133 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
134 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
135 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
136     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
137 static void g_raid_start(struct bio *bp);
138 static void g_raid_start_request(struct bio *bp);
139 static void g_raid_disk_done(struct bio *bp);
140 static void g_raid_poll(struct g_raid_softc *sc);
141 
142 static const char *
143 g_raid_node_event2str(int event)
144 {
145 
146 	switch (event) {
147 	case G_RAID_NODE_E_WAKE:
148 		return ("WAKE");
149 	case G_RAID_NODE_E_START:
150 		return ("START");
151 	default:
152 		return ("INVALID");
153 	}
154 }
155 
156 const char *
157 g_raid_disk_state2str(int state)
158 {
159 
160 	switch (state) {
161 	case G_RAID_DISK_S_NONE:
162 		return ("NONE");
163 	case G_RAID_DISK_S_OFFLINE:
164 		return ("OFFLINE");
165 	case G_RAID_DISK_S_FAILED:
166 		return ("FAILED");
167 	case G_RAID_DISK_S_STALE_FAILED:
168 		return ("STALE_FAILED");
169 	case G_RAID_DISK_S_SPARE:
170 		return ("SPARE");
171 	case G_RAID_DISK_S_STALE:
172 		return ("STALE");
173 	case G_RAID_DISK_S_ACTIVE:
174 		return ("ACTIVE");
175 	default:
176 		return ("INVALID");
177 	}
178 }
179 
180 static const char *
181 g_raid_disk_event2str(int event)
182 {
183 
184 	switch (event) {
185 	case G_RAID_DISK_E_DISCONNECTED:
186 		return ("DISCONNECTED");
187 	default:
188 		return ("INVALID");
189 	}
190 }
191 
192 const char *
193 g_raid_subdisk_state2str(int state)
194 {
195 
196 	switch (state) {
197 	case G_RAID_SUBDISK_S_NONE:
198 		return ("NONE");
199 	case G_RAID_SUBDISK_S_FAILED:
200 		return ("FAILED");
201 	case G_RAID_SUBDISK_S_NEW:
202 		return ("NEW");
203 	case G_RAID_SUBDISK_S_REBUILD:
204 		return ("REBUILD");
205 	case G_RAID_SUBDISK_S_UNINITIALIZED:
206 		return ("UNINITIALIZED");
207 	case G_RAID_SUBDISK_S_STALE:
208 		return ("STALE");
209 	case G_RAID_SUBDISK_S_RESYNC:
210 		return ("RESYNC");
211 	case G_RAID_SUBDISK_S_ACTIVE:
212 		return ("ACTIVE");
213 	default:
214 		return ("INVALID");
215 	}
216 }
217 
218 static const char *
219 g_raid_subdisk_event2str(int event)
220 {
221 
222 	switch (event) {
223 	case G_RAID_SUBDISK_E_NEW:
224 		return ("NEW");
225 	case G_RAID_SUBDISK_E_FAILED:
226 		return ("FAILED");
227 	case G_RAID_SUBDISK_E_DISCONNECTED:
228 		return ("DISCONNECTED");
229 	default:
230 		return ("INVALID");
231 	}
232 }
233 
234 const char *
235 g_raid_volume_state2str(int state)
236 {
237 
238 	switch (state) {
239 	case G_RAID_VOLUME_S_STARTING:
240 		return ("STARTING");
241 	case G_RAID_VOLUME_S_BROKEN:
242 		return ("BROKEN");
243 	case G_RAID_VOLUME_S_DEGRADED:
244 		return ("DEGRADED");
245 	case G_RAID_VOLUME_S_SUBOPTIMAL:
246 		return ("SUBOPTIMAL");
247 	case G_RAID_VOLUME_S_OPTIMAL:
248 		return ("OPTIMAL");
249 	case G_RAID_VOLUME_S_UNSUPPORTED:
250 		return ("UNSUPPORTED");
251 	case G_RAID_VOLUME_S_STOPPED:
252 		return ("STOPPED");
253 	default:
254 		return ("INVALID");
255 	}
256 }
257 
258 static const char *
259 g_raid_volume_event2str(int event)
260 {
261 
262 	switch (event) {
263 	case G_RAID_VOLUME_E_UP:
264 		return ("UP");
265 	case G_RAID_VOLUME_E_DOWN:
266 		return ("DOWN");
267 	case G_RAID_VOLUME_E_START:
268 		return ("START");
269 	case G_RAID_VOLUME_E_STARTMD:
270 		return ("STARTMD");
271 	default:
272 		return ("INVALID");
273 	}
274 }
275 
276 const char *
277 g_raid_volume_level2str(int level, int qual)
278 {
279 
280 	switch (level) {
281 	case G_RAID_VOLUME_RL_RAID0:
282 		return ("RAID0");
283 	case G_RAID_VOLUME_RL_RAID1:
284 		return ("RAID1");
285 	case G_RAID_VOLUME_RL_RAID3:
286 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
287 			return ("RAID3-P0");
288 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
289 			return ("RAID3-PN");
290 		return ("RAID3");
291 	case G_RAID_VOLUME_RL_RAID4:
292 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
293 			return ("RAID4-P0");
294 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
295 			return ("RAID4-PN");
296 		return ("RAID4");
297 	case G_RAID_VOLUME_RL_RAID5:
298 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
299 			return ("RAID5-RA");
300 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
301 			return ("RAID5-RS");
302 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
303 			return ("RAID5-LA");
304 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
305 			return ("RAID5-LS");
306 		return ("RAID5");
307 	case G_RAID_VOLUME_RL_RAID6:
308 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
309 			return ("RAID6-RA");
310 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
311 			return ("RAID6-RS");
312 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
313 			return ("RAID6-LA");
314 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
315 			return ("RAID6-LS");
316 		return ("RAID6");
317 	case G_RAID_VOLUME_RL_RAIDMDF:
318 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
319 			return ("RAIDMDF-RA");
320 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
321 			return ("RAIDMDF-RS");
322 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
323 			return ("RAIDMDF-LA");
324 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
325 			return ("RAIDMDF-LS");
326 		return ("RAIDMDF");
327 	case G_RAID_VOLUME_RL_RAID1E:
328 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
329 			return ("RAID1E-A");
330 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
331 			return ("RAID1E-O");
332 		return ("RAID1E");
333 	case G_RAID_VOLUME_RL_SINGLE:
334 		return ("SINGLE");
335 	case G_RAID_VOLUME_RL_CONCAT:
336 		return ("CONCAT");
337 	case G_RAID_VOLUME_RL_RAID5E:
338 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
339 			return ("RAID5E-RA");
340 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
341 			return ("RAID5E-RS");
342 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
343 			return ("RAID5E-LA");
344 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
345 			return ("RAID5E-LS");
346 		return ("RAID5E");
347 	case G_RAID_VOLUME_RL_RAID5EE:
348 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
349 			return ("RAID5EE-RA");
350 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
351 			return ("RAID5EE-RS");
352 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
353 			return ("RAID5EE-LA");
354 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
355 			return ("RAID5EE-LS");
356 		return ("RAID5EE");
357 	case G_RAID_VOLUME_RL_RAID5R:
358 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
359 			return ("RAID5R-RA");
360 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
361 			return ("RAID5R-RS");
362 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
363 			return ("RAID5R-LA");
364 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
365 			return ("RAID5R-LS");
366 		return ("RAID5E");
367 	default:
368 		return ("UNKNOWN");
369 	}
370 }
371 
372 int
373 g_raid_volume_str2level(const char *str, int *level, int *qual)
374 {
375 
376 	*level = G_RAID_VOLUME_RL_UNKNOWN;
377 	*qual = G_RAID_VOLUME_RLQ_NONE;
378 	if (strcasecmp(str, "RAID0") == 0)
379 		*level = G_RAID_VOLUME_RL_RAID0;
380 	else if (strcasecmp(str, "RAID1") == 0)
381 		*level = G_RAID_VOLUME_RL_RAID1;
382 	else if (strcasecmp(str, "RAID3-P0") == 0) {
383 		*level = G_RAID_VOLUME_RL_RAID3;
384 		*qual = G_RAID_VOLUME_RLQ_R3P0;
385 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
386 		   strcasecmp(str, "RAID3") == 0) {
387 		*level = G_RAID_VOLUME_RL_RAID3;
388 		*qual = G_RAID_VOLUME_RLQ_R3PN;
389 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
390 		*level = G_RAID_VOLUME_RL_RAID4;
391 		*qual = G_RAID_VOLUME_RLQ_R4P0;
392 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
393 		   strcasecmp(str, "RAID4") == 0) {
394 		*level = G_RAID_VOLUME_RL_RAID4;
395 		*qual = G_RAID_VOLUME_RLQ_R4PN;
396 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
397 		*level = G_RAID_VOLUME_RL_RAID5;
398 		*qual = G_RAID_VOLUME_RLQ_R5RA;
399 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
400 		*level = G_RAID_VOLUME_RL_RAID5;
401 		*qual = G_RAID_VOLUME_RLQ_R5RS;
402 	} else if (strcasecmp(str, "RAID5") == 0 ||
403 		   strcasecmp(str, "RAID5-LA") == 0) {
404 		*level = G_RAID_VOLUME_RL_RAID5;
405 		*qual = G_RAID_VOLUME_RLQ_R5LA;
406 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
407 		*level = G_RAID_VOLUME_RL_RAID5;
408 		*qual = G_RAID_VOLUME_RLQ_R5LS;
409 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
410 		*level = G_RAID_VOLUME_RL_RAID6;
411 		*qual = G_RAID_VOLUME_RLQ_R6RA;
412 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
413 		*level = G_RAID_VOLUME_RL_RAID6;
414 		*qual = G_RAID_VOLUME_RLQ_R6RS;
415 	} else if (strcasecmp(str, "RAID6") == 0 ||
416 		   strcasecmp(str, "RAID6-LA") == 0) {
417 		*level = G_RAID_VOLUME_RL_RAID6;
418 		*qual = G_RAID_VOLUME_RLQ_R6LA;
419 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
420 		*level = G_RAID_VOLUME_RL_RAID6;
421 		*qual = G_RAID_VOLUME_RLQ_R6LS;
422 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
423 		*level = G_RAID_VOLUME_RL_RAIDMDF;
424 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
425 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
426 		*level = G_RAID_VOLUME_RL_RAIDMDF;
427 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
428 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
429 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
430 		*level = G_RAID_VOLUME_RL_RAIDMDF;
431 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
432 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
433 		*level = G_RAID_VOLUME_RL_RAIDMDF;
434 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
435 	} else if (strcasecmp(str, "RAID10") == 0 ||
436 		   strcasecmp(str, "RAID1E") == 0 ||
437 		   strcasecmp(str, "RAID1E-A") == 0) {
438 		*level = G_RAID_VOLUME_RL_RAID1E;
439 		*qual = G_RAID_VOLUME_RLQ_R1EA;
440 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
441 		*level = G_RAID_VOLUME_RL_RAID1E;
442 		*qual = G_RAID_VOLUME_RLQ_R1EO;
443 	} else if (strcasecmp(str, "SINGLE") == 0)
444 		*level = G_RAID_VOLUME_RL_SINGLE;
445 	else if (strcasecmp(str, "CONCAT") == 0)
446 		*level = G_RAID_VOLUME_RL_CONCAT;
447 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
448 		*level = G_RAID_VOLUME_RL_RAID5E;
449 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
450 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
451 		*level = G_RAID_VOLUME_RL_RAID5E;
452 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
453 	} else if (strcasecmp(str, "RAID5E") == 0 ||
454 		   strcasecmp(str, "RAID5E-LA") == 0) {
455 		*level = G_RAID_VOLUME_RL_RAID5E;
456 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
457 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
458 		*level = G_RAID_VOLUME_RL_RAID5E;
459 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
460 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
461 		*level = G_RAID_VOLUME_RL_RAID5EE;
462 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
463 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
464 		*level = G_RAID_VOLUME_RL_RAID5EE;
465 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
466 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
467 		   strcasecmp(str, "RAID5EE-LA") == 0) {
468 		*level = G_RAID_VOLUME_RL_RAID5EE;
469 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
470 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
471 		*level = G_RAID_VOLUME_RL_RAID5EE;
472 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
473 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
474 		*level = G_RAID_VOLUME_RL_RAID5R;
475 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
476 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
477 		*level = G_RAID_VOLUME_RL_RAID5R;
478 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
479 	} else if (strcasecmp(str, "RAID5R") == 0 ||
480 		   strcasecmp(str, "RAID5R-LA") == 0) {
481 		*level = G_RAID_VOLUME_RL_RAID5R;
482 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
483 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
484 		*level = G_RAID_VOLUME_RL_RAID5R;
485 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
486 	} else
487 		return (-1);
488 	return (0);
489 }
490 
491 const char *
492 g_raid_get_diskname(struct g_raid_disk *disk)
493 {
494 
495 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
496 		return ("[unknown]");
497 	return (disk->d_consumer->provider->name);
498 }
499 
500 void
501 g_raid_report_disk_state(struct g_raid_disk *disk)
502 {
503 	struct g_raid_subdisk *sd;
504 	int len, state;
505 	uint32_t s;
506 
507 	if (disk->d_consumer == NULL)
508 		return;
509 	if (disk->d_state == G_RAID_DISK_S_FAILED ||
510 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
511 		s = G_STATE_FAILED;
512 	} else {
513 		state = G_RAID_SUBDISK_S_ACTIVE;
514 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
515 			if (sd->sd_state < state)
516 				state = sd->sd_state;
517 		}
518 		if (state == G_RAID_SUBDISK_S_FAILED)
519 			s = G_STATE_FAILED;
520 		else if (state == G_RAID_SUBDISK_S_NEW ||
521 		    state == G_RAID_SUBDISK_S_REBUILD)
522 			s = G_STATE_REBUILD;
523 		else if (state == G_RAID_SUBDISK_S_STALE ||
524 		    state == G_RAID_SUBDISK_S_RESYNC)
525 			s = G_STATE_RESYNC;
526 		else
527 			s = G_STATE_ACTIVE;
528 	}
529 	len = sizeof(s);
530 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
531 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
532 	    g_raid_get_diskname(disk), s);
533 }
534 
535 void
536 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
537 {
538 
539 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
540 	    g_raid_get_diskname(disk),
541 	    g_raid_disk_state2str(disk->d_state),
542 	    g_raid_disk_state2str(state));
543 	disk->d_state = state;
544 	g_raid_report_disk_state(disk);
545 }
546 
547 void
548 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
549 {
550 
551 	G_RAID_DEBUG1(0, sd->sd_softc,
552 	    "Subdisk %s:%d-%s state changed from %s to %s.",
553 	    sd->sd_volume->v_name, sd->sd_pos,
554 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
555 	    g_raid_subdisk_state2str(sd->sd_state),
556 	    g_raid_subdisk_state2str(state));
557 	sd->sd_state = state;
558 	if (sd->sd_disk)
559 		g_raid_report_disk_state(sd->sd_disk);
560 }
561 
562 void
563 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
564 {
565 
566 	G_RAID_DEBUG1(0, vol->v_softc,
567 	    "Volume %s state changed from %s to %s.",
568 	    vol->v_name,
569 	    g_raid_volume_state2str(vol->v_state),
570 	    g_raid_volume_state2str(state));
571 	vol->v_state = state;
572 }
573 
574 /*
575  * --- Events handling functions ---
576  * Events in geom_raid are used to maintain subdisks and volumes status
577  * from one thread to simplify locking.
578  */
579 static void
580 g_raid_event_free(struct g_raid_event *ep)
581 {
582 
583 	free(ep, M_RAID);
584 }
585 
586 int
587 g_raid_event_send(void *arg, int event, int flags)
588 {
589 	struct g_raid_softc *sc;
590 	struct g_raid_event *ep;
591 	int error;
592 
593 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
594 		sc = ((struct g_raid_volume *)arg)->v_softc;
595 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
596 		sc = ((struct g_raid_disk *)arg)->d_softc;
597 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
598 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
599 	} else {
600 		sc = arg;
601 	}
602 	ep = malloc(sizeof(*ep), M_RAID,
603 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
604 	if (ep == NULL)
605 		return (ENOMEM);
606 	ep->e_tgt = arg;
607 	ep->e_event = event;
608 	ep->e_flags = flags;
609 	ep->e_error = 0;
610 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
611 	mtx_lock(&sc->sc_queue_mtx);
612 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
613 	mtx_unlock(&sc->sc_queue_mtx);
614 	wakeup(sc);
615 
616 	if ((flags & G_RAID_EVENT_WAIT) == 0)
617 		return (0);
618 
619 	sx_assert(&sc->sc_lock, SX_XLOCKED);
620 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
621 	sx_xunlock(&sc->sc_lock);
622 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
623 		mtx_lock(&sc->sc_queue_mtx);
624 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
625 		    hz * 5);
626 	}
627 	error = ep->e_error;
628 	g_raid_event_free(ep);
629 	sx_xlock(&sc->sc_lock);
630 	return (error);
631 }
632 
633 static void
634 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
635 {
636 	struct g_raid_event *ep, *tmpep;
637 
638 	sx_assert(&sc->sc_lock, SX_XLOCKED);
639 
640 	mtx_lock(&sc->sc_queue_mtx);
641 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
642 		if (ep->e_tgt != tgt)
643 			continue;
644 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
645 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
646 			g_raid_event_free(ep);
647 		else {
648 			ep->e_error = ECANCELED;
649 			wakeup(ep);
650 		}
651 	}
652 	mtx_unlock(&sc->sc_queue_mtx);
653 }
654 
655 static int
656 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
657 {
658 	struct g_raid_event *ep;
659 	int	res = 0;
660 
661 	sx_assert(&sc->sc_lock, SX_XLOCKED);
662 
663 	mtx_lock(&sc->sc_queue_mtx);
664 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
665 		if (ep->e_tgt != tgt)
666 			continue;
667 		res = 1;
668 		break;
669 	}
670 	mtx_unlock(&sc->sc_queue_mtx);
671 	return (res);
672 }
673 
674 /*
675  * Return the number of disks in given state.
676  * If state is equal to -1, count all connected disks.
677  */
678 u_int
679 g_raid_ndisks(struct g_raid_softc *sc, int state)
680 {
681 	struct g_raid_disk *disk;
682 	u_int n;
683 
684 	sx_assert(&sc->sc_lock, SX_LOCKED);
685 
686 	n = 0;
687 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
688 		if (disk->d_state == state || state == -1)
689 			n++;
690 	}
691 	return (n);
692 }
693 
694 /*
695  * Return the number of subdisks in given state.
696  * If state is equal to -1, count all connected disks.
697  */
698 u_int
699 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
700 {
701 	struct g_raid_subdisk *subdisk;
702 	struct g_raid_softc *sc;
703 	u_int i, n ;
704 
705 	sc = vol->v_softc;
706 	sx_assert(&sc->sc_lock, SX_LOCKED);
707 
708 	n = 0;
709 	for (i = 0; i < vol->v_disks_count; i++) {
710 		subdisk = &vol->v_subdisks[i];
711 		if ((state == -1 &&
712 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
713 		    subdisk->sd_state == state)
714 			n++;
715 	}
716 	return (n);
717 }
718 
719 /*
720  * Return the first subdisk in given state.
721  * If state is equal to -1, then the first connected disks.
722  */
723 struct g_raid_subdisk *
724 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
725 {
726 	struct g_raid_subdisk *sd;
727 	struct g_raid_softc *sc;
728 	u_int i;
729 
730 	sc = vol->v_softc;
731 	sx_assert(&sc->sc_lock, SX_LOCKED);
732 
733 	for (i = 0; i < vol->v_disks_count; i++) {
734 		sd = &vol->v_subdisks[i];
735 		if ((state == -1 &&
736 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
737 		    sd->sd_state == state)
738 			return (sd);
739 	}
740 	return (NULL);
741 }
742 
743 struct g_consumer *
744 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
745 {
746 	struct g_consumer *cp;
747 	struct g_provider *pp;
748 
749 	g_topology_assert();
750 
751 	if (strncmp(name, "/dev/", 5) == 0)
752 		name += 5;
753 	pp = g_provider_by_name(name);
754 	if (pp == NULL)
755 		return (NULL);
756 	cp = g_new_consumer(sc->sc_geom);
757 	if (g_attach(cp, pp) != 0) {
758 		g_destroy_consumer(cp);
759 		return (NULL);
760 	}
761 	if (g_access(cp, 1, 1, 1) != 0) {
762 		g_detach(cp);
763 		g_destroy_consumer(cp);
764 		return (NULL);
765 	}
766 	return (cp);
767 }
768 
769 static u_int
770 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
771 {
772 	struct bio *bp;
773 	u_int nreqs = 0;
774 
775 	mtx_lock(&sc->sc_queue_mtx);
776 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
777 		if (bp->bio_from == cp)
778 			nreqs++;
779 	}
780 	mtx_unlock(&sc->sc_queue_mtx);
781 	return (nreqs);
782 }
783 
784 u_int
785 g_raid_nopens(struct g_raid_softc *sc)
786 {
787 	struct g_raid_volume *vol;
788 	u_int opens;
789 
790 	opens = 0;
791 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
792 		if (vol->v_provider_open != 0)
793 			opens++;
794 	}
795 	return (opens);
796 }
797 
798 static int
799 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
800 {
801 
802 	if (cp->index > 0) {
803 		G_RAID_DEBUG1(2, sc,
804 		    "I/O requests for %s exist, can't destroy it now.",
805 		    cp->provider->name);
806 		return (1);
807 	}
808 	if (g_raid_nrequests(sc, cp) > 0) {
809 		G_RAID_DEBUG1(2, sc,
810 		    "I/O requests for %s in queue, can't destroy it now.",
811 		    cp->provider->name);
812 		return (1);
813 	}
814 	return (0);
815 }
816 
817 static void
818 g_raid_destroy_consumer(void *arg, int flags __unused)
819 {
820 	struct g_consumer *cp;
821 
822 	g_topology_assert();
823 
824 	cp = arg;
825 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
826 	g_detach(cp);
827 	g_destroy_consumer(cp);
828 }
829 
830 void
831 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
832 {
833 	struct g_provider *pp;
834 	int retaste_wait;
835 
836 	g_topology_assert_not();
837 
838 	g_topology_lock();
839 	cp->private = NULL;
840 	if (g_raid_consumer_is_busy(sc, cp))
841 		goto out;
842 	pp = cp->provider;
843 	retaste_wait = 0;
844 	if (cp->acw == 1) {
845 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
846 			retaste_wait = 1;
847 	}
848 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
849 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
850 	if (retaste_wait) {
851 		/*
852 		 * After retaste event was send (inside g_access()), we can send
853 		 * event to detach and destroy consumer.
854 		 * A class, which has consumer to the given provider connected
855 		 * will not receive retaste event for the provider.
856 		 * This is the way how I ignore retaste events when I close
857 		 * consumers opened for write: I detach and destroy consumer
858 		 * after retaste event is sent.
859 		 */
860 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
861 		goto out;
862 	}
863 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
864 	g_detach(cp);
865 	g_destroy_consumer(cp);
866 out:
867 	g_topology_unlock();
868 }
869 
870 static void
871 g_raid_orphan(struct g_consumer *cp)
872 {
873 	struct g_raid_disk *disk;
874 
875 	g_topology_assert();
876 
877 	disk = cp->private;
878 	if (disk == NULL)
879 		return;
880 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
881 	    G_RAID_EVENT_DISK);
882 }
883 
884 static int
885 g_raid_clean(struct g_raid_volume *vol, int acw)
886 {
887 	struct g_raid_softc *sc;
888 	int timeout;
889 
890 	sc = vol->v_softc;
891 	g_topology_assert_not();
892 	sx_assert(&sc->sc_lock, SX_XLOCKED);
893 
894 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
895 //		return (0);
896 	if (!vol->v_dirty)
897 		return (0);
898 	if (vol->v_writes > 0)
899 		return (0);
900 	if (acw > 0 || (acw == -1 &&
901 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
902 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
903 		if (timeout > 0)
904 			return (timeout);
905 	}
906 	vol->v_dirty = 0;
907 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
908 	    vol->v_name);
909 	g_raid_write_metadata(sc, vol, NULL, NULL);
910 	return (0);
911 }
912 
913 static void
914 g_raid_dirty(struct g_raid_volume *vol)
915 {
916 	struct g_raid_softc *sc;
917 
918 	sc = vol->v_softc;
919 	g_topology_assert_not();
920 	sx_assert(&sc->sc_lock, SX_XLOCKED);
921 
922 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
923 //		return;
924 	vol->v_dirty = 1;
925 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
926 	    vol->v_name);
927 	g_raid_write_metadata(sc, vol, NULL, NULL);
928 }
929 
930 void
931 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
932 {
933 	struct g_raid_softc *sc;
934 	struct g_raid_volume *vol;
935 	struct g_raid_subdisk *sd;
936 	struct bio_queue_head queue;
937 	struct bio *cbp;
938 	int i;
939 
940 	vol = tr->tro_volume;
941 	sc = vol->v_softc;
942 
943 	/*
944 	 * Allocate all bios before sending any request, so we can return
945 	 * ENOMEM in nice and clean way.
946 	 */
947 	bioq_init(&queue);
948 	for (i = 0; i < vol->v_disks_count; i++) {
949 		sd = &vol->v_subdisks[i];
950 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
951 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
952 			continue;
953 		cbp = g_clone_bio(bp);
954 		if (cbp == NULL)
955 			goto failure;
956 		cbp->bio_caller1 = sd;
957 		bioq_insert_tail(&queue, cbp);
958 	}
959 	for (cbp = bioq_first(&queue); cbp != NULL;
960 	    cbp = bioq_first(&queue)) {
961 		bioq_remove(&queue, cbp);
962 		sd = cbp->bio_caller1;
963 		cbp->bio_caller1 = NULL;
964 		g_raid_subdisk_iostart(sd, cbp);
965 	}
966 	return;
967 failure:
968 	for (cbp = bioq_first(&queue); cbp != NULL;
969 	    cbp = bioq_first(&queue)) {
970 		bioq_remove(&queue, cbp);
971 		g_destroy_bio(cbp);
972 	}
973 	if (bp->bio_error == 0)
974 		bp->bio_error = ENOMEM;
975 	g_raid_iodone(bp, bp->bio_error);
976 }
977 
978 static void
979 g_raid_tr_kerneldump_common_done(struct bio *bp)
980 {
981 
982 	bp->bio_flags |= BIO_DONE;
983 }
984 
985 int
986 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
987     void *virtual, vm_offset_t physical, off_t offset, size_t length)
988 {
989 	struct g_raid_softc *sc;
990 	struct g_raid_volume *vol;
991 	struct bio bp;
992 
993 	vol = tr->tro_volume;
994 	sc = vol->v_softc;
995 
996 	bzero(&bp, sizeof(bp));
997 	bp.bio_cmd = BIO_WRITE;
998 	bp.bio_done = g_raid_tr_kerneldump_common_done;
999 	bp.bio_attribute = NULL;
1000 	bp.bio_offset = offset;
1001 	bp.bio_length = length;
1002 	bp.bio_data = virtual;
1003 	bp.bio_to = vol->v_provider;
1004 
1005 	g_raid_start(&bp);
1006 	while (!(bp.bio_flags & BIO_DONE)) {
1007 		G_RAID_DEBUG1(4, sc, "Poll...");
1008 		g_raid_poll(sc);
1009 		DELAY(10);
1010 	}
1011 
1012 	return (bp.bio_error != 0 ? EIO : 0);
1013 }
1014 
1015 static int
1016 g_raid_dump(void *arg,
1017     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1018 {
1019 	struct g_raid_volume *vol;
1020 	int error;
1021 
1022 	vol = (struct g_raid_volume *)arg;
1023 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1024 	    (long long unsigned)offset, (long long unsigned)length);
1025 
1026 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1027 	    virtual, physical, offset, length);
1028 	return (error);
1029 }
1030 
1031 static void
1032 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1033 {
1034 	struct g_kerneldump *gkd;
1035 	struct g_provider *pp;
1036 	struct g_raid_volume *vol;
1037 
1038 	gkd = (struct g_kerneldump*)bp->bio_data;
1039 	pp = bp->bio_to;
1040 	vol = pp->private;
1041 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1042 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1043 	gkd->di.dumper = g_raid_dump;
1044 	gkd->di.priv = vol;
1045 	gkd->di.blocksize = vol->v_sectorsize;
1046 	gkd->di.maxiosize = DFLTPHYS;
1047 	gkd->di.mediaoffset = gkd->offset;
1048 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1049 		gkd->length = vol->v_mediasize - gkd->offset;
1050 	gkd->di.mediasize = gkd->length;
1051 	g_io_deliver(bp, 0);
1052 }
1053 
1054 static void
1055 g_raid_start(struct bio *bp)
1056 {
1057 	struct g_raid_softc *sc;
1058 
1059 	sc = bp->bio_to->geom->softc;
1060 	/*
1061 	 * If sc == NULL or there are no valid disks, provider's error
1062 	 * should be set and g_raid_start() should not be called at all.
1063 	 */
1064 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1065 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1066 //	    bp->bio_to->error, bp->bio_to->name));
1067 	G_RAID_LOGREQ(3, bp, "Request received.");
1068 
1069 	switch (bp->bio_cmd) {
1070 	case BIO_READ:
1071 	case BIO_WRITE:
1072 	case BIO_DELETE:
1073 	case BIO_FLUSH:
1074 		break;
1075 	case BIO_GETATTR:
1076 		if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1077 			g_raid_kerneldump(sc, bp);
1078 		else
1079 			g_io_deliver(bp, EOPNOTSUPP);
1080 		return;
1081 	default:
1082 		g_io_deliver(bp, EOPNOTSUPP);
1083 		return;
1084 	}
1085 	mtx_lock(&sc->sc_queue_mtx);
1086 	bioq_disksort(&sc->sc_queue, bp);
1087 	mtx_unlock(&sc->sc_queue_mtx);
1088 	if (!dumping) {
1089 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1090 		wakeup(sc);
1091 	}
1092 }
1093 
1094 static int
1095 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1096 {
1097 	/*
1098 	 * 5 cases:
1099 	 * (1) bp entirely below NO
1100 	 * (2) bp entirely above NO
1101 	 * (3) bp start below, but end in range YES
1102 	 * (4) bp entirely within YES
1103 	 * (5) bp starts within, ends above YES
1104 	 *
1105 	 * lock range 10-19 (offset 10 length 10)
1106 	 * (1) 1-5: first if kicks it out
1107 	 * (2) 30-35: second if kicks it out
1108 	 * (3) 5-15: passes both ifs
1109 	 * (4) 12-14: passes both ifs
1110 	 * (5) 19-20: passes both
1111 	 */
1112 	off_t lend = lstart + len - 1;
1113 	off_t bstart = bp->bio_offset;
1114 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1115 
1116 	if (bend < lstart)
1117 		return (0);
1118 	if (lend < bstart)
1119 		return (0);
1120 	return (1);
1121 }
1122 
1123 static int
1124 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1125 {
1126 	struct g_raid_lock *lp;
1127 
1128 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1129 
1130 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1131 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1132 			return (1);
1133 	}
1134 	return (0);
1135 }
1136 
1137 static void
1138 g_raid_start_request(struct bio *bp)
1139 {
1140 	struct g_raid_softc *sc;
1141 	struct g_raid_volume *vol;
1142 
1143 	sc = bp->bio_to->geom->softc;
1144 	sx_assert(&sc->sc_lock, SX_LOCKED);
1145 	vol = bp->bio_to->private;
1146 
1147 	/*
1148 	 * Check to see if this item is in a locked range.  If so,
1149 	 * queue it to our locked queue and return.  We'll requeue
1150 	 * it when the range is unlocked.  Internal I/O for the
1151 	 * rebuild/rescan/recovery process is excluded from this
1152 	 * check so we can actually do the recovery.
1153 	 */
1154 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1155 	    g_raid_is_in_locked_range(vol, bp)) {
1156 		G_RAID_LOGREQ(3, bp, "Defer request.");
1157 		bioq_insert_tail(&vol->v_locked, bp);
1158 		return;
1159 	}
1160 
1161 	/*
1162 	 * If we're actually going to do the write/delete, then
1163 	 * update the idle stats for the volume.
1164 	 */
1165 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1166 		if (!vol->v_dirty)
1167 			g_raid_dirty(vol);
1168 		vol->v_writes++;
1169 	}
1170 
1171 	/*
1172 	 * Put request onto inflight queue, so we can check if new
1173 	 * synchronization requests don't collide with it.  Then tell
1174 	 * the transformation layer to start the I/O.
1175 	 */
1176 	bioq_insert_tail(&vol->v_inflight, bp);
1177 	G_RAID_LOGREQ(4, bp, "Request started");
1178 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1179 }
1180 
1181 static void
1182 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1183 {
1184 	off_t off, len;
1185 	struct bio *nbp;
1186 	struct g_raid_lock *lp;
1187 
1188 	vol->v_pending_lock = 0;
1189 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1190 		if (lp->l_pending) {
1191 			off = lp->l_offset;
1192 			len = lp->l_length;
1193 			lp->l_pending = 0;
1194 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1195 				if (g_raid_bio_overlaps(nbp, off, len))
1196 					lp->l_pending++;
1197 			}
1198 			if (lp->l_pending) {
1199 				vol->v_pending_lock = 1;
1200 				G_RAID_DEBUG1(4, vol->v_softc,
1201 				    "Deferred lock(%jd, %jd) has %d pending",
1202 				    (intmax_t)off, (intmax_t)(off + len),
1203 				    lp->l_pending);
1204 				continue;
1205 			}
1206 			G_RAID_DEBUG1(4, vol->v_softc,
1207 			    "Deferred lock of %jd to %jd completed",
1208 			    (intmax_t)off, (intmax_t)(off + len));
1209 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1210 		}
1211 	}
1212 }
1213 
1214 void
1215 g_raid_iodone(struct bio *bp, int error)
1216 {
1217 	struct g_raid_softc *sc;
1218 	struct g_raid_volume *vol;
1219 
1220 	sc = bp->bio_to->geom->softc;
1221 	sx_assert(&sc->sc_lock, SX_LOCKED);
1222 	vol = bp->bio_to->private;
1223 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1224 
1225 	/* Update stats if we done write/delete. */
1226 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1227 		vol->v_writes--;
1228 		vol->v_last_write = time_uptime;
1229 	}
1230 
1231 	bioq_remove(&vol->v_inflight, bp);
1232 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1233 		g_raid_finish_with_locked_ranges(vol, bp);
1234 	getmicrouptime(&vol->v_last_done);
1235 	g_io_deliver(bp, error);
1236 }
1237 
1238 int
1239 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1240     struct bio *ignore, void *argp)
1241 {
1242 	struct g_raid_softc *sc;
1243 	struct g_raid_lock *lp;
1244 	struct bio *bp;
1245 
1246 	sc = vol->v_softc;
1247 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1248 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1249 	lp->l_offset = off;
1250 	lp->l_length = len;
1251 	lp->l_callback_arg = argp;
1252 
1253 	lp->l_pending = 0;
1254 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1255 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1256 			lp->l_pending++;
1257 	}
1258 
1259 	/*
1260 	 * If there are any writes that are pending, we return EBUSY.  All
1261 	 * callers will have to wait until all pending writes clear.
1262 	 */
1263 	if (lp->l_pending > 0) {
1264 		vol->v_pending_lock = 1;
1265 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1266 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1267 		return (EBUSY);
1268 	}
1269 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1270 	    (intmax_t)off, (intmax_t)(off+len));
1271 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1272 	return (0);
1273 }
1274 
1275 int
1276 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1277 {
1278 	struct g_raid_lock *lp;
1279 	struct g_raid_softc *sc;
1280 	struct bio *bp;
1281 
1282 	sc = vol->v_softc;
1283 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1284 		if (lp->l_offset == off && lp->l_length == len) {
1285 			LIST_REMOVE(lp, l_next);
1286 			/* XXX
1287 			 * Right now we just put them all back on the queue
1288 			 * and hope for the best.  We hope this because any
1289 			 * locked ranges will go right back on this list
1290 			 * when the worker thread runs.
1291 			 * XXX
1292 			 */
1293 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1294 			    (intmax_t)lp->l_offset,
1295 			    (intmax_t)(lp->l_offset+lp->l_length));
1296 			mtx_lock(&sc->sc_queue_mtx);
1297 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1298 				bioq_disksort(&sc->sc_queue, bp);
1299 			mtx_unlock(&sc->sc_queue_mtx);
1300 			free(lp, M_RAID);
1301 			return (0);
1302 		}
1303 	}
1304 	return (EINVAL);
1305 }
1306 
1307 void
1308 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1309 {
1310 	struct g_consumer *cp;
1311 	struct g_raid_disk *disk, *tdisk;
1312 
1313 	bp->bio_caller1 = sd;
1314 
1315 	/*
1316 	 * Make sure that the disk is present. Generally it is a task of
1317 	 * transformation layers to not send requests to absent disks, but
1318 	 * it is better to be safe and report situation then sorry.
1319 	 */
1320 	if (sd->sd_disk == NULL) {
1321 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1322 nodisk:
1323 		bp->bio_from = NULL;
1324 		bp->bio_to = NULL;
1325 		bp->bio_error = ENXIO;
1326 		g_raid_disk_done(bp);
1327 		return;
1328 	}
1329 	disk = sd->sd_disk;
1330 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1331 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1332 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1333 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1334 		goto nodisk;
1335 	}
1336 
1337 	cp = disk->d_consumer;
1338 	bp->bio_from = cp;
1339 	bp->bio_to = cp->provider;
1340 	cp->index++;
1341 
1342 	/* Update average disks load. */
1343 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1344 		if (tdisk->d_consumer == NULL)
1345 			tdisk->d_load = 0;
1346 		else
1347 			tdisk->d_load = (tdisk->d_consumer->index *
1348 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1349 	}
1350 
1351 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1352 	if (dumping) {
1353 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1354 		if (bp->bio_cmd == BIO_WRITE) {
1355 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1356 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1357 		} else
1358 			bp->bio_error = EOPNOTSUPP;
1359 		g_raid_disk_done(bp);
1360 	} else {
1361 		bp->bio_done = g_raid_disk_done;
1362 		bp->bio_offset += sd->sd_offset;
1363 		G_RAID_LOGREQ(3, bp, "Sending request.");
1364 		g_io_request(bp, cp);
1365 	}
1366 }
1367 
1368 int
1369 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1370     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1371 {
1372 
1373 	if (sd->sd_disk == NULL)
1374 		return (ENXIO);
1375 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1376 		return (EOPNOTSUPP);
1377 	return (dump_write(&sd->sd_disk->d_kd.di,
1378 	    virtual, physical,
1379 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1380 	    length));
1381 }
1382 
1383 static void
1384 g_raid_disk_done(struct bio *bp)
1385 {
1386 	struct g_raid_softc *sc;
1387 	struct g_raid_subdisk *sd;
1388 
1389 	sd = bp->bio_caller1;
1390 	sc = sd->sd_softc;
1391 	mtx_lock(&sc->sc_queue_mtx);
1392 	bioq_disksort(&sc->sc_queue, bp);
1393 	mtx_unlock(&sc->sc_queue_mtx);
1394 	if (!dumping)
1395 		wakeup(sc);
1396 }
1397 
1398 static void
1399 g_raid_disk_done_request(struct bio *bp)
1400 {
1401 	struct g_raid_softc *sc;
1402 	struct g_raid_disk *disk;
1403 	struct g_raid_subdisk *sd;
1404 	struct g_raid_volume *vol;
1405 
1406 	g_topology_assert_not();
1407 
1408 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1409 	sd = bp->bio_caller1;
1410 	sc = sd->sd_softc;
1411 	vol = sd->sd_volume;
1412 	if (bp->bio_from != NULL) {
1413 		bp->bio_from->index--;
1414 		disk = bp->bio_from->private;
1415 		if (disk == NULL)
1416 			g_raid_kill_consumer(sc, bp->bio_from);
1417 	}
1418 	bp->bio_offset -= sd->sd_offset;
1419 
1420 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1421 }
1422 
1423 static void
1424 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1425 {
1426 
1427 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1428 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1429 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1430 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1431 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1432 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1433 	else
1434 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1435 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1436 		KASSERT(ep->e_error == 0,
1437 		    ("Error cannot be handled."));
1438 		g_raid_event_free(ep);
1439 	} else {
1440 		ep->e_flags |= G_RAID_EVENT_DONE;
1441 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1442 		mtx_lock(&sc->sc_queue_mtx);
1443 		wakeup(ep);
1444 		mtx_unlock(&sc->sc_queue_mtx);
1445 	}
1446 }
1447 
1448 /*
1449  * Worker thread.
1450  */
1451 static void
1452 g_raid_worker(void *arg)
1453 {
1454 	struct g_raid_softc *sc;
1455 	struct g_raid_event *ep;
1456 	struct g_raid_volume *vol;
1457 	struct bio *bp;
1458 	struct timeval now, t;
1459 	int timeout, rv;
1460 
1461 	sc = arg;
1462 	thread_lock(curthread);
1463 	sched_prio(curthread, PRIBIO);
1464 	thread_unlock(curthread);
1465 
1466 	sx_xlock(&sc->sc_lock);
1467 	for (;;) {
1468 		mtx_lock(&sc->sc_queue_mtx);
1469 		/*
1470 		 * First take a look at events.
1471 		 * This is important to handle events before any I/O requests.
1472 		 */
1473 		bp = NULL;
1474 		vol = NULL;
1475 		rv = 0;
1476 		ep = TAILQ_FIRST(&sc->sc_events);
1477 		if (ep != NULL)
1478 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1479 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1480 			;
1481 		else {
1482 			getmicrouptime(&now);
1483 			t = now;
1484 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1485 				if (bioq_first(&vol->v_inflight) == NULL &&
1486 				    vol->v_tr &&
1487 				    timevalcmp(&vol->v_last_done, &t, < ))
1488 					t = vol->v_last_done;
1489 			}
1490 			timevalsub(&t, &now);
1491 			timeout = g_raid_idle_threshold +
1492 			    t.tv_sec * 1000000 + t.tv_usec;
1493 			if (timeout > 0) {
1494 				/*
1495 				 * Two steps to avoid overflows at HZ=1000
1496 				 * and idle timeouts > 2.1s.  Some rounding
1497 				 * errors can occur, but they are < 1tick,
1498 				 * which is deemed to be close enough for
1499 				 * this purpose.
1500 				 */
1501 				int micpertic = 1000000 / hz;
1502 				timeout = (timeout + micpertic - 1) / micpertic;
1503 				sx_xunlock(&sc->sc_lock);
1504 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1505 				    PRIBIO | PDROP, "-", timeout);
1506 				sx_xlock(&sc->sc_lock);
1507 				goto process;
1508 			} else
1509 				rv = EWOULDBLOCK;
1510 		}
1511 		mtx_unlock(&sc->sc_queue_mtx);
1512 process:
1513 		if (ep != NULL) {
1514 			g_raid_handle_event(sc, ep);
1515 		} else if (bp != NULL) {
1516 			if (bp->bio_to != NULL &&
1517 			    bp->bio_to->geom == sc->sc_geom)
1518 				g_raid_start_request(bp);
1519 			else
1520 				g_raid_disk_done_request(bp);
1521 		} else if (rv == EWOULDBLOCK) {
1522 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1523 				if (vol->v_writes == 0 && vol->v_dirty)
1524 					g_raid_clean(vol, -1);
1525 				if (bioq_first(&vol->v_inflight) == NULL &&
1526 				    vol->v_tr) {
1527 					t.tv_sec = g_raid_idle_threshold / 1000000;
1528 					t.tv_usec = g_raid_idle_threshold % 1000000;
1529 					timevaladd(&t, &vol->v_last_done);
1530 					getmicrouptime(&now);
1531 					if (timevalcmp(&t, &now, <= )) {
1532 						G_RAID_TR_IDLE(vol->v_tr);
1533 						vol->v_last_done = now;
1534 					}
1535 				}
1536 			}
1537 		}
1538 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1539 			g_raid_destroy_node(sc, 1);	/* May not return. */
1540 	}
1541 }
1542 
1543 static void
1544 g_raid_poll(struct g_raid_softc *sc)
1545 {
1546 	struct g_raid_event *ep;
1547 	struct bio *bp;
1548 
1549 	sx_xlock(&sc->sc_lock);
1550 	mtx_lock(&sc->sc_queue_mtx);
1551 	/*
1552 	 * First take a look at events.
1553 	 * This is important to handle events before any I/O requests.
1554 	 */
1555 	ep = TAILQ_FIRST(&sc->sc_events);
1556 	if (ep != NULL) {
1557 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1558 		mtx_unlock(&sc->sc_queue_mtx);
1559 		g_raid_handle_event(sc, ep);
1560 		goto out;
1561 	}
1562 	bp = bioq_takefirst(&sc->sc_queue);
1563 	if (bp != NULL) {
1564 		mtx_unlock(&sc->sc_queue_mtx);
1565 		if (bp->bio_from == NULL ||
1566 		    bp->bio_from->geom != sc->sc_geom)
1567 			g_raid_start_request(bp);
1568 		else
1569 			g_raid_disk_done_request(bp);
1570 	}
1571 out:
1572 	sx_xunlock(&sc->sc_lock);
1573 }
1574 
1575 static void
1576 g_raid_launch_provider(struct g_raid_volume *vol)
1577 {
1578 	struct g_raid_disk *disk;
1579 	struct g_raid_softc *sc;
1580 	struct g_provider *pp;
1581 	char name[G_RAID_MAX_VOLUMENAME];
1582 	off_t off;
1583 
1584 	sc = vol->v_softc;
1585 	sx_assert(&sc->sc_lock, SX_LOCKED);
1586 
1587 	g_topology_lock();
1588 	/* Try to name provider with volume name. */
1589 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1590 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1591 	    g_provider_by_name(name) != NULL) {
1592 		/* Otherwise use sequential volume number. */
1593 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1594 	}
1595 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1596 	pp->private = vol;
1597 	pp->mediasize = vol->v_mediasize;
1598 	pp->sectorsize = vol->v_sectorsize;
1599 	pp->stripesize = 0;
1600 	pp->stripeoffset = 0;
1601 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1602 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1603 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1604 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1605 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1606 		    disk->d_consumer != NULL &&
1607 		    disk->d_consumer->provider != NULL) {
1608 			pp->stripesize = disk->d_consumer->provider->stripesize;
1609 			off = disk->d_consumer->provider->stripeoffset;
1610 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1611 			if (off > 0)
1612 				pp->stripeoffset %= off;
1613 		}
1614 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1615 			pp->stripesize *= (vol->v_disks_count - 1);
1616 			pp->stripeoffset *= (vol->v_disks_count - 1);
1617 		}
1618 	} else
1619 		pp->stripesize = vol->v_strip_size;
1620 	vol->v_provider = pp;
1621 	g_error_provider(pp, 0);
1622 	g_topology_unlock();
1623 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1624 	    pp->name, vol->v_name);
1625 }
1626 
1627 static void
1628 g_raid_destroy_provider(struct g_raid_volume *vol)
1629 {
1630 	struct g_raid_softc *sc;
1631 	struct g_provider *pp;
1632 	struct bio *bp, *tmp;
1633 
1634 	g_topology_assert_not();
1635 	sc = vol->v_softc;
1636 	pp = vol->v_provider;
1637 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1638 
1639 	g_topology_lock();
1640 	g_error_provider(pp, ENXIO);
1641 	mtx_lock(&sc->sc_queue_mtx);
1642 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1643 		if (bp->bio_to != pp)
1644 			continue;
1645 		bioq_remove(&sc->sc_queue, bp);
1646 		g_io_deliver(bp, ENXIO);
1647 	}
1648 	mtx_unlock(&sc->sc_queue_mtx);
1649 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1650 	    pp->name, vol->v_name);
1651 	g_wither_provider(pp, ENXIO);
1652 	g_topology_unlock();
1653 	vol->v_provider = NULL;
1654 }
1655 
1656 /*
1657  * Update device state.
1658  */
1659 static int
1660 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1661 {
1662 	struct g_raid_softc *sc;
1663 
1664 	sc = vol->v_softc;
1665 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1666 
1667 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1668 	    g_raid_volume_event2str(event),
1669 	    vol->v_name);
1670 	switch (event) {
1671 	case G_RAID_VOLUME_E_DOWN:
1672 		if (vol->v_provider != NULL)
1673 			g_raid_destroy_provider(vol);
1674 		break;
1675 	case G_RAID_VOLUME_E_UP:
1676 		if (vol->v_provider == NULL)
1677 			g_raid_launch_provider(vol);
1678 		break;
1679 	case G_RAID_VOLUME_E_START:
1680 		if (vol->v_tr)
1681 			G_RAID_TR_START(vol->v_tr);
1682 		return (0);
1683 	default:
1684 		if (sc->sc_md)
1685 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1686 		return (0);
1687 	}
1688 
1689 	/* Manage root mount release. */
1690 	if (vol->v_starting) {
1691 		vol->v_starting = 0;
1692 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1693 		root_mount_rel(vol->v_rootmount);
1694 		vol->v_rootmount = NULL;
1695 	}
1696 	if (vol->v_stopping && vol->v_provider_open == 0)
1697 		g_raid_destroy_volume(vol);
1698 	return (0);
1699 }
1700 
1701 /*
1702  * Update subdisk state.
1703  */
1704 static int
1705 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1706 {
1707 	struct g_raid_softc *sc;
1708 	struct g_raid_volume *vol;
1709 
1710 	sc = sd->sd_softc;
1711 	vol = sd->sd_volume;
1712 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1713 
1714 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1715 	    g_raid_subdisk_event2str(event),
1716 	    vol->v_name, sd->sd_pos,
1717 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1718 	if (vol->v_tr)
1719 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1720 
1721 	return (0);
1722 }
1723 
1724 /*
1725  * Update disk state.
1726  */
1727 static int
1728 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1729 {
1730 	struct g_raid_softc *sc;
1731 
1732 	sc = disk->d_softc;
1733 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1734 
1735 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1736 	    g_raid_disk_event2str(event),
1737 	    g_raid_get_diskname(disk));
1738 
1739 	if (sc->sc_md)
1740 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1741 	return (0);
1742 }
1743 
1744 /*
1745  * Node event.
1746  */
1747 static int
1748 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1749 {
1750 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1751 
1752 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1753 	    g_raid_node_event2str(event));
1754 
1755 	if (event == G_RAID_NODE_E_WAKE)
1756 		return (0);
1757 	if (sc->sc_md)
1758 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1759 	return (0);
1760 }
1761 
1762 static int
1763 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1764 {
1765 	struct g_raid_volume *vol;
1766 	struct g_raid_softc *sc;
1767 	int dcw, opens, error = 0;
1768 
1769 	g_topology_assert();
1770 	sc = pp->geom->softc;
1771 	vol = pp->private;
1772 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1773 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1774 
1775 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1776 	    acr, acw, ace);
1777 	dcw = pp->acw + acw;
1778 
1779 	g_topology_unlock();
1780 	sx_xlock(&sc->sc_lock);
1781 	/* Deny new opens while dying. */
1782 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1783 		error = ENXIO;
1784 		goto out;
1785 	}
1786 	if (dcw == 0 && vol->v_dirty)
1787 		g_raid_clean(vol, dcw);
1788 	vol->v_provider_open += acr + acw + ace;
1789 	/* Handle delayed node destruction. */
1790 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1791 	    vol->v_provider_open == 0) {
1792 		/* Count open volumes. */
1793 		opens = g_raid_nopens(sc);
1794 		if (opens == 0) {
1795 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1796 			/* Wake up worker to make it selfdestruct. */
1797 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1798 		}
1799 	}
1800 	/* Handle open volume destruction. */
1801 	if (vol->v_stopping && vol->v_provider_open == 0)
1802 		g_raid_destroy_volume(vol);
1803 out:
1804 	sx_xunlock(&sc->sc_lock);
1805 	g_topology_lock();
1806 	return (error);
1807 }
1808 
1809 struct g_raid_softc *
1810 g_raid_create_node(struct g_class *mp,
1811     const char *name, struct g_raid_md_object *md)
1812 {
1813 	struct g_raid_softc *sc;
1814 	struct g_geom *gp;
1815 	int error;
1816 
1817 	g_topology_assert();
1818 	G_RAID_DEBUG(1, "Creating array %s.", name);
1819 
1820 	gp = g_new_geomf(mp, "%s", name);
1821 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1822 	gp->start = g_raid_start;
1823 	gp->orphan = g_raid_orphan;
1824 	gp->access = g_raid_access;
1825 	gp->dumpconf = g_raid_dumpconf;
1826 
1827 	sc->sc_md = md;
1828 	sc->sc_geom = gp;
1829 	sc->sc_flags = 0;
1830 	TAILQ_INIT(&sc->sc_volumes);
1831 	TAILQ_INIT(&sc->sc_disks);
1832 	sx_init(&sc->sc_lock, "graid:lock");
1833 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1834 	TAILQ_INIT(&sc->sc_events);
1835 	bioq_init(&sc->sc_queue);
1836 	gp->softc = sc;
1837 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1838 	    "g_raid %s", name);
1839 	if (error != 0) {
1840 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1841 		mtx_destroy(&sc->sc_queue_mtx);
1842 		sx_destroy(&sc->sc_lock);
1843 		g_destroy_geom(sc->sc_geom);
1844 		free(sc, M_RAID);
1845 		return (NULL);
1846 	}
1847 
1848 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1849 	return (sc);
1850 }
1851 
1852 struct g_raid_volume *
1853 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1854 {
1855 	struct g_raid_volume	*vol, *vol1;
1856 	int i;
1857 
1858 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1859 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1860 	vol->v_softc = sc;
1861 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1862 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1863 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1864 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1865 	vol->v_rotate_parity = 1;
1866 	bioq_init(&vol->v_inflight);
1867 	bioq_init(&vol->v_locked);
1868 	LIST_INIT(&vol->v_locks);
1869 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1870 		vol->v_subdisks[i].sd_softc = sc;
1871 		vol->v_subdisks[i].sd_volume = vol;
1872 		vol->v_subdisks[i].sd_pos = i;
1873 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1874 	}
1875 
1876 	/* Find free ID for this volume. */
1877 	g_topology_lock();
1878 	vol1 = vol;
1879 	if (id >= 0) {
1880 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1881 			if (vol1->v_global_id == id)
1882 				break;
1883 		}
1884 	}
1885 	if (vol1 != NULL) {
1886 		for (id = 0; ; id++) {
1887 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1888 				if (vol1->v_global_id == id)
1889 					break;
1890 			}
1891 			if (vol1 == NULL)
1892 				break;
1893 		}
1894 	}
1895 	vol->v_global_id = id;
1896 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1897 	g_topology_unlock();
1898 
1899 	/* Delay root mounting. */
1900 	vol->v_rootmount = root_mount_hold("GRAID");
1901 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1902 	vol->v_starting = 1;
1903 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1904 	return (vol);
1905 }
1906 
1907 struct g_raid_disk *
1908 g_raid_create_disk(struct g_raid_softc *sc)
1909 {
1910 	struct g_raid_disk	*disk;
1911 
1912 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1913 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1914 	disk->d_softc = sc;
1915 	disk->d_state = G_RAID_DISK_S_NONE;
1916 	TAILQ_INIT(&disk->d_subdisks);
1917 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1918 	return (disk);
1919 }
1920 
1921 int g_raid_start_volume(struct g_raid_volume *vol)
1922 {
1923 	struct g_raid_tr_class *class;
1924 	struct g_raid_tr_object *obj;
1925 	int status;
1926 
1927 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1928 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1929 		if (!class->trc_enable)
1930 			continue;
1931 		G_RAID_DEBUG1(2, vol->v_softc,
1932 		    "Tasting volume %s for %s transformation.",
1933 		    vol->v_name, class->name);
1934 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1935 		    M_WAITOK);
1936 		obj->tro_class = class;
1937 		obj->tro_volume = vol;
1938 		status = G_RAID_TR_TASTE(obj, vol);
1939 		if (status != G_RAID_TR_TASTE_FAIL)
1940 			break;
1941 		kobj_delete((kobj_t)obj, M_RAID);
1942 	}
1943 	if (class == NULL) {
1944 		G_RAID_DEBUG1(0, vol->v_softc,
1945 		    "No transformation module found for %s.",
1946 		    vol->v_name);
1947 		vol->v_tr = NULL;
1948 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
1949 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
1950 		    G_RAID_EVENT_VOLUME);
1951 		return (-1);
1952 	}
1953 	G_RAID_DEBUG1(2, vol->v_softc,
1954 	    "Transformation module %s chosen for %s.",
1955 	    class->name, vol->v_name);
1956 	vol->v_tr = obj;
1957 	return (0);
1958 }
1959 
1960 int
1961 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
1962 {
1963 	struct g_raid_volume *vol, *tmpv;
1964 	struct g_raid_disk *disk, *tmpd;
1965 	int error = 0;
1966 
1967 	sc->sc_stopping = G_RAID_DESTROY_HARD;
1968 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
1969 		if (g_raid_destroy_volume(vol))
1970 			error = EBUSY;
1971 	}
1972 	if (error)
1973 		return (error);
1974 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
1975 		if (g_raid_destroy_disk(disk))
1976 			error = EBUSY;
1977 	}
1978 	if (error)
1979 		return (error);
1980 	if (sc->sc_md) {
1981 		G_RAID_MD_FREE(sc->sc_md);
1982 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
1983 		sc->sc_md = NULL;
1984 	}
1985 	if (sc->sc_geom != NULL) {
1986 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
1987 		g_topology_lock();
1988 		sc->sc_geom->softc = NULL;
1989 		g_wither_geom(sc->sc_geom, ENXIO);
1990 		g_topology_unlock();
1991 		sc->sc_geom = NULL;
1992 	} else
1993 		G_RAID_DEBUG(1, "Array destroyed.");
1994 	if (worker) {
1995 		g_raid_event_cancel(sc, sc);
1996 		mtx_destroy(&sc->sc_queue_mtx);
1997 		sx_xunlock(&sc->sc_lock);
1998 		sx_destroy(&sc->sc_lock);
1999 		wakeup(&sc->sc_stopping);
2000 		free(sc, M_RAID);
2001 		curthread->td_pflags &= ~TDP_GEOM;
2002 		G_RAID_DEBUG(1, "Thread exiting.");
2003 		kproc_exit(0);
2004 	} else {
2005 		/* Wake up worker to make it selfdestruct. */
2006 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2007 	}
2008 	return (0);
2009 }
2010 
2011 int
2012 g_raid_destroy_volume(struct g_raid_volume *vol)
2013 {
2014 	struct g_raid_softc *sc;
2015 	struct g_raid_disk *disk;
2016 	int i;
2017 
2018 	sc = vol->v_softc;
2019 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2020 	vol->v_stopping = 1;
2021 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2022 		if (vol->v_tr) {
2023 			G_RAID_TR_STOP(vol->v_tr);
2024 			return (EBUSY);
2025 		} else
2026 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2027 	}
2028 	if (g_raid_event_check(sc, vol) != 0)
2029 		return (EBUSY);
2030 	if (vol->v_provider != NULL)
2031 		return (EBUSY);
2032 	if (vol->v_provider_open != 0)
2033 		return (EBUSY);
2034 	if (vol->v_tr) {
2035 		G_RAID_TR_FREE(vol->v_tr);
2036 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2037 		vol->v_tr = NULL;
2038 	}
2039 	if (vol->v_rootmount)
2040 		root_mount_rel(vol->v_rootmount);
2041 	g_topology_lock();
2042 	LIST_REMOVE(vol, v_global_next);
2043 	g_topology_unlock();
2044 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2045 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2046 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2047 		disk = vol->v_subdisks[i].sd_disk;
2048 		if (disk == NULL)
2049 			continue;
2050 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2051 	}
2052 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2053 	if (sc->sc_md)
2054 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2055 	g_raid_event_cancel(sc, vol);
2056 	free(vol, M_RAID);
2057 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2058 		/* Wake up worker to let it selfdestruct. */
2059 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2060 	}
2061 	return (0);
2062 }
2063 
2064 int
2065 g_raid_destroy_disk(struct g_raid_disk *disk)
2066 {
2067 	struct g_raid_softc *sc;
2068 	struct g_raid_subdisk *sd, *tmp;
2069 
2070 	sc = disk->d_softc;
2071 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2072 	if (disk->d_consumer) {
2073 		g_raid_kill_consumer(sc, disk->d_consumer);
2074 		disk->d_consumer = NULL;
2075 	}
2076 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2077 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2078 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2079 		    G_RAID_EVENT_SUBDISK);
2080 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2081 		sd->sd_disk = NULL;
2082 	}
2083 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2084 	if (sc->sc_md)
2085 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2086 	g_raid_event_cancel(sc, disk);
2087 	free(disk, M_RAID);
2088 	return (0);
2089 }
2090 
2091 int
2092 g_raid_destroy(struct g_raid_softc *sc, int how)
2093 {
2094 	int opens;
2095 
2096 	g_topology_assert_not();
2097 	if (sc == NULL)
2098 		return (ENXIO);
2099 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2100 
2101 	/* Count open volumes. */
2102 	opens = g_raid_nopens(sc);
2103 
2104 	/* React on some opened volumes. */
2105 	if (opens > 0) {
2106 		switch (how) {
2107 		case G_RAID_DESTROY_SOFT:
2108 			G_RAID_DEBUG1(1, sc,
2109 			    "%d volumes are still open.",
2110 			    opens);
2111 			return (EBUSY);
2112 		case G_RAID_DESTROY_DELAYED:
2113 			G_RAID_DEBUG1(1, sc,
2114 			    "Array will be destroyed on last close.");
2115 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2116 			return (EBUSY);
2117 		case G_RAID_DESTROY_HARD:
2118 			G_RAID_DEBUG1(1, sc,
2119 			    "%d volumes are still open.",
2120 			    opens);
2121 		}
2122 	}
2123 
2124 	/* Mark node for destruction. */
2125 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2126 	/* Wake up worker to let it selfdestruct. */
2127 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2128 	/* Sleep until node destroyed. */
2129 	sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2130 	    PRIBIO | PDROP, "r:destroy", 0);
2131 	return (0);
2132 }
2133 
2134 static void
2135 g_raid_taste_orphan(struct g_consumer *cp)
2136 {
2137 
2138 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2139 	    cp->provider->name));
2140 }
2141 
2142 static struct g_geom *
2143 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2144 {
2145 	struct g_consumer *cp;
2146 	struct g_geom *gp, *geom;
2147 	struct g_raid_md_class *class;
2148 	struct g_raid_md_object *obj;
2149 	int status;
2150 
2151 	g_topology_assert();
2152 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2153 	if (!g_raid_enable)
2154 		return (NULL);
2155 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2156 
2157 	gp = g_new_geomf(mp, "raid:taste");
2158 	/*
2159 	 * This orphan function should be never called.
2160 	 */
2161 	gp->orphan = g_raid_taste_orphan;
2162 	cp = g_new_consumer(gp);
2163 	g_attach(cp, pp);
2164 
2165 	geom = NULL;
2166 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2167 		if (!class->mdc_enable)
2168 			continue;
2169 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2170 		    pp->name, class->name);
2171 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2172 		    M_WAITOK);
2173 		obj->mdo_class = class;
2174 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2175 		if (status != G_RAID_MD_TASTE_NEW)
2176 			kobj_delete((kobj_t)obj, M_RAID);
2177 		if (status != G_RAID_MD_TASTE_FAIL)
2178 			break;
2179 	}
2180 
2181 	g_detach(cp);
2182 	g_destroy_consumer(cp);
2183 	g_destroy_geom(gp);
2184 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2185 	return (geom);
2186 }
2187 
2188 int
2189 g_raid_create_node_format(const char *format, struct gctl_req *req,
2190     struct g_geom **gp)
2191 {
2192 	struct g_raid_md_class *class;
2193 	struct g_raid_md_object *obj;
2194 	int status;
2195 
2196 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2197 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2198 		if (strcasecmp(class->name, format) == 0)
2199 			break;
2200 	}
2201 	if (class == NULL) {
2202 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2203 		return (G_RAID_MD_TASTE_FAIL);
2204 	}
2205 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2206 	    M_WAITOK);
2207 	obj->mdo_class = class;
2208 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2209 	if (status != G_RAID_MD_TASTE_NEW)
2210 		kobj_delete((kobj_t)obj, M_RAID);
2211 	return (status);
2212 }
2213 
2214 static int
2215 g_raid_destroy_geom(struct gctl_req *req __unused,
2216     struct g_class *mp __unused, struct g_geom *gp)
2217 {
2218 	struct g_raid_softc *sc;
2219 	int error;
2220 
2221 	g_topology_unlock();
2222 	sc = gp->softc;
2223 	sx_xlock(&sc->sc_lock);
2224 	g_cancel_event(sc);
2225 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2226 	if (error != 0)
2227 		sx_xunlock(&sc->sc_lock);
2228 	g_topology_lock();
2229 	return (error);
2230 }
2231 
2232 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2233     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2234 {
2235 
2236 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2237 		return;
2238 	if (sc->sc_md)
2239 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2240 }
2241 
2242 void g_raid_fail_disk(struct g_raid_softc *sc,
2243     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2244 {
2245 
2246 	if (disk == NULL)
2247 		disk = sd->sd_disk;
2248 	if (disk == NULL) {
2249 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2250 		return;
2251 	}
2252 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2253 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2254 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2255 		return;
2256 	}
2257 	if (sc->sc_md)
2258 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2259 }
2260 
2261 static void
2262 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2263     struct g_consumer *cp, struct g_provider *pp)
2264 {
2265 	struct g_raid_softc *sc;
2266 	struct g_raid_volume *vol;
2267 	struct g_raid_subdisk *sd;
2268 	struct g_raid_disk *disk;
2269 	int i, s;
2270 
2271 	g_topology_assert();
2272 
2273 	sc = gp->softc;
2274 	if (sc == NULL)
2275 		return;
2276 	if (pp != NULL) {
2277 		vol = pp->private;
2278 		g_topology_unlock();
2279 		sx_xlock(&sc->sc_lock);
2280 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2281 		    vol->v_name);
2282 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2283 		    g_raid_volume_level2str(vol->v_raid_level,
2284 		    vol->v_raid_level_qualifier));
2285 		sbuf_printf(sb,
2286 		    "%s<Transformation>%s</Transformation>\n", indent,
2287 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2288 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2289 		    vol->v_disks_count);
2290 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2291 		    vol->v_strip_size);
2292 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2293 		    g_raid_volume_state2str(vol->v_state));
2294 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2295 		    vol->v_dirty ? "Yes" : "No");
2296 		sbuf_printf(sb, "%s<Subdisks>", indent);
2297 		for (i = 0; i < vol->v_disks_count; i++) {
2298 			sd = &vol->v_subdisks[i];
2299 			if (sd->sd_disk != NULL &&
2300 			    sd->sd_disk->d_consumer != NULL) {
2301 				sbuf_printf(sb, "%s ",
2302 				    g_raid_get_diskname(sd->sd_disk));
2303 			} else {
2304 				sbuf_printf(sb, "NONE ");
2305 			}
2306 			sbuf_printf(sb, "(%s",
2307 			    g_raid_subdisk_state2str(sd->sd_state));
2308 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2309 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2310 				sbuf_printf(sb, " %d%%",
2311 				    (int)(sd->sd_rebuild_pos * 100 /
2312 				     sd->sd_size));
2313 			}
2314 			sbuf_printf(sb, ")");
2315 			if (i + 1 < vol->v_disks_count)
2316 				sbuf_printf(sb, ", ");
2317 		}
2318 		sbuf_printf(sb, "</Subdisks>\n");
2319 		sx_xunlock(&sc->sc_lock);
2320 		g_topology_lock();
2321 	} else if (cp != NULL) {
2322 		disk = cp->private;
2323 		if (disk == NULL)
2324 			return;
2325 		g_topology_unlock();
2326 		sx_xlock(&sc->sc_lock);
2327 		sbuf_printf(sb, "%s<State>%s", indent,
2328 		    g_raid_disk_state2str(disk->d_state));
2329 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2330 			sbuf_printf(sb, " (");
2331 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2332 				sbuf_printf(sb, "%s",
2333 				    g_raid_subdisk_state2str(sd->sd_state));
2334 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2335 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2336 					sbuf_printf(sb, " %d%%",
2337 					    (int)(sd->sd_rebuild_pos * 100 /
2338 					     sd->sd_size));
2339 				}
2340 				if (TAILQ_NEXT(sd, sd_next))
2341 					sbuf_printf(sb, ", ");
2342 			}
2343 			sbuf_printf(sb, ")");
2344 		}
2345 		sbuf_printf(sb, "</State>\n");
2346 		sbuf_printf(sb, "%s<Subdisks>", indent);
2347 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2348 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2349 			    sd->sd_volume->v_global_id,
2350 			    sd->sd_volume->v_name,
2351 			    sd->sd_pos, sd->sd_offset);
2352 			if (TAILQ_NEXT(sd, sd_next))
2353 				sbuf_printf(sb, ", ");
2354 		}
2355 		sbuf_printf(sb, "</Subdisks>\n");
2356 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2357 		    disk->d_read_errs);
2358 		sx_xunlock(&sc->sc_lock);
2359 		g_topology_lock();
2360 	} else {
2361 		g_topology_unlock();
2362 		sx_xlock(&sc->sc_lock);
2363 		if (sc->sc_md) {
2364 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2365 			    sc->sc_md->mdo_class->name);
2366 		}
2367 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2368 			s = 0xff;
2369 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2370 				if (vol->v_state < s)
2371 					s = vol->v_state;
2372 			}
2373 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2374 			    g_raid_volume_state2str(s));
2375 		}
2376 		sx_xunlock(&sc->sc_lock);
2377 		g_topology_lock();
2378 	}
2379 }
2380 
2381 static void
2382 g_raid_shutdown_pre_sync(void *arg, int howto)
2383 {
2384 	struct g_class *mp;
2385 	struct g_geom *gp, *gp2;
2386 	struct g_raid_softc *sc;
2387 	int error;
2388 
2389 	mp = arg;
2390 	DROP_GIANT();
2391 	g_topology_lock();
2392 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2393 		if ((sc = gp->softc) == NULL)
2394 			continue;
2395 		g_topology_unlock();
2396 		sx_xlock(&sc->sc_lock);
2397 		g_cancel_event(sc);
2398 		error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2399 		if (error != 0)
2400 			sx_xunlock(&sc->sc_lock);
2401 		g_topology_lock();
2402 	}
2403 	g_topology_unlock();
2404 	PICKUP_GIANT();
2405 }
2406 
2407 static void
2408 g_raid_init(struct g_class *mp)
2409 {
2410 
2411 	g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync,
2412 	    g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST);
2413 	if (g_raid_pre_sync == NULL)
2414 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2415 	g_raid_started = 1;
2416 }
2417 
2418 static void
2419 g_raid_fini(struct g_class *mp)
2420 {
2421 
2422 	if (g_raid_pre_sync != NULL)
2423 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync);
2424 	g_raid_started = 0;
2425 }
2426 
2427 int
2428 g_raid_md_modevent(module_t mod, int type, void *arg)
2429 {
2430 	struct g_raid_md_class *class, *c, *nc;
2431 	int error;
2432 
2433 	error = 0;
2434 	class = arg;
2435 	switch (type) {
2436 	case MOD_LOAD:
2437 		c = LIST_FIRST(&g_raid_md_classes);
2438 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2439 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2440 		else {
2441 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2442 			    nc->mdc_priority < class->mdc_priority)
2443 				c = nc;
2444 			LIST_INSERT_AFTER(c, class, mdc_list);
2445 		}
2446 		if (g_raid_started)
2447 			g_retaste(&g_raid_class);
2448 		break;
2449 	case MOD_UNLOAD:
2450 		LIST_REMOVE(class, mdc_list);
2451 		break;
2452 	default:
2453 		error = EOPNOTSUPP;
2454 		break;
2455 	}
2456 
2457 	return (error);
2458 }
2459 
2460 int
2461 g_raid_tr_modevent(module_t mod, int type, void *arg)
2462 {
2463 	struct g_raid_tr_class *class, *c, *nc;
2464 	int error;
2465 
2466 	error = 0;
2467 	class = arg;
2468 	switch (type) {
2469 	case MOD_LOAD:
2470 		c = LIST_FIRST(&g_raid_tr_classes);
2471 		if (c == NULL || c->trc_priority > class->trc_priority)
2472 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2473 		else {
2474 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2475 			    nc->trc_priority < class->trc_priority)
2476 				c = nc;
2477 			LIST_INSERT_AFTER(c, class, trc_list);
2478 		}
2479 		break;
2480 	case MOD_UNLOAD:
2481 		LIST_REMOVE(class, trc_list);
2482 		break;
2483 	default:
2484 		error = EOPNOTSUPP;
2485 		break;
2486 	}
2487 
2488 	return (error);
2489 }
2490 
2491 /*
2492  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2493  * to reduce module priority, allowing submodules to register them first.
2494  */
2495 static moduledata_t g_raid_mod = {
2496 	"g_raid",
2497 	g_modevent,
2498 	&g_raid_class
2499 };
2500 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2501 MODULE_VERSION(geom_raid, 0);
2502