xref: /freebsd/sys/geom/raid/g_raid.c (revision e40139ff33b48b56a24c808b166b04b8ee6f5b21)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/bio.h>
40 #include <sys/sbuf.h>
41 #include <sys/sysctl.h>
42 #include <sys/malloc.h>
43 #include <sys/eventhandler.h>
44 #include <vm/uma.h>
45 #include <geom/geom.h>
46 #include <geom/geom_dbg.h>
47 #include <sys/proc.h>
48 #include <sys/kthread.h>
49 #include <sys/sched.h>
50 #include <geom/raid/g_raid.h>
51 #include "g_raid_md_if.h"
52 #include "g_raid_tr_if.h"
53 
54 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
55 
56 SYSCTL_DECL(_kern_geom);
57 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
58 int g_raid_enable = 1;
59 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN,
60     &g_raid_enable, 0, "Enable on-disk metadata taste");
61 u_int g_raid_aggressive_spare = 0;
62 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN,
63     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
64 u_int g_raid_debug = 0;
65 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0,
66     "Debug level");
67 int g_raid_read_err_thresh = 10;
68 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN,
69     &g_raid_read_err_thresh, 0,
70     "Number of read errors equated to disk failure");
71 u_int g_raid_start_timeout = 30;
72 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN,
73     &g_raid_start_timeout, 0,
74     "Time to wait for all array components");
75 static u_int g_raid_clean_time = 5;
76 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN,
77     &g_raid_clean_time, 0, "Mark volume as clean when idling");
78 static u_int g_raid_disconnect_on_failure = 1;
79 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN,
80     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
81 static u_int g_raid_name_format = 0;
82 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN,
83     &g_raid_name_format, 0, "Providers name format.");
84 static u_int g_raid_idle_threshold = 1000000;
85 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN,
86     &g_raid_idle_threshold, 1000000,
87     "Time in microseconds to consider a volume idle.");
88 
89 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
90 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
91 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
92 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
93 } while (0)
94 
95 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
96     LIST_HEAD_INITIALIZER(g_raid_md_classes);
97 
98 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
99     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
100 
101 LIST_HEAD(, g_raid_volume) g_raid_volumes =
102     LIST_HEAD_INITIALIZER(g_raid_volumes);
103 
104 static eventhandler_tag g_raid_post_sync = NULL;
105 static int g_raid_started = 0;
106 static int g_raid_shutdown = 0;
107 
108 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
109     struct g_geom *gp);
110 static g_taste_t g_raid_taste;
111 static void g_raid_init(struct g_class *mp);
112 static void g_raid_fini(struct g_class *mp);
113 
114 struct g_class g_raid_class = {
115 	.name = G_RAID_CLASS_NAME,
116 	.version = G_VERSION,
117 	.ctlreq = g_raid_ctl,
118 	.taste = g_raid_taste,
119 	.destroy_geom = g_raid_destroy_geom,
120 	.init = g_raid_init,
121 	.fini = g_raid_fini
122 };
123 
124 static void g_raid_destroy_provider(struct g_raid_volume *vol);
125 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
126 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
127 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
128 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
129 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
130     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
131 static void g_raid_start(struct bio *bp);
132 static void g_raid_start_request(struct bio *bp);
133 static void g_raid_disk_done(struct bio *bp);
134 static void g_raid_poll(struct g_raid_softc *sc);
135 
136 static const char *
137 g_raid_node_event2str(int event)
138 {
139 
140 	switch (event) {
141 	case G_RAID_NODE_E_WAKE:
142 		return ("WAKE");
143 	case G_RAID_NODE_E_START:
144 		return ("START");
145 	default:
146 		return ("INVALID");
147 	}
148 }
149 
150 const char *
151 g_raid_disk_state2str(int state)
152 {
153 
154 	switch (state) {
155 	case G_RAID_DISK_S_NONE:
156 		return ("NONE");
157 	case G_RAID_DISK_S_OFFLINE:
158 		return ("OFFLINE");
159 	case G_RAID_DISK_S_DISABLED:
160 		return ("DISABLED");
161 	case G_RAID_DISK_S_FAILED:
162 		return ("FAILED");
163 	case G_RAID_DISK_S_STALE_FAILED:
164 		return ("STALE_FAILED");
165 	case G_RAID_DISK_S_SPARE:
166 		return ("SPARE");
167 	case G_RAID_DISK_S_STALE:
168 		return ("STALE");
169 	case G_RAID_DISK_S_ACTIVE:
170 		return ("ACTIVE");
171 	default:
172 		return ("INVALID");
173 	}
174 }
175 
176 static const char *
177 g_raid_disk_event2str(int event)
178 {
179 
180 	switch (event) {
181 	case G_RAID_DISK_E_DISCONNECTED:
182 		return ("DISCONNECTED");
183 	default:
184 		return ("INVALID");
185 	}
186 }
187 
188 const char *
189 g_raid_subdisk_state2str(int state)
190 {
191 
192 	switch (state) {
193 	case G_RAID_SUBDISK_S_NONE:
194 		return ("NONE");
195 	case G_RAID_SUBDISK_S_FAILED:
196 		return ("FAILED");
197 	case G_RAID_SUBDISK_S_NEW:
198 		return ("NEW");
199 	case G_RAID_SUBDISK_S_REBUILD:
200 		return ("REBUILD");
201 	case G_RAID_SUBDISK_S_UNINITIALIZED:
202 		return ("UNINITIALIZED");
203 	case G_RAID_SUBDISK_S_STALE:
204 		return ("STALE");
205 	case G_RAID_SUBDISK_S_RESYNC:
206 		return ("RESYNC");
207 	case G_RAID_SUBDISK_S_ACTIVE:
208 		return ("ACTIVE");
209 	default:
210 		return ("INVALID");
211 	}
212 }
213 
214 static const char *
215 g_raid_subdisk_event2str(int event)
216 {
217 
218 	switch (event) {
219 	case G_RAID_SUBDISK_E_NEW:
220 		return ("NEW");
221 	case G_RAID_SUBDISK_E_FAILED:
222 		return ("FAILED");
223 	case G_RAID_SUBDISK_E_DISCONNECTED:
224 		return ("DISCONNECTED");
225 	default:
226 		return ("INVALID");
227 	}
228 }
229 
230 const char *
231 g_raid_volume_state2str(int state)
232 {
233 
234 	switch (state) {
235 	case G_RAID_VOLUME_S_STARTING:
236 		return ("STARTING");
237 	case G_RAID_VOLUME_S_BROKEN:
238 		return ("BROKEN");
239 	case G_RAID_VOLUME_S_DEGRADED:
240 		return ("DEGRADED");
241 	case G_RAID_VOLUME_S_SUBOPTIMAL:
242 		return ("SUBOPTIMAL");
243 	case G_RAID_VOLUME_S_OPTIMAL:
244 		return ("OPTIMAL");
245 	case G_RAID_VOLUME_S_UNSUPPORTED:
246 		return ("UNSUPPORTED");
247 	case G_RAID_VOLUME_S_STOPPED:
248 		return ("STOPPED");
249 	default:
250 		return ("INVALID");
251 	}
252 }
253 
254 static const char *
255 g_raid_volume_event2str(int event)
256 {
257 
258 	switch (event) {
259 	case G_RAID_VOLUME_E_UP:
260 		return ("UP");
261 	case G_RAID_VOLUME_E_DOWN:
262 		return ("DOWN");
263 	case G_RAID_VOLUME_E_START:
264 		return ("START");
265 	case G_RAID_VOLUME_E_STARTMD:
266 		return ("STARTMD");
267 	default:
268 		return ("INVALID");
269 	}
270 }
271 
272 const char *
273 g_raid_volume_level2str(int level, int qual)
274 {
275 
276 	switch (level) {
277 	case G_RAID_VOLUME_RL_RAID0:
278 		return ("RAID0");
279 	case G_RAID_VOLUME_RL_RAID1:
280 		return ("RAID1");
281 	case G_RAID_VOLUME_RL_RAID3:
282 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
283 			return ("RAID3-P0");
284 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
285 			return ("RAID3-PN");
286 		return ("RAID3");
287 	case G_RAID_VOLUME_RL_RAID4:
288 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
289 			return ("RAID4-P0");
290 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
291 			return ("RAID4-PN");
292 		return ("RAID4");
293 	case G_RAID_VOLUME_RL_RAID5:
294 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
295 			return ("RAID5-RA");
296 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
297 			return ("RAID5-RS");
298 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
299 			return ("RAID5-LA");
300 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
301 			return ("RAID5-LS");
302 		return ("RAID5");
303 	case G_RAID_VOLUME_RL_RAID6:
304 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
305 			return ("RAID6-RA");
306 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
307 			return ("RAID6-RS");
308 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
309 			return ("RAID6-LA");
310 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
311 			return ("RAID6-LS");
312 		return ("RAID6");
313 	case G_RAID_VOLUME_RL_RAIDMDF:
314 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
315 			return ("RAIDMDF-RA");
316 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
317 			return ("RAIDMDF-RS");
318 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
319 			return ("RAIDMDF-LA");
320 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
321 			return ("RAIDMDF-LS");
322 		return ("RAIDMDF");
323 	case G_RAID_VOLUME_RL_RAID1E:
324 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
325 			return ("RAID1E-A");
326 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
327 			return ("RAID1E-O");
328 		return ("RAID1E");
329 	case G_RAID_VOLUME_RL_SINGLE:
330 		return ("SINGLE");
331 	case G_RAID_VOLUME_RL_CONCAT:
332 		return ("CONCAT");
333 	case G_RAID_VOLUME_RL_RAID5E:
334 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
335 			return ("RAID5E-RA");
336 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
337 			return ("RAID5E-RS");
338 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
339 			return ("RAID5E-LA");
340 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
341 			return ("RAID5E-LS");
342 		return ("RAID5E");
343 	case G_RAID_VOLUME_RL_RAID5EE:
344 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
345 			return ("RAID5EE-RA");
346 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
347 			return ("RAID5EE-RS");
348 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
349 			return ("RAID5EE-LA");
350 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
351 			return ("RAID5EE-LS");
352 		return ("RAID5EE");
353 	case G_RAID_VOLUME_RL_RAID5R:
354 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
355 			return ("RAID5R-RA");
356 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
357 			return ("RAID5R-RS");
358 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
359 			return ("RAID5R-LA");
360 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
361 			return ("RAID5R-LS");
362 		return ("RAID5E");
363 	default:
364 		return ("UNKNOWN");
365 	}
366 }
367 
368 int
369 g_raid_volume_str2level(const char *str, int *level, int *qual)
370 {
371 
372 	*level = G_RAID_VOLUME_RL_UNKNOWN;
373 	*qual = G_RAID_VOLUME_RLQ_NONE;
374 	if (strcasecmp(str, "RAID0") == 0)
375 		*level = G_RAID_VOLUME_RL_RAID0;
376 	else if (strcasecmp(str, "RAID1") == 0)
377 		*level = G_RAID_VOLUME_RL_RAID1;
378 	else if (strcasecmp(str, "RAID3-P0") == 0) {
379 		*level = G_RAID_VOLUME_RL_RAID3;
380 		*qual = G_RAID_VOLUME_RLQ_R3P0;
381 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
382 		   strcasecmp(str, "RAID3") == 0) {
383 		*level = G_RAID_VOLUME_RL_RAID3;
384 		*qual = G_RAID_VOLUME_RLQ_R3PN;
385 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
386 		*level = G_RAID_VOLUME_RL_RAID4;
387 		*qual = G_RAID_VOLUME_RLQ_R4P0;
388 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
389 		   strcasecmp(str, "RAID4") == 0) {
390 		*level = G_RAID_VOLUME_RL_RAID4;
391 		*qual = G_RAID_VOLUME_RLQ_R4PN;
392 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
393 		*level = G_RAID_VOLUME_RL_RAID5;
394 		*qual = G_RAID_VOLUME_RLQ_R5RA;
395 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
396 		*level = G_RAID_VOLUME_RL_RAID5;
397 		*qual = G_RAID_VOLUME_RLQ_R5RS;
398 	} else if (strcasecmp(str, "RAID5") == 0 ||
399 		   strcasecmp(str, "RAID5-LA") == 0) {
400 		*level = G_RAID_VOLUME_RL_RAID5;
401 		*qual = G_RAID_VOLUME_RLQ_R5LA;
402 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
403 		*level = G_RAID_VOLUME_RL_RAID5;
404 		*qual = G_RAID_VOLUME_RLQ_R5LS;
405 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
406 		*level = G_RAID_VOLUME_RL_RAID6;
407 		*qual = G_RAID_VOLUME_RLQ_R6RA;
408 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
409 		*level = G_RAID_VOLUME_RL_RAID6;
410 		*qual = G_RAID_VOLUME_RLQ_R6RS;
411 	} else if (strcasecmp(str, "RAID6") == 0 ||
412 		   strcasecmp(str, "RAID6-LA") == 0) {
413 		*level = G_RAID_VOLUME_RL_RAID6;
414 		*qual = G_RAID_VOLUME_RLQ_R6LA;
415 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
416 		*level = G_RAID_VOLUME_RL_RAID6;
417 		*qual = G_RAID_VOLUME_RLQ_R6LS;
418 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
419 		*level = G_RAID_VOLUME_RL_RAIDMDF;
420 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
421 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
422 		*level = G_RAID_VOLUME_RL_RAIDMDF;
423 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
424 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
425 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
426 		*level = G_RAID_VOLUME_RL_RAIDMDF;
427 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
428 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
429 		*level = G_RAID_VOLUME_RL_RAIDMDF;
430 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
431 	} else if (strcasecmp(str, "RAID10") == 0 ||
432 		   strcasecmp(str, "RAID1E") == 0 ||
433 		   strcasecmp(str, "RAID1E-A") == 0) {
434 		*level = G_RAID_VOLUME_RL_RAID1E;
435 		*qual = G_RAID_VOLUME_RLQ_R1EA;
436 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
437 		*level = G_RAID_VOLUME_RL_RAID1E;
438 		*qual = G_RAID_VOLUME_RLQ_R1EO;
439 	} else if (strcasecmp(str, "SINGLE") == 0)
440 		*level = G_RAID_VOLUME_RL_SINGLE;
441 	else if (strcasecmp(str, "CONCAT") == 0)
442 		*level = G_RAID_VOLUME_RL_CONCAT;
443 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
444 		*level = G_RAID_VOLUME_RL_RAID5E;
445 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
446 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
447 		*level = G_RAID_VOLUME_RL_RAID5E;
448 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
449 	} else if (strcasecmp(str, "RAID5E") == 0 ||
450 		   strcasecmp(str, "RAID5E-LA") == 0) {
451 		*level = G_RAID_VOLUME_RL_RAID5E;
452 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
453 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
454 		*level = G_RAID_VOLUME_RL_RAID5E;
455 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
456 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
457 		*level = G_RAID_VOLUME_RL_RAID5EE;
458 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
459 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
460 		*level = G_RAID_VOLUME_RL_RAID5EE;
461 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
462 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
463 		   strcasecmp(str, "RAID5EE-LA") == 0) {
464 		*level = G_RAID_VOLUME_RL_RAID5EE;
465 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
466 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
467 		*level = G_RAID_VOLUME_RL_RAID5EE;
468 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
469 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
470 		*level = G_RAID_VOLUME_RL_RAID5R;
471 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
472 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
473 		*level = G_RAID_VOLUME_RL_RAID5R;
474 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
475 	} else if (strcasecmp(str, "RAID5R") == 0 ||
476 		   strcasecmp(str, "RAID5R-LA") == 0) {
477 		*level = G_RAID_VOLUME_RL_RAID5R;
478 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
479 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
480 		*level = G_RAID_VOLUME_RL_RAID5R;
481 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
482 	} else
483 		return (-1);
484 	return (0);
485 }
486 
487 const char *
488 g_raid_get_diskname(struct g_raid_disk *disk)
489 {
490 
491 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
492 		return ("[unknown]");
493 	return (disk->d_consumer->provider->name);
494 }
495 
496 void
497 g_raid_get_disk_info(struct g_raid_disk *disk)
498 {
499 	struct g_consumer *cp = disk->d_consumer;
500 	int error, len;
501 
502 	/* Read kernel dumping information. */
503 	disk->d_kd.offset = 0;
504 	disk->d_kd.length = OFF_MAX;
505 	len = sizeof(disk->d_kd);
506 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
507 	if (error)
508 		disk->d_kd.di.dumper = NULL;
509 	if (disk->d_kd.di.dumper == NULL)
510 		G_RAID_DEBUG1(2, disk->d_softc,
511 		    "Dumping not supported by %s: %d.",
512 		    cp->provider->name, error);
513 
514 	/* Read BIO_DELETE support. */
515 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
516 	if (error)
517 		disk->d_candelete = 0;
518 	if (!disk->d_candelete)
519 		G_RAID_DEBUG1(2, disk->d_softc,
520 		    "BIO_DELETE not supported by %s: %d.",
521 		    cp->provider->name, error);
522 }
523 
524 void
525 g_raid_report_disk_state(struct g_raid_disk *disk)
526 {
527 	struct g_raid_subdisk *sd;
528 	int len, state;
529 	uint32_t s;
530 
531 	if (disk->d_consumer == NULL)
532 		return;
533 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
534 		s = G_STATE_ACTIVE; /* XXX */
535 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
536 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
537 		s = G_STATE_FAILED;
538 	} else {
539 		state = G_RAID_SUBDISK_S_ACTIVE;
540 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
541 			if (sd->sd_state < state)
542 				state = sd->sd_state;
543 		}
544 		if (state == G_RAID_SUBDISK_S_FAILED)
545 			s = G_STATE_FAILED;
546 		else if (state == G_RAID_SUBDISK_S_NEW ||
547 		    state == G_RAID_SUBDISK_S_REBUILD)
548 			s = G_STATE_REBUILD;
549 		else if (state == G_RAID_SUBDISK_S_STALE ||
550 		    state == G_RAID_SUBDISK_S_RESYNC)
551 			s = G_STATE_RESYNC;
552 		else
553 			s = G_STATE_ACTIVE;
554 	}
555 	len = sizeof(s);
556 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
557 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
558 	    g_raid_get_diskname(disk), s);
559 }
560 
561 void
562 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
563 {
564 
565 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
566 	    g_raid_get_diskname(disk),
567 	    g_raid_disk_state2str(disk->d_state),
568 	    g_raid_disk_state2str(state));
569 	disk->d_state = state;
570 	g_raid_report_disk_state(disk);
571 }
572 
573 void
574 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
575 {
576 
577 	G_RAID_DEBUG1(0, sd->sd_softc,
578 	    "Subdisk %s:%d-%s state changed from %s to %s.",
579 	    sd->sd_volume->v_name, sd->sd_pos,
580 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
581 	    g_raid_subdisk_state2str(sd->sd_state),
582 	    g_raid_subdisk_state2str(state));
583 	sd->sd_state = state;
584 	if (sd->sd_disk)
585 		g_raid_report_disk_state(sd->sd_disk);
586 }
587 
588 void
589 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
590 {
591 
592 	G_RAID_DEBUG1(0, vol->v_softc,
593 	    "Volume %s state changed from %s to %s.",
594 	    vol->v_name,
595 	    g_raid_volume_state2str(vol->v_state),
596 	    g_raid_volume_state2str(state));
597 	vol->v_state = state;
598 }
599 
600 /*
601  * --- Events handling functions ---
602  * Events in geom_raid are used to maintain subdisks and volumes status
603  * from one thread to simplify locking.
604  */
605 static void
606 g_raid_event_free(struct g_raid_event *ep)
607 {
608 
609 	free(ep, M_RAID);
610 }
611 
612 int
613 g_raid_event_send(void *arg, int event, int flags)
614 {
615 	struct g_raid_softc *sc;
616 	struct g_raid_event *ep;
617 	int error;
618 
619 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
620 		sc = ((struct g_raid_volume *)arg)->v_softc;
621 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
622 		sc = ((struct g_raid_disk *)arg)->d_softc;
623 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
624 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
625 	} else {
626 		sc = arg;
627 	}
628 	ep = malloc(sizeof(*ep), M_RAID,
629 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
630 	if (ep == NULL)
631 		return (ENOMEM);
632 	ep->e_tgt = arg;
633 	ep->e_event = event;
634 	ep->e_flags = flags;
635 	ep->e_error = 0;
636 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
637 	mtx_lock(&sc->sc_queue_mtx);
638 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
639 	mtx_unlock(&sc->sc_queue_mtx);
640 	wakeup(sc);
641 
642 	if ((flags & G_RAID_EVENT_WAIT) == 0)
643 		return (0);
644 
645 	sx_assert(&sc->sc_lock, SX_XLOCKED);
646 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
647 	sx_xunlock(&sc->sc_lock);
648 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
649 		mtx_lock(&sc->sc_queue_mtx);
650 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
651 		    hz * 5);
652 	}
653 	error = ep->e_error;
654 	g_raid_event_free(ep);
655 	sx_xlock(&sc->sc_lock);
656 	return (error);
657 }
658 
659 static void
660 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
661 {
662 	struct g_raid_event *ep, *tmpep;
663 
664 	sx_assert(&sc->sc_lock, SX_XLOCKED);
665 
666 	mtx_lock(&sc->sc_queue_mtx);
667 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
668 		if (ep->e_tgt != tgt)
669 			continue;
670 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
671 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
672 			g_raid_event_free(ep);
673 		else {
674 			ep->e_error = ECANCELED;
675 			wakeup(ep);
676 		}
677 	}
678 	mtx_unlock(&sc->sc_queue_mtx);
679 }
680 
681 static int
682 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
683 {
684 	struct g_raid_event *ep;
685 	int	res = 0;
686 
687 	sx_assert(&sc->sc_lock, SX_XLOCKED);
688 
689 	mtx_lock(&sc->sc_queue_mtx);
690 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
691 		if (ep->e_tgt != tgt)
692 			continue;
693 		res = 1;
694 		break;
695 	}
696 	mtx_unlock(&sc->sc_queue_mtx);
697 	return (res);
698 }
699 
700 /*
701  * Return the number of disks in given state.
702  * If state is equal to -1, count all connected disks.
703  */
704 u_int
705 g_raid_ndisks(struct g_raid_softc *sc, int state)
706 {
707 	struct g_raid_disk *disk;
708 	u_int n;
709 
710 	sx_assert(&sc->sc_lock, SX_LOCKED);
711 
712 	n = 0;
713 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
714 		if (disk->d_state == state || state == -1)
715 			n++;
716 	}
717 	return (n);
718 }
719 
720 /*
721  * Return the number of subdisks in given state.
722  * If state is equal to -1, count all connected disks.
723  */
724 u_int
725 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
726 {
727 	struct g_raid_subdisk *subdisk;
728 	struct g_raid_softc *sc;
729 	u_int i, n ;
730 
731 	sc = vol->v_softc;
732 	sx_assert(&sc->sc_lock, SX_LOCKED);
733 
734 	n = 0;
735 	for (i = 0; i < vol->v_disks_count; i++) {
736 		subdisk = &vol->v_subdisks[i];
737 		if ((state == -1 &&
738 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
739 		    subdisk->sd_state == state)
740 			n++;
741 	}
742 	return (n);
743 }
744 
745 /*
746  * Return the first subdisk in given state.
747  * If state is equal to -1, then the first connected disks.
748  */
749 struct g_raid_subdisk *
750 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
751 {
752 	struct g_raid_subdisk *sd;
753 	struct g_raid_softc *sc;
754 	u_int i;
755 
756 	sc = vol->v_softc;
757 	sx_assert(&sc->sc_lock, SX_LOCKED);
758 
759 	for (i = 0; i < vol->v_disks_count; i++) {
760 		sd = &vol->v_subdisks[i];
761 		if ((state == -1 &&
762 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
763 		    sd->sd_state == state)
764 			return (sd);
765 	}
766 	return (NULL);
767 }
768 
769 struct g_consumer *
770 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
771 {
772 	struct g_consumer *cp;
773 	struct g_provider *pp;
774 
775 	g_topology_assert();
776 
777 	if (strncmp(name, "/dev/", 5) == 0)
778 		name += 5;
779 	pp = g_provider_by_name(name);
780 	if (pp == NULL)
781 		return (NULL);
782 	cp = g_new_consumer(sc->sc_geom);
783 	cp->flags |= G_CF_DIRECT_RECEIVE;
784 	if (g_attach(cp, pp) != 0) {
785 		g_destroy_consumer(cp);
786 		return (NULL);
787 	}
788 	if (g_access(cp, 1, 1, 1) != 0) {
789 		g_detach(cp);
790 		g_destroy_consumer(cp);
791 		return (NULL);
792 	}
793 	return (cp);
794 }
795 
796 static u_int
797 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
798 {
799 	struct bio *bp;
800 	u_int nreqs = 0;
801 
802 	mtx_lock(&sc->sc_queue_mtx);
803 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
804 		if (bp->bio_from == cp)
805 			nreqs++;
806 	}
807 	mtx_unlock(&sc->sc_queue_mtx);
808 	return (nreqs);
809 }
810 
811 u_int
812 g_raid_nopens(struct g_raid_softc *sc)
813 {
814 	struct g_raid_volume *vol;
815 	u_int opens;
816 
817 	opens = 0;
818 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
819 		if (vol->v_provider_open != 0)
820 			opens++;
821 	}
822 	return (opens);
823 }
824 
825 static int
826 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
827 {
828 
829 	if (cp->index > 0) {
830 		G_RAID_DEBUG1(2, sc,
831 		    "I/O requests for %s exist, can't destroy it now.",
832 		    cp->provider->name);
833 		return (1);
834 	}
835 	if (g_raid_nrequests(sc, cp) > 0) {
836 		G_RAID_DEBUG1(2, sc,
837 		    "I/O requests for %s in queue, can't destroy it now.",
838 		    cp->provider->name);
839 		return (1);
840 	}
841 	return (0);
842 }
843 
844 static void
845 g_raid_destroy_consumer(void *arg, int flags __unused)
846 {
847 	struct g_consumer *cp;
848 
849 	g_topology_assert();
850 
851 	cp = arg;
852 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
853 	g_detach(cp);
854 	g_destroy_consumer(cp);
855 }
856 
857 void
858 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
859 {
860 	struct g_provider *pp;
861 	int retaste_wait;
862 
863 	g_topology_assert_not();
864 
865 	g_topology_lock();
866 	cp->private = NULL;
867 	if (g_raid_consumer_is_busy(sc, cp))
868 		goto out;
869 	pp = cp->provider;
870 	retaste_wait = 0;
871 	if (cp->acw == 1) {
872 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
873 			retaste_wait = 1;
874 	}
875 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
876 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
877 	if (retaste_wait) {
878 		/*
879 		 * After retaste event was send (inside g_access()), we can send
880 		 * event to detach and destroy consumer.
881 		 * A class, which has consumer to the given provider connected
882 		 * will not receive retaste event for the provider.
883 		 * This is the way how I ignore retaste events when I close
884 		 * consumers opened for write: I detach and destroy consumer
885 		 * after retaste event is sent.
886 		 */
887 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
888 		goto out;
889 	}
890 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
891 	g_detach(cp);
892 	g_destroy_consumer(cp);
893 out:
894 	g_topology_unlock();
895 }
896 
897 static void
898 g_raid_orphan(struct g_consumer *cp)
899 {
900 	struct g_raid_disk *disk;
901 
902 	g_topology_assert();
903 
904 	disk = cp->private;
905 	if (disk == NULL)
906 		return;
907 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
908 	    G_RAID_EVENT_DISK);
909 }
910 
911 static void
912 g_raid_clean(struct g_raid_volume *vol, int acw)
913 {
914 	struct g_raid_softc *sc;
915 	int timeout;
916 
917 	sc = vol->v_softc;
918 	g_topology_assert_not();
919 	sx_assert(&sc->sc_lock, SX_XLOCKED);
920 
921 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
922 //		return;
923 	if (!vol->v_dirty)
924 		return;
925 	if (vol->v_writes > 0)
926 		return;
927 	if (acw > 0 || (acw == -1 &&
928 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
929 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
930 		if (!g_raid_shutdown && timeout > 0)
931 			return;
932 	}
933 	vol->v_dirty = 0;
934 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
935 	    vol->v_name);
936 	g_raid_write_metadata(sc, vol, NULL, NULL);
937 }
938 
939 static void
940 g_raid_dirty(struct g_raid_volume *vol)
941 {
942 	struct g_raid_softc *sc;
943 
944 	sc = vol->v_softc;
945 	g_topology_assert_not();
946 	sx_assert(&sc->sc_lock, SX_XLOCKED);
947 
948 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
949 //		return;
950 	vol->v_dirty = 1;
951 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
952 	    vol->v_name);
953 	g_raid_write_metadata(sc, vol, NULL, NULL);
954 }
955 
956 void
957 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
958 {
959 	struct g_raid_volume *vol;
960 	struct g_raid_subdisk *sd;
961 	struct bio_queue_head queue;
962 	struct bio *cbp;
963 	int i;
964 
965 	vol = tr->tro_volume;
966 
967 	/*
968 	 * Allocate all bios before sending any request, so we can return
969 	 * ENOMEM in nice and clean way.
970 	 */
971 	bioq_init(&queue);
972 	for (i = 0; i < vol->v_disks_count; i++) {
973 		sd = &vol->v_subdisks[i];
974 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
975 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
976 			continue;
977 		cbp = g_clone_bio(bp);
978 		if (cbp == NULL)
979 			goto failure;
980 		cbp->bio_caller1 = sd;
981 		bioq_insert_tail(&queue, cbp);
982 	}
983 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
984 		sd = cbp->bio_caller1;
985 		cbp->bio_caller1 = NULL;
986 		g_raid_subdisk_iostart(sd, cbp);
987 	}
988 	return;
989 failure:
990 	while ((cbp = bioq_takefirst(&queue)) != NULL)
991 		g_destroy_bio(cbp);
992 	if (bp->bio_error == 0)
993 		bp->bio_error = ENOMEM;
994 	g_raid_iodone(bp, bp->bio_error);
995 }
996 
997 static void
998 g_raid_tr_kerneldump_common_done(struct bio *bp)
999 {
1000 
1001 	bp->bio_flags |= BIO_DONE;
1002 }
1003 
1004 int
1005 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1006     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1007 {
1008 	struct g_raid_softc *sc;
1009 	struct g_raid_volume *vol;
1010 	struct bio bp;
1011 
1012 	vol = tr->tro_volume;
1013 	sc = vol->v_softc;
1014 
1015 	g_reset_bio(&bp);
1016 	bp.bio_cmd = BIO_WRITE;
1017 	bp.bio_done = g_raid_tr_kerneldump_common_done;
1018 	bp.bio_attribute = NULL;
1019 	bp.bio_offset = offset;
1020 	bp.bio_length = length;
1021 	bp.bio_data = virtual;
1022 	bp.bio_to = vol->v_provider;
1023 
1024 	g_raid_start(&bp);
1025 	while (!(bp.bio_flags & BIO_DONE)) {
1026 		G_RAID_DEBUG1(4, sc, "Poll...");
1027 		g_raid_poll(sc);
1028 		DELAY(10);
1029 	}
1030 
1031 	return (bp.bio_error != 0 ? EIO : 0);
1032 }
1033 
1034 static int
1035 g_raid_dump(void *arg,
1036     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1037 {
1038 	struct g_raid_volume *vol;
1039 	int error;
1040 
1041 	vol = (struct g_raid_volume *)arg;
1042 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1043 	    (long long unsigned)offset, (long long unsigned)length);
1044 
1045 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1046 	    virtual, physical, offset, length);
1047 	return (error);
1048 }
1049 
1050 static void
1051 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1052 {
1053 	struct g_kerneldump *gkd;
1054 	struct g_provider *pp;
1055 	struct g_raid_volume *vol;
1056 
1057 	gkd = (struct g_kerneldump*)bp->bio_data;
1058 	pp = bp->bio_to;
1059 	vol = pp->private;
1060 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1061 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1062 	gkd->di.dumper = g_raid_dump;
1063 	gkd->di.priv = vol;
1064 	gkd->di.blocksize = vol->v_sectorsize;
1065 	gkd->di.maxiosize = DFLTPHYS;
1066 	gkd->di.mediaoffset = gkd->offset;
1067 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1068 		gkd->length = vol->v_mediasize - gkd->offset;
1069 	gkd->di.mediasize = gkd->length;
1070 	g_io_deliver(bp, 0);
1071 }
1072 
1073 static void
1074 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1075 {
1076 	struct g_provider *pp;
1077 	struct g_raid_volume *vol;
1078 	struct g_raid_subdisk *sd;
1079 	int i, val;
1080 
1081 	pp = bp->bio_to;
1082 	vol = pp->private;
1083 	for (i = 0; i < vol->v_disks_count; i++) {
1084 		sd = &vol->v_subdisks[i];
1085 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1086 			continue;
1087 		if (sd->sd_disk->d_candelete)
1088 			break;
1089 	}
1090 	val = i < vol->v_disks_count;
1091 	g_handleattr(bp, "GEOM::candelete", &val, sizeof(val));
1092 }
1093 
1094 static void
1095 g_raid_start(struct bio *bp)
1096 {
1097 	struct g_raid_softc *sc;
1098 
1099 	sc = bp->bio_to->geom->softc;
1100 	/*
1101 	 * If sc == NULL or there are no valid disks, provider's error
1102 	 * should be set and g_raid_start() should not be called at all.
1103 	 */
1104 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1105 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1106 //	    bp->bio_to->error, bp->bio_to->name));
1107 	G_RAID_LOGREQ(3, bp, "Request received.");
1108 
1109 	switch (bp->bio_cmd) {
1110 	case BIO_READ:
1111 	case BIO_WRITE:
1112 	case BIO_DELETE:
1113 	case BIO_FLUSH:
1114 	case BIO_SPEEDUP:
1115 		break;
1116 	case BIO_GETATTR:
1117 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1118 			g_raid_candelete(sc, bp);
1119 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1120 			g_raid_kerneldump(sc, bp);
1121 		else
1122 			g_io_deliver(bp, EOPNOTSUPP);
1123 		return;
1124 	default:
1125 		g_io_deliver(bp, EOPNOTSUPP);
1126 		return;
1127 	}
1128 	mtx_lock(&sc->sc_queue_mtx);
1129 	bioq_insert_tail(&sc->sc_queue, bp);
1130 	mtx_unlock(&sc->sc_queue_mtx);
1131 	if (!dumping) {
1132 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1133 		wakeup(sc);
1134 	}
1135 }
1136 
1137 static int
1138 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1139 {
1140 	/*
1141 	 * 5 cases:
1142 	 * (1) bp entirely below NO
1143 	 * (2) bp entirely above NO
1144 	 * (3) bp start below, but end in range YES
1145 	 * (4) bp entirely within YES
1146 	 * (5) bp starts within, ends above YES
1147 	 *
1148 	 * lock range 10-19 (offset 10 length 10)
1149 	 * (1) 1-5: first if kicks it out
1150 	 * (2) 30-35: second if kicks it out
1151 	 * (3) 5-15: passes both ifs
1152 	 * (4) 12-14: passes both ifs
1153 	 * (5) 19-20: passes both
1154 	 */
1155 	off_t lend = lstart + len - 1;
1156 	off_t bstart = bp->bio_offset;
1157 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1158 
1159 	if (bend < lstart)
1160 		return (0);
1161 	if (lend < bstart)
1162 		return (0);
1163 	return (1);
1164 }
1165 
1166 static int
1167 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1168 {
1169 	struct g_raid_lock *lp;
1170 
1171 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1172 
1173 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1174 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1175 			return (1);
1176 	}
1177 	return (0);
1178 }
1179 
1180 static void
1181 g_raid_start_request(struct bio *bp)
1182 {
1183 	struct g_raid_softc *sc;
1184 	struct g_raid_volume *vol;
1185 
1186 	sc = bp->bio_to->geom->softc;
1187 	sx_assert(&sc->sc_lock, SX_LOCKED);
1188 	vol = bp->bio_to->private;
1189 
1190 	/*
1191 	 * Check to see if this item is in a locked range.  If so,
1192 	 * queue it to our locked queue and return.  We'll requeue
1193 	 * it when the range is unlocked.  Internal I/O for the
1194 	 * rebuild/rescan/recovery process is excluded from this
1195 	 * check so we can actually do the recovery.
1196 	 */
1197 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1198 	    g_raid_is_in_locked_range(vol, bp)) {
1199 		G_RAID_LOGREQ(3, bp, "Defer request.");
1200 		bioq_insert_tail(&vol->v_locked, bp);
1201 		return;
1202 	}
1203 
1204 	/*
1205 	 * If we're actually going to do the write/delete, then
1206 	 * update the idle stats for the volume.
1207 	 */
1208 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1209 		if (!vol->v_dirty)
1210 			g_raid_dirty(vol);
1211 		vol->v_writes++;
1212 	}
1213 
1214 	/*
1215 	 * Put request onto inflight queue, so we can check if new
1216 	 * synchronization requests don't collide with it.  Then tell
1217 	 * the transformation layer to start the I/O.
1218 	 */
1219 	bioq_insert_tail(&vol->v_inflight, bp);
1220 	G_RAID_LOGREQ(4, bp, "Request started");
1221 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1222 }
1223 
1224 static void
1225 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1226 {
1227 	off_t off, len;
1228 	struct bio *nbp;
1229 	struct g_raid_lock *lp;
1230 
1231 	vol->v_pending_lock = 0;
1232 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1233 		if (lp->l_pending) {
1234 			off = lp->l_offset;
1235 			len = lp->l_length;
1236 			lp->l_pending = 0;
1237 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1238 				if (g_raid_bio_overlaps(nbp, off, len))
1239 					lp->l_pending++;
1240 			}
1241 			if (lp->l_pending) {
1242 				vol->v_pending_lock = 1;
1243 				G_RAID_DEBUG1(4, vol->v_softc,
1244 				    "Deferred lock(%jd, %jd) has %d pending",
1245 				    (intmax_t)off, (intmax_t)(off + len),
1246 				    lp->l_pending);
1247 				continue;
1248 			}
1249 			G_RAID_DEBUG1(4, vol->v_softc,
1250 			    "Deferred lock of %jd to %jd completed",
1251 			    (intmax_t)off, (intmax_t)(off + len));
1252 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1253 		}
1254 	}
1255 }
1256 
1257 void
1258 g_raid_iodone(struct bio *bp, int error)
1259 {
1260 	struct g_raid_softc *sc;
1261 	struct g_raid_volume *vol;
1262 
1263 	sc = bp->bio_to->geom->softc;
1264 	sx_assert(&sc->sc_lock, SX_LOCKED);
1265 	vol = bp->bio_to->private;
1266 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1267 
1268 	/* Update stats if we done write/delete. */
1269 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1270 		vol->v_writes--;
1271 		vol->v_last_write = time_uptime;
1272 	}
1273 
1274 	bioq_remove(&vol->v_inflight, bp);
1275 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1276 		g_raid_finish_with_locked_ranges(vol, bp);
1277 	getmicrouptime(&vol->v_last_done);
1278 	g_io_deliver(bp, error);
1279 }
1280 
1281 int
1282 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1283     struct bio *ignore, void *argp)
1284 {
1285 	struct g_raid_softc *sc;
1286 	struct g_raid_lock *lp;
1287 	struct bio *bp;
1288 
1289 	sc = vol->v_softc;
1290 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1291 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1292 	lp->l_offset = off;
1293 	lp->l_length = len;
1294 	lp->l_callback_arg = argp;
1295 
1296 	lp->l_pending = 0;
1297 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1298 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1299 			lp->l_pending++;
1300 	}
1301 
1302 	/*
1303 	 * If there are any writes that are pending, we return EBUSY.  All
1304 	 * callers will have to wait until all pending writes clear.
1305 	 */
1306 	if (lp->l_pending > 0) {
1307 		vol->v_pending_lock = 1;
1308 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1309 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1310 		return (EBUSY);
1311 	}
1312 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1313 	    (intmax_t)off, (intmax_t)(off+len));
1314 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1315 	return (0);
1316 }
1317 
1318 int
1319 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1320 {
1321 	struct g_raid_lock *lp;
1322 	struct g_raid_softc *sc;
1323 	struct bio *bp;
1324 
1325 	sc = vol->v_softc;
1326 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1327 		if (lp->l_offset == off && lp->l_length == len) {
1328 			LIST_REMOVE(lp, l_next);
1329 			/* XXX
1330 			 * Right now we just put them all back on the queue
1331 			 * and hope for the best.  We hope this because any
1332 			 * locked ranges will go right back on this list
1333 			 * when the worker thread runs.
1334 			 * XXX
1335 			 */
1336 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1337 			    (intmax_t)lp->l_offset,
1338 			    (intmax_t)(lp->l_offset+lp->l_length));
1339 			mtx_lock(&sc->sc_queue_mtx);
1340 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1341 				bioq_insert_tail(&sc->sc_queue, bp);
1342 			mtx_unlock(&sc->sc_queue_mtx);
1343 			free(lp, M_RAID);
1344 			return (0);
1345 		}
1346 	}
1347 	return (EINVAL);
1348 }
1349 
1350 void
1351 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1352 {
1353 	struct g_consumer *cp;
1354 	struct g_raid_disk *disk, *tdisk;
1355 
1356 	bp->bio_caller1 = sd;
1357 
1358 	/*
1359 	 * Make sure that the disk is present. Generally it is a task of
1360 	 * transformation layers to not send requests to absent disks, but
1361 	 * it is better to be safe and report situation then sorry.
1362 	 */
1363 	if (sd->sd_disk == NULL) {
1364 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1365 nodisk:
1366 		bp->bio_from = NULL;
1367 		bp->bio_to = NULL;
1368 		bp->bio_error = ENXIO;
1369 		g_raid_disk_done(bp);
1370 		return;
1371 	}
1372 	disk = sd->sd_disk;
1373 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1374 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1375 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1376 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1377 		goto nodisk;
1378 	}
1379 
1380 	cp = disk->d_consumer;
1381 	bp->bio_from = cp;
1382 	bp->bio_to = cp->provider;
1383 	cp->index++;
1384 
1385 	/* Update average disks load. */
1386 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1387 		if (tdisk->d_consumer == NULL)
1388 			tdisk->d_load = 0;
1389 		else
1390 			tdisk->d_load = (tdisk->d_consumer->index *
1391 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1392 	}
1393 
1394 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1395 	if (dumping) {
1396 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1397 		if (bp->bio_cmd == BIO_WRITE) {
1398 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1399 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1400 		} else
1401 			bp->bio_error = EOPNOTSUPP;
1402 		g_raid_disk_done(bp);
1403 	} else {
1404 		bp->bio_done = g_raid_disk_done;
1405 		bp->bio_offset += sd->sd_offset;
1406 		G_RAID_LOGREQ(3, bp, "Sending request.");
1407 		g_io_request(bp, cp);
1408 	}
1409 }
1410 
1411 int
1412 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1413     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1414 {
1415 
1416 	if (sd->sd_disk == NULL)
1417 		return (ENXIO);
1418 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1419 		return (EOPNOTSUPP);
1420 	return (dump_write(&sd->sd_disk->d_kd.di,
1421 	    virtual, physical,
1422 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1423 	    length));
1424 }
1425 
1426 static void
1427 g_raid_disk_done(struct bio *bp)
1428 {
1429 	struct g_raid_softc *sc;
1430 	struct g_raid_subdisk *sd;
1431 
1432 	sd = bp->bio_caller1;
1433 	sc = sd->sd_softc;
1434 	mtx_lock(&sc->sc_queue_mtx);
1435 	bioq_insert_tail(&sc->sc_queue, bp);
1436 	mtx_unlock(&sc->sc_queue_mtx);
1437 	if (!dumping)
1438 		wakeup(sc);
1439 }
1440 
1441 static void
1442 g_raid_disk_done_request(struct bio *bp)
1443 {
1444 	struct g_raid_softc *sc;
1445 	struct g_raid_disk *disk;
1446 	struct g_raid_subdisk *sd;
1447 	struct g_raid_volume *vol;
1448 
1449 	g_topology_assert_not();
1450 
1451 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1452 	sd = bp->bio_caller1;
1453 	sc = sd->sd_softc;
1454 	vol = sd->sd_volume;
1455 	if (bp->bio_from != NULL) {
1456 		bp->bio_from->index--;
1457 		disk = bp->bio_from->private;
1458 		if (disk == NULL)
1459 			g_raid_kill_consumer(sc, bp->bio_from);
1460 	}
1461 	bp->bio_offset -= sd->sd_offset;
1462 
1463 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1464 }
1465 
1466 static void
1467 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1468 {
1469 
1470 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1471 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1472 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1473 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1474 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1475 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1476 	else
1477 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1478 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1479 		KASSERT(ep->e_error == 0,
1480 		    ("Error cannot be handled."));
1481 		g_raid_event_free(ep);
1482 	} else {
1483 		ep->e_flags |= G_RAID_EVENT_DONE;
1484 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1485 		mtx_lock(&sc->sc_queue_mtx);
1486 		wakeup(ep);
1487 		mtx_unlock(&sc->sc_queue_mtx);
1488 	}
1489 }
1490 
1491 /*
1492  * Worker thread.
1493  */
1494 static void
1495 g_raid_worker(void *arg)
1496 {
1497 	struct g_raid_softc *sc;
1498 	struct g_raid_event *ep;
1499 	struct g_raid_volume *vol;
1500 	struct bio *bp;
1501 	struct timeval now, t;
1502 	int timeout, rv;
1503 
1504 	sc = arg;
1505 	thread_lock(curthread);
1506 	sched_prio(curthread, PRIBIO);
1507 	thread_unlock(curthread);
1508 
1509 	sx_xlock(&sc->sc_lock);
1510 	for (;;) {
1511 		mtx_lock(&sc->sc_queue_mtx);
1512 		/*
1513 		 * First take a look at events.
1514 		 * This is important to handle events before any I/O requests.
1515 		 */
1516 		bp = NULL;
1517 		vol = NULL;
1518 		rv = 0;
1519 		ep = TAILQ_FIRST(&sc->sc_events);
1520 		if (ep != NULL)
1521 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1522 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1523 			;
1524 		else {
1525 			getmicrouptime(&now);
1526 			t = now;
1527 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1528 				if (bioq_first(&vol->v_inflight) == NULL &&
1529 				    vol->v_tr &&
1530 				    timevalcmp(&vol->v_last_done, &t, < ))
1531 					t = vol->v_last_done;
1532 			}
1533 			timevalsub(&t, &now);
1534 			timeout = g_raid_idle_threshold +
1535 			    t.tv_sec * 1000000 + t.tv_usec;
1536 			if (timeout > 0) {
1537 				/*
1538 				 * Two steps to avoid overflows at HZ=1000
1539 				 * and idle timeouts > 2.1s.  Some rounding
1540 				 * errors can occur, but they are < 1tick,
1541 				 * which is deemed to be close enough for
1542 				 * this purpose.
1543 				 */
1544 				int micpertic = 1000000 / hz;
1545 				timeout = (timeout + micpertic - 1) / micpertic;
1546 				sx_xunlock(&sc->sc_lock);
1547 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1548 				    PRIBIO | PDROP, "-", timeout);
1549 				sx_xlock(&sc->sc_lock);
1550 				goto process;
1551 			} else
1552 				rv = EWOULDBLOCK;
1553 		}
1554 		mtx_unlock(&sc->sc_queue_mtx);
1555 process:
1556 		if (ep != NULL) {
1557 			g_raid_handle_event(sc, ep);
1558 		} else if (bp != NULL) {
1559 			if (bp->bio_to != NULL &&
1560 			    bp->bio_to->geom == sc->sc_geom)
1561 				g_raid_start_request(bp);
1562 			else
1563 				g_raid_disk_done_request(bp);
1564 		} else if (rv == EWOULDBLOCK) {
1565 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1566 				g_raid_clean(vol, -1);
1567 				if (bioq_first(&vol->v_inflight) == NULL &&
1568 				    vol->v_tr) {
1569 					t.tv_sec = g_raid_idle_threshold / 1000000;
1570 					t.tv_usec = g_raid_idle_threshold % 1000000;
1571 					timevaladd(&t, &vol->v_last_done);
1572 					getmicrouptime(&now);
1573 					if (timevalcmp(&t, &now, <= )) {
1574 						G_RAID_TR_IDLE(vol->v_tr);
1575 						vol->v_last_done = now;
1576 					}
1577 				}
1578 			}
1579 		}
1580 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1581 			g_raid_destroy_node(sc, 1);	/* May not return. */
1582 	}
1583 }
1584 
1585 static void
1586 g_raid_poll(struct g_raid_softc *sc)
1587 {
1588 	struct g_raid_event *ep;
1589 	struct bio *bp;
1590 
1591 	sx_xlock(&sc->sc_lock);
1592 	mtx_lock(&sc->sc_queue_mtx);
1593 	/*
1594 	 * First take a look at events.
1595 	 * This is important to handle events before any I/O requests.
1596 	 */
1597 	ep = TAILQ_FIRST(&sc->sc_events);
1598 	if (ep != NULL) {
1599 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1600 		mtx_unlock(&sc->sc_queue_mtx);
1601 		g_raid_handle_event(sc, ep);
1602 		goto out;
1603 	}
1604 	bp = bioq_takefirst(&sc->sc_queue);
1605 	if (bp != NULL) {
1606 		mtx_unlock(&sc->sc_queue_mtx);
1607 		if (bp->bio_from == NULL ||
1608 		    bp->bio_from->geom != sc->sc_geom)
1609 			g_raid_start_request(bp);
1610 		else
1611 			g_raid_disk_done_request(bp);
1612 	}
1613 out:
1614 	sx_xunlock(&sc->sc_lock);
1615 }
1616 
1617 static void
1618 g_raid_launch_provider(struct g_raid_volume *vol)
1619 {
1620 	struct g_raid_disk *disk;
1621 	struct g_raid_subdisk *sd;
1622 	struct g_raid_softc *sc;
1623 	struct g_provider *pp;
1624 	char name[G_RAID_MAX_VOLUMENAME];
1625 	off_t off;
1626 	int i;
1627 
1628 	sc = vol->v_softc;
1629 	sx_assert(&sc->sc_lock, SX_LOCKED);
1630 
1631 	g_topology_lock();
1632 	/* Try to name provider with volume name. */
1633 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1634 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1635 	    g_provider_by_name(name) != NULL) {
1636 		/* Otherwise use sequential volume number. */
1637 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1638 	}
1639 
1640 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1641 	pp->flags |= G_PF_DIRECT_RECEIVE;
1642 	if (vol->v_tr->tro_class->trc_accept_unmapped) {
1643 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1644 		for (i = 0; i < vol->v_disks_count; i++) {
1645 			sd = &vol->v_subdisks[i];
1646 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1647 				continue;
1648 			if ((sd->sd_disk->d_consumer->provider->flags &
1649 			    G_PF_ACCEPT_UNMAPPED) == 0)
1650 				pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
1651 		}
1652 	}
1653 	pp->private = vol;
1654 	pp->mediasize = vol->v_mediasize;
1655 	pp->sectorsize = vol->v_sectorsize;
1656 	pp->stripesize = 0;
1657 	pp->stripeoffset = 0;
1658 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1659 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1660 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1661 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1662 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1663 		    disk->d_consumer != NULL &&
1664 		    disk->d_consumer->provider != NULL) {
1665 			pp->stripesize = disk->d_consumer->provider->stripesize;
1666 			off = disk->d_consumer->provider->stripeoffset;
1667 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1668 			if (off > 0)
1669 				pp->stripeoffset %= off;
1670 		}
1671 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1672 			pp->stripesize *= (vol->v_disks_count - 1);
1673 			pp->stripeoffset *= (vol->v_disks_count - 1);
1674 		}
1675 	} else
1676 		pp->stripesize = vol->v_strip_size;
1677 	vol->v_provider = pp;
1678 	g_error_provider(pp, 0);
1679 	g_topology_unlock();
1680 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1681 	    pp->name, vol->v_name);
1682 }
1683 
1684 static void
1685 g_raid_destroy_provider(struct g_raid_volume *vol)
1686 {
1687 	struct g_raid_softc *sc;
1688 	struct g_provider *pp;
1689 	struct bio *bp, *tmp;
1690 
1691 	g_topology_assert_not();
1692 	sc = vol->v_softc;
1693 	pp = vol->v_provider;
1694 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1695 
1696 	g_topology_lock();
1697 	g_error_provider(pp, ENXIO);
1698 	mtx_lock(&sc->sc_queue_mtx);
1699 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1700 		if (bp->bio_to != pp)
1701 			continue;
1702 		bioq_remove(&sc->sc_queue, bp);
1703 		g_io_deliver(bp, ENXIO);
1704 	}
1705 	mtx_unlock(&sc->sc_queue_mtx);
1706 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1707 	    pp->name, vol->v_name);
1708 	g_wither_provider(pp, ENXIO);
1709 	g_topology_unlock();
1710 	vol->v_provider = NULL;
1711 }
1712 
1713 /*
1714  * Update device state.
1715  */
1716 static int
1717 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1718 {
1719 	struct g_raid_softc *sc;
1720 
1721 	sc = vol->v_softc;
1722 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1723 
1724 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1725 	    g_raid_volume_event2str(event),
1726 	    vol->v_name);
1727 	switch (event) {
1728 	case G_RAID_VOLUME_E_DOWN:
1729 		if (vol->v_provider != NULL)
1730 			g_raid_destroy_provider(vol);
1731 		break;
1732 	case G_RAID_VOLUME_E_UP:
1733 		if (vol->v_provider == NULL)
1734 			g_raid_launch_provider(vol);
1735 		break;
1736 	case G_RAID_VOLUME_E_START:
1737 		if (vol->v_tr)
1738 			G_RAID_TR_START(vol->v_tr);
1739 		return (0);
1740 	default:
1741 		if (sc->sc_md)
1742 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1743 		return (0);
1744 	}
1745 
1746 	/* Manage root mount release. */
1747 	if (vol->v_starting) {
1748 		vol->v_starting = 0;
1749 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1750 		root_mount_rel(vol->v_rootmount);
1751 		vol->v_rootmount = NULL;
1752 	}
1753 	if (vol->v_stopping && vol->v_provider_open == 0)
1754 		g_raid_destroy_volume(vol);
1755 	return (0);
1756 }
1757 
1758 /*
1759  * Update subdisk state.
1760  */
1761 static int
1762 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1763 {
1764 	struct g_raid_softc *sc;
1765 	struct g_raid_volume *vol;
1766 
1767 	sc = sd->sd_softc;
1768 	vol = sd->sd_volume;
1769 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1770 
1771 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1772 	    g_raid_subdisk_event2str(event),
1773 	    vol->v_name, sd->sd_pos,
1774 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1775 	if (vol->v_tr)
1776 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1777 
1778 	return (0);
1779 }
1780 
1781 /*
1782  * Update disk state.
1783  */
1784 static int
1785 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1786 {
1787 	struct g_raid_softc *sc;
1788 
1789 	sc = disk->d_softc;
1790 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1791 
1792 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1793 	    g_raid_disk_event2str(event),
1794 	    g_raid_get_diskname(disk));
1795 
1796 	if (sc->sc_md)
1797 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1798 	return (0);
1799 }
1800 
1801 /*
1802  * Node event.
1803  */
1804 static int
1805 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1806 {
1807 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1808 
1809 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1810 	    g_raid_node_event2str(event));
1811 
1812 	if (event == G_RAID_NODE_E_WAKE)
1813 		return (0);
1814 	if (sc->sc_md)
1815 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1816 	return (0);
1817 }
1818 
1819 static int
1820 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1821 {
1822 	struct g_raid_volume *vol;
1823 	struct g_raid_softc *sc;
1824 	int dcw, opens, error = 0;
1825 
1826 	g_topology_assert();
1827 	sc = pp->geom->softc;
1828 	vol = pp->private;
1829 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1830 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1831 
1832 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1833 	    acr, acw, ace);
1834 	dcw = pp->acw + acw;
1835 
1836 	g_topology_unlock();
1837 	sx_xlock(&sc->sc_lock);
1838 	/* Deny new opens while dying. */
1839 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1840 		error = ENXIO;
1841 		goto out;
1842 	}
1843 	/* Deny write opens for read-only volumes. */
1844 	if (vol->v_read_only && acw > 0) {
1845 		error = EROFS;
1846 		goto out;
1847 	}
1848 	if (dcw == 0)
1849 		g_raid_clean(vol, dcw);
1850 	vol->v_provider_open += acr + acw + ace;
1851 	/* Handle delayed node destruction. */
1852 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1853 	    vol->v_provider_open == 0) {
1854 		/* Count open volumes. */
1855 		opens = g_raid_nopens(sc);
1856 		if (opens == 0) {
1857 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1858 			/* Wake up worker to make it selfdestruct. */
1859 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1860 		}
1861 	}
1862 	/* Handle open volume destruction. */
1863 	if (vol->v_stopping && vol->v_provider_open == 0)
1864 		g_raid_destroy_volume(vol);
1865 out:
1866 	sx_xunlock(&sc->sc_lock);
1867 	g_topology_lock();
1868 	return (error);
1869 }
1870 
1871 struct g_raid_softc *
1872 g_raid_create_node(struct g_class *mp,
1873     const char *name, struct g_raid_md_object *md)
1874 {
1875 	struct g_raid_softc *sc;
1876 	struct g_geom *gp;
1877 	int error;
1878 
1879 	g_topology_assert();
1880 	G_RAID_DEBUG(1, "Creating array %s.", name);
1881 
1882 	gp = g_new_geomf(mp, "%s", name);
1883 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1884 	gp->start = g_raid_start;
1885 	gp->orphan = g_raid_orphan;
1886 	gp->access = g_raid_access;
1887 	gp->dumpconf = g_raid_dumpconf;
1888 
1889 	sc->sc_md = md;
1890 	sc->sc_geom = gp;
1891 	sc->sc_flags = 0;
1892 	TAILQ_INIT(&sc->sc_volumes);
1893 	TAILQ_INIT(&sc->sc_disks);
1894 	sx_init(&sc->sc_lock, "graid:lock");
1895 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1896 	TAILQ_INIT(&sc->sc_events);
1897 	bioq_init(&sc->sc_queue);
1898 	gp->softc = sc;
1899 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1900 	    "g_raid %s", name);
1901 	if (error != 0) {
1902 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1903 		mtx_destroy(&sc->sc_queue_mtx);
1904 		sx_destroy(&sc->sc_lock);
1905 		g_destroy_geom(sc->sc_geom);
1906 		free(sc, M_RAID);
1907 		return (NULL);
1908 	}
1909 
1910 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1911 	return (sc);
1912 }
1913 
1914 struct g_raid_volume *
1915 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1916 {
1917 	struct g_raid_volume	*vol, *vol1;
1918 	int i;
1919 
1920 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1921 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1922 	vol->v_softc = sc;
1923 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1924 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1925 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1926 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1927 	vol->v_rotate_parity = 1;
1928 	bioq_init(&vol->v_inflight);
1929 	bioq_init(&vol->v_locked);
1930 	LIST_INIT(&vol->v_locks);
1931 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1932 		vol->v_subdisks[i].sd_softc = sc;
1933 		vol->v_subdisks[i].sd_volume = vol;
1934 		vol->v_subdisks[i].sd_pos = i;
1935 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1936 	}
1937 
1938 	/* Find free ID for this volume. */
1939 	g_topology_lock();
1940 	vol1 = vol;
1941 	if (id >= 0) {
1942 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1943 			if (vol1->v_global_id == id)
1944 				break;
1945 		}
1946 	}
1947 	if (vol1 != NULL) {
1948 		for (id = 0; ; id++) {
1949 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1950 				if (vol1->v_global_id == id)
1951 					break;
1952 			}
1953 			if (vol1 == NULL)
1954 				break;
1955 		}
1956 	}
1957 	vol->v_global_id = id;
1958 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1959 	g_topology_unlock();
1960 
1961 	/* Delay root mounting. */
1962 	vol->v_rootmount = root_mount_hold("GRAID");
1963 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1964 	vol->v_starting = 1;
1965 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1966 	return (vol);
1967 }
1968 
1969 struct g_raid_disk *
1970 g_raid_create_disk(struct g_raid_softc *sc)
1971 {
1972 	struct g_raid_disk	*disk;
1973 
1974 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1975 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1976 	disk->d_softc = sc;
1977 	disk->d_state = G_RAID_DISK_S_NONE;
1978 	TAILQ_INIT(&disk->d_subdisks);
1979 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1980 	return (disk);
1981 }
1982 
1983 int g_raid_start_volume(struct g_raid_volume *vol)
1984 {
1985 	struct g_raid_tr_class *class;
1986 	struct g_raid_tr_object *obj;
1987 	int status;
1988 
1989 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1990 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1991 		if (!class->trc_enable)
1992 			continue;
1993 		G_RAID_DEBUG1(2, vol->v_softc,
1994 		    "Tasting volume %s for %s transformation.",
1995 		    vol->v_name, class->name);
1996 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1997 		    M_WAITOK);
1998 		obj->tro_class = class;
1999 		obj->tro_volume = vol;
2000 		status = G_RAID_TR_TASTE(obj, vol);
2001 		if (status != G_RAID_TR_TASTE_FAIL)
2002 			break;
2003 		kobj_delete((kobj_t)obj, M_RAID);
2004 	}
2005 	if (class == NULL) {
2006 		G_RAID_DEBUG1(0, vol->v_softc,
2007 		    "No transformation module found for %s.",
2008 		    vol->v_name);
2009 		vol->v_tr = NULL;
2010 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2011 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2012 		    G_RAID_EVENT_VOLUME);
2013 		return (-1);
2014 	}
2015 	G_RAID_DEBUG1(2, vol->v_softc,
2016 	    "Transformation module %s chosen for %s.",
2017 	    class->name, vol->v_name);
2018 	vol->v_tr = obj;
2019 	return (0);
2020 }
2021 
2022 int
2023 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2024 {
2025 	struct g_raid_volume *vol, *tmpv;
2026 	struct g_raid_disk *disk, *tmpd;
2027 	int error = 0;
2028 
2029 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2030 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2031 		if (g_raid_destroy_volume(vol))
2032 			error = EBUSY;
2033 	}
2034 	if (error)
2035 		return (error);
2036 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2037 		if (g_raid_destroy_disk(disk))
2038 			error = EBUSY;
2039 	}
2040 	if (error)
2041 		return (error);
2042 	if (sc->sc_md) {
2043 		G_RAID_MD_FREE(sc->sc_md);
2044 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2045 		sc->sc_md = NULL;
2046 	}
2047 	if (sc->sc_geom != NULL) {
2048 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2049 		g_topology_lock();
2050 		sc->sc_geom->softc = NULL;
2051 		g_wither_geom(sc->sc_geom, ENXIO);
2052 		g_topology_unlock();
2053 		sc->sc_geom = NULL;
2054 	} else
2055 		G_RAID_DEBUG(1, "Array destroyed.");
2056 	if (worker) {
2057 		g_raid_event_cancel(sc, sc);
2058 		mtx_destroy(&sc->sc_queue_mtx);
2059 		sx_xunlock(&sc->sc_lock);
2060 		sx_destroy(&sc->sc_lock);
2061 		wakeup(&sc->sc_stopping);
2062 		free(sc, M_RAID);
2063 		curthread->td_pflags &= ~TDP_GEOM;
2064 		G_RAID_DEBUG(1, "Thread exiting.");
2065 		kproc_exit(0);
2066 	} else {
2067 		/* Wake up worker to make it selfdestruct. */
2068 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2069 	}
2070 	return (0);
2071 }
2072 
2073 int
2074 g_raid_destroy_volume(struct g_raid_volume *vol)
2075 {
2076 	struct g_raid_softc *sc;
2077 	struct g_raid_disk *disk;
2078 	int i;
2079 
2080 	sc = vol->v_softc;
2081 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2082 	vol->v_stopping = 1;
2083 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2084 		if (vol->v_tr) {
2085 			G_RAID_TR_STOP(vol->v_tr);
2086 			return (EBUSY);
2087 		} else
2088 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2089 	}
2090 	if (g_raid_event_check(sc, vol) != 0)
2091 		return (EBUSY);
2092 	if (vol->v_provider != NULL)
2093 		return (EBUSY);
2094 	if (vol->v_provider_open != 0)
2095 		return (EBUSY);
2096 	if (vol->v_tr) {
2097 		G_RAID_TR_FREE(vol->v_tr);
2098 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2099 		vol->v_tr = NULL;
2100 	}
2101 	if (vol->v_rootmount)
2102 		root_mount_rel(vol->v_rootmount);
2103 	g_topology_lock();
2104 	LIST_REMOVE(vol, v_global_next);
2105 	g_topology_unlock();
2106 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2107 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2108 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2109 		disk = vol->v_subdisks[i].sd_disk;
2110 		if (disk == NULL)
2111 			continue;
2112 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2113 	}
2114 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2115 	if (sc->sc_md)
2116 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2117 	g_raid_event_cancel(sc, vol);
2118 	free(vol, M_RAID);
2119 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2120 		/* Wake up worker to let it selfdestruct. */
2121 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2122 	}
2123 	return (0);
2124 }
2125 
2126 int
2127 g_raid_destroy_disk(struct g_raid_disk *disk)
2128 {
2129 	struct g_raid_softc *sc;
2130 	struct g_raid_subdisk *sd, *tmp;
2131 
2132 	sc = disk->d_softc;
2133 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2134 	if (disk->d_consumer) {
2135 		g_raid_kill_consumer(sc, disk->d_consumer);
2136 		disk->d_consumer = NULL;
2137 	}
2138 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2139 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2140 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2141 		    G_RAID_EVENT_SUBDISK);
2142 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2143 		sd->sd_disk = NULL;
2144 	}
2145 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2146 	if (sc->sc_md)
2147 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2148 	g_raid_event_cancel(sc, disk);
2149 	free(disk, M_RAID);
2150 	return (0);
2151 }
2152 
2153 int
2154 g_raid_destroy(struct g_raid_softc *sc, int how)
2155 {
2156 	int error, opens;
2157 
2158 	g_topology_assert_not();
2159 	if (sc == NULL)
2160 		return (ENXIO);
2161 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2162 
2163 	/* Count open volumes. */
2164 	opens = g_raid_nopens(sc);
2165 
2166 	/* React on some opened volumes. */
2167 	if (opens > 0) {
2168 		switch (how) {
2169 		case G_RAID_DESTROY_SOFT:
2170 			G_RAID_DEBUG1(1, sc,
2171 			    "%d volumes are still open.",
2172 			    opens);
2173 			sx_xunlock(&sc->sc_lock);
2174 			return (EBUSY);
2175 		case G_RAID_DESTROY_DELAYED:
2176 			G_RAID_DEBUG1(1, sc,
2177 			    "Array will be destroyed on last close.");
2178 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2179 			sx_xunlock(&sc->sc_lock);
2180 			return (EBUSY);
2181 		case G_RAID_DESTROY_HARD:
2182 			G_RAID_DEBUG1(1, sc,
2183 			    "%d volumes are still open.",
2184 			    opens);
2185 		}
2186 	}
2187 
2188 	/* Mark node for destruction. */
2189 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2190 	/* Wake up worker to let it selfdestruct. */
2191 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2192 	/* Sleep until node destroyed. */
2193 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2194 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2195 	return (error == EWOULDBLOCK ? EBUSY : 0);
2196 }
2197 
2198 static void
2199 g_raid_taste_orphan(struct g_consumer *cp)
2200 {
2201 
2202 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2203 	    cp->provider->name));
2204 }
2205 
2206 static struct g_geom *
2207 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2208 {
2209 	struct g_consumer *cp;
2210 	struct g_geom *gp, *geom;
2211 	struct g_raid_md_class *class;
2212 	struct g_raid_md_object *obj;
2213 	int status;
2214 
2215 	g_topology_assert();
2216 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2217 	if (!g_raid_enable)
2218 		return (NULL);
2219 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2220 
2221 	geom = NULL;
2222 	status = G_RAID_MD_TASTE_FAIL;
2223 	gp = g_new_geomf(mp, "raid:taste");
2224 	/*
2225 	 * This orphan function should be never called.
2226 	 */
2227 	gp->orphan = g_raid_taste_orphan;
2228 	cp = g_new_consumer(gp);
2229 	cp->flags |= G_CF_DIRECT_RECEIVE;
2230 	g_attach(cp, pp);
2231 	if (g_access(cp, 1, 0, 0) != 0)
2232 		goto ofail;
2233 
2234 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2235 		if (!class->mdc_enable)
2236 			continue;
2237 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2238 		    pp->name, class->name);
2239 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2240 		    M_WAITOK);
2241 		obj->mdo_class = class;
2242 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2243 		if (status != G_RAID_MD_TASTE_NEW)
2244 			kobj_delete((kobj_t)obj, M_RAID);
2245 		if (status != G_RAID_MD_TASTE_FAIL)
2246 			break;
2247 	}
2248 
2249 	if (status == G_RAID_MD_TASTE_FAIL)
2250 		(void)g_access(cp, -1, 0, 0);
2251 ofail:
2252 	g_detach(cp);
2253 	g_destroy_consumer(cp);
2254 	g_destroy_geom(gp);
2255 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2256 	return (geom);
2257 }
2258 
2259 int
2260 g_raid_create_node_format(const char *format, struct gctl_req *req,
2261     struct g_geom **gp)
2262 {
2263 	struct g_raid_md_class *class;
2264 	struct g_raid_md_object *obj;
2265 	int status;
2266 
2267 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2268 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2269 		if (strcasecmp(class->name, format) == 0)
2270 			break;
2271 	}
2272 	if (class == NULL) {
2273 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2274 		return (G_RAID_MD_TASTE_FAIL);
2275 	}
2276 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2277 	    M_WAITOK);
2278 	obj->mdo_class = class;
2279 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2280 	if (status != G_RAID_MD_TASTE_NEW)
2281 		kobj_delete((kobj_t)obj, M_RAID);
2282 	return (status);
2283 }
2284 
2285 static int
2286 g_raid_destroy_geom(struct gctl_req *req __unused,
2287     struct g_class *mp __unused, struct g_geom *gp)
2288 {
2289 	struct g_raid_softc *sc;
2290 	int error;
2291 
2292 	g_topology_unlock();
2293 	sc = gp->softc;
2294 	sx_xlock(&sc->sc_lock);
2295 	g_cancel_event(sc);
2296 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2297 	g_topology_lock();
2298 	return (error);
2299 }
2300 
2301 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2302     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2303 {
2304 
2305 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2306 		return;
2307 	if (sc->sc_md)
2308 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2309 }
2310 
2311 void g_raid_fail_disk(struct g_raid_softc *sc,
2312     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2313 {
2314 
2315 	if (disk == NULL)
2316 		disk = sd->sd_disk;
2317 	if (disk == NULL) {
2318 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2319 		return;
2320 	}
2321 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2322 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2323 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2324 		return;
2325 	}
2326 	if (sc->sc_md)
2327 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2328 }
2329 
2330 static void
2331 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2332     struct g_consumer *cp, struct g_provider *pp)
2333 {
2334 	struct g_raid_softc *sc;
2335 	struct g_raid_volume *vol;
2336 	struct g_raid_subdisk *sd;
2337 	struct g_raid_disk *disk;
2338 	int i, s;
2339 
2340 	g_topology_assert();
2341 
2342 	sc = gp->softc;
2343 	if (sc == NULL)
2344 		return;
2345 	if (pp != NULL) {
2346 		vol = pp->private;
2347 		g_topology_unlock();
2348 		sx_xlock(&sc->sc_lock);
2349 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2350 		    sc->sc_md->mdo_class->name,
2351 		    g_raid_volume_level2str(vol->v_raid_level,
2352 		    vol->v_raid_level_qualifier));
2353 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2354 		    vol->v_name);
2355 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2356 		    g_raid_volume_level2str(vol->v_raid_level,
2357 		    vol->v_raid_level_qualifier));
2358 		sbuf_printf(sb,
2359 		    "%s<Transformation>%s</Transformation>\n", indent,
2360 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2361 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2362 		    vol->v_disks_count);
2363 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2364 		    vol->v_strip_size);
2365 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2366 		    g_raid_volume_state2str(vol->v_state));
2367 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2368 		    vol->v_dirty ? "Yes" : "No");
2369 		sbuf_printf(sb, "%s<Subdisks>", indent);
2370 		for (i = 0; i < vol->v_disks_count; i++) {
2371 			sd = &vol->v_subdisks[i];
2372 			if (sd->sd_disk != NULL &&
2373 			    sd->sd_disk->d_consumer != NULL) {
2374 				sbuf_printf(sb, "%s ",
2375 				    g_raid_get_diskname(sd->sd_disk));
2376 			} else {
2377 				sbuf_cat(sb, "NONE ");
2378 			}
2379 			sbuf_printf(sb, "(%s",
2380 			    g_raid_subdisk_state2str(sd->sd_state));
2381 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2382 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2383 				sbuf_printf(sb, " %d%%",
2384 				    (int)(sd->sd_rebuild_pos * 100 /
2385 				     sd->sd_size));
2386 			}
2387 			sbuf_cat(sb, ")");
2388 			if (i + 1 < vol->v_disks_count)
2389 				sbuf_cat(sb, ", ");
2390 		}
2391 		sbuf_cat(sb, "</Subdisks>\n");
2392 		sx_xunlock(&sc->sc_lock);
2393 		g_topology_lock();
2394 	} else if (cp != NULL) {
2395 		disk = cp->private;
2396 		if (disk == NULL)
2397 			return;
2398 		g_topology_unlock();
2399 		sx_xlock(&sc->sc_lock);
2400 		sbuf_printf(sb, "%s<State>%s", indent,
2401 		    g_raid_disk_state2str(disk->d_state));
2402 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2403 			sbuf_cat(sb, " (");
2404 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2405 				sbuf_printf(sb, "%s",
2406 				    g_raid_subdisk_state2str(sd->sd_state));
2407 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2408 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2409 					sbuf_printf(sb, " %d%%",
2410 					    (int)(sd->sd_rebuild_pos * 100 /
2411 					     sd->sd_size));
2412 				}
2413 				if (TAILQ_NEXT(sd, sd_next))
2414 					sbuf_cat(sb, ", ");
2415 			}
2416 			sbuf_cat(sb, ")");
2417 		}
2418 		sbuf_cat(sb, "</State>\n");
2419 		sbuf_printf(sb, "%s<Subdisks>", indent);
2420 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2421 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2422 			    sd->sd_volume->v_global_id,
2423 			    sd->sd_volume->v_name,
2424 			    sd->sd_pos, (uintmax_t)sd->sd_offset);
2425 			if (TAILQ_NEXT(sd, sd_next))
2426 				sbuf_cat(sb, ", ");
2427 		}
2428 		sbuf_cat(sb, "</Subdisks>\n");
2429 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2430 		    disk->d_read_errs);
2431 		sx_xunlock(&sc->sc_lock);
2432 		g_topology_lock();
2433 	} else {
2434 		g_topology_unlock();
2435 		sx_xlock(&sc->sc_lock);
2436 		if (sc->sc_md) {
2437 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2438 			    sc->sc_md->mdo_class->name);
2439 		}
2440 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2441 			s = 0xff;
2442 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2443 				if (vol->v_state < s)
2444 					s = vol->v_state;
2445 			}
2446 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2447 			    g_raid_volume_state2str(s));
2448 		}
2449 		sx_xunlock(&sc->sc_lock);
2450 		g_topology_lock();
2451 	}
2452 }
2453 
2454 static void
2455 g_raid_shutdown_post_sync(void *arg, int howto)
2456 {
2457 	struct g_class *mp;
2458 	struct g_geom *gp, *gp2;
2459 	struct g_raid_softc *sc;
2460 	struct g_raid_volume *vol;
2461 
2462 	mp = arg;
2463 	g_topology_lock();
2464 	g_raid_shutdown = 1;
2465 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2466 		if ((sc = gp->softc) == NULL)
2467 			continue;
2468 		g_topology_unlock();
2469 		sx_xlock(&sc->sc_lock);
2470 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2471 			g_raid_clean(vol, -1);
2472 		g_cancel_event(sc);
2473 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2474 		g_topology_lock();
2475 	}
2476 	g_topology_unlock();
2477 }
2478 
2479 static void
2480 g_raid_init(struct g_class *mp)
2481 {
2482 
2483 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2484 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2485 	if (g_raid_post_sync == NULL)
2486 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2487 	g_raid_started = 1;
2488 }
2489 
2490 static void
2491 g_raid_fini(struct g_class *mp)
2492 {
2493 
2494 	if (g_raid_post_sync != NULL)
2495 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2496 	g_raid_started = 0;
2497 }
2498 
2499 int
2500 g_raid_md_modevent(module_t mod, int type, void *arg)
2501 {
2502 	struct g_raid_md_class *class, *c, *nc;
2503 	int error;
2504 
2505 	error = 0;
2506 	class = arg;
2507 	switch (type) {
2508 	case MOD_LOAD:
2509 		c = LIST_FIRST(&g_raid_md_classes);
2510 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2511 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2512 		else {
2513 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2514 			    nc->mdc_priority < class->mdc_priority)
2515 				c = nc;
2516 			LIST_INSERT_AFTER(c, class, mdc_list);
2517 		}
2518 		if (g_raid_started)
2519 			g_retaste(&g_raid_class);
2520 		break;
2521 	case MOD_UNLOAD:
2522 		LIST_REMOVE(class, mdc_list);
2523 		break;
2524 	default:
2525 		error = EOPNOTSUPP;
2526 		break;
2527 	}
2528 
2529 	return (error);
2530 }
2531 
2532 int
2533 g_raid_tr_modevent(module_t mod, int type, void *arg)
2534 {
2535 	struct g_raid_tr_class *class, *c, *nc;
2536 	int error;
2537 
2538 	error = 0;
2539 	class = arg;
2540 	switch (type) {
2541 	case MOD_LOAD:
2542 		c = LIST_FIRST(&g_raid_tr_classes);
2543 		if (c == NULL || c->trc_priority > class->trc_priority)
2544 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2545 		else {
2546 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2547 			    nc->trc_priority < class->trc_priority)
2548 				c = nc;
2549 			LIST_INSERT_AFTER(c, class, trc_list);
2550 		}
2551 		break;
2552 	case MOD_UNLOAD:
2553 		LIST_REMOVE(class, trc_list);
2554 		break;
2555 	default:
2556 		error = EOPNOTSUPP;
2557 		break;
2558 	}
2559 
2560 	return (error);
2561 }
2562 
2563 /*
2564  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2565  * to reduce module priority, allowing submodules to register them first.
2566  */
2567 static moduledata_t g_raid_mod = {
2568 	"g_raid",
2569 	g_modevent,
2570 	&g_raid_class
2571 };
2572 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2573 MODULE_VERSION(geom_raid, 0);
2574