xref: /freebsd/sys/geom/raid/g_raid.c (revision fe267a559009cbf34f9341666fe4d88a92c02d5e)
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 <sys/proc.h>
47 #include <sys/kthread.h>
48 #include <sys/sched.h>
49 #include <geom/raid/g_raid.h>
50 #include "g_raid_md_if.h"
51 #include "g_raid_tr_if.h"
52 
53 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
54 
55 SYSCTL_DECL(_kern_geom);
56 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
57 int g_raid_enable = 1;
58 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN,
59     &g_raid_enable, 0, "Enable on-disk metadata taste");
60 u_int g_raid_aggressive_spare = 0;
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN,
62     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0,
65     "Debug level");
66 int g_raid_read_err_thresh = 10;
67 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN,
68     &g_raid_read_err_thresh, 0,
69     "Number of read errors equated to disk failure");
70 u_int g_raid_start_timeout = 30;
71 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN,
72     &g_raid_start_timeout, 0,
73     "Time to wait for all array components");
74 static u_int g_raid_clean_time = 5;
75 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN,
76     &g_raid_clean_time, 0, "Mark volume as clean when idling");
77 static u_int g_raid_disconnect_on_failure = 1;
78 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN,
79     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
80 static u_int g_raid_name_format = 0;
81 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN,
82     &g_raid_name_format, 0, "Providers name format.");
83 static u_int g_raid_idle_threshold = 1000000;
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN,
85     &g_raid_idle_threshold, 1000000,
86     "Time in microseconds to consider a volume idle.");
87 
88 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
89 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
90 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
91 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
92 } while (0)
93 
94 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
95     LIST_HEAD_INITIALIZER(g_raid_md_classes);
96 
97 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
98     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
99 
100 LIST_HEAD(, g_raid_volume) g_raid_volumes =
101     LIST_HEAD_INITIALIZER(g_raid_volumes);
102 
103 static eventhandler_tag g_raid_post_sync = NULL;
104 static int g_raid_started = 0;
105 static int g_raid_shutdown = 0;
106 
107 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
108     struct g_geom *gp);
109 static g_taste_t g_raid_taste;
110 static void g_raid_init(struct g_class *mp);
111 static void g_raid_fini(struct g_class *mp);
112 
113 struct g_class g_raid_class = {
114 	.name = G_RAID_CLASS_NAME,
115 	.version = G_VERSION,
116 	.ctlreq = g_raid_ctl,
117 	.taste = g_raid_taste,
118 	.destroy_geom = g_raid_destroy_geom,
119 	.init = g_raid_init,
120 	.fini = g_raid_fini
121 };
122 
123 static void g_raid_destroy_provider(struct g_raid_volume *vol);
124 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
125 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
126 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
127 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
128 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
129     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
130 static void g_raid_start(struct bio *bp);
131 static void g_raid_start_request(struct bio *bp);
132 static void g_raid_disk_done(struct bio *bp);
133 static void g_raid_poll(struct g_raid_softc *sc);
134 
135 static const char *
136 g_raid_node_event2str(int event)
137 {
138 
139 	switch (event) {
140 	case G_RAID_NODE_E_WAKE:
141 		return ("WAKE");
142 	case G_RAID_NODE_E_START:
143 		return ("START");
144 	default:
145 		return ("INVALID");
146 	}
147 }
148 
149 const char *
150 g_raid_disk_state2str(int state)
151 {
152 
153 	switch (state) {
154 	case G_RAID_DISK_S_NONE:
155 		return ("NONE");
156 	case G_RAID_DISK_S_OFFLINE:
157 		return ("OFFLINE");
158 	case G_RAID_DISK_S_DISABLED:
159 		return ("DISABLED");
160 	case G_RAID_DISK_S_FAILED:
161 		return ("FAILED");
162 	case G_RAID_DISK_S_STALE_FAILED:
163 		return ("STALE_FAILED");
164 	case G_RAID_DISK_S_SPARE:
165 		return ("SPARE");
166 	case G_RAID_DISK_S_STALE:
167 		return ("STALE");
168 	case G_RAID_DISK_S_ACTIVE:
169 		return ("ACTIVE");
170 	default:
171 		return ("INVALID");
172 	}
173 }
174 
175 static const char *
176 g_raid_disk_event2str(int event)
177 {
178 
179 	switch (event) {
180 	case G_RAID_DISK_E_DISCONNECTED:
181 		return ("DISCONNECTED");
182 	default:
183 		return ("INVALID");
184 	}
185 }
186 
187 const char *
188 g_raid_subdisk_state2str(int state)
189 {
190 
191 	switch (state) {
192 	case G_RAID_SUBDISK_S_NONE:
193 		return ("NONE");
194 	case G_RAID_SUBDISK_S_FAILED:
195 		return ("FAILED");
196 	case G_RAID_SUBDISK_S_NEW:
197 		return ("NEW");
198 	case G_RAID_SUBDISK_S_REBUILD:
199 		return ("REBUILD");
200 	case G_RAID_SUBDISK_S_UNINITIALIZED:
201 		return ("UNINITIALIZED");
202 	case G_RAID_SUBDISK_S_STALE:
203 		return ("STALE");
204 	case G_RAID_SUBDISK_S_RESYNC:
205 		return ("RESYNC");
206 	case G_RAID_SUBDISK_S_ACTIVE:
207 		return ("ACTIVE");
208 	default:
209 		return ("INVALID");
210 	}
211 }
212 
213 static const char *
214 g_raid_subdisk_event2str(int event)
215 {
216 
217 	switch (event) {
218 	case G_RAID_SUBDISK_E_NEW:
219 		return ("NEW");
220 	case G_RAID_SUBDISK_E_FAILED:
221 		return ("FAILED");
222 	case G_RAID_SUBDISK_E_DISCONNECTED:
223 		return ("DISCONNECTED");
224 	default:
225 		return ("INVALID");
226 	}
227 }
228 
229 const char *
230 g_raid_volume_state2str(int state)
231 {
232 
233 	switch (state) {
234 	case G_RAID_VOLUME_S_STARTING:
235 		return ("STARTING");
236 	case G_RAID_VOLUME_S_BROKEN:
237 		return ("BROKEN");
238 	case G_RAID_VOLUME_S_DEGRADED:
239 		return ("DEGRADED");
240 	case G_RAID_VOLUME_S_SUBOPTIMAL:
241 		return ("SUBOPTIMAL");
242 	case G_RAID_VOLUME_S_OPTIMAL:
243 		return ("OPTIMAL");
244 	case G_RAID_VOLUME_S_UNSUPPORTED:
245 		return ("UNSUPPORTED");
246 	case G_RAID_VOLUME_S_STOPPED:
247 		return ("STOPPED");
248 	default:
249 		return ("INVALID");
250 	}
251 }
252 
253 static const char *
254 g_raid_volume_event2str(int event)
255 {
256 
257 	switch (event) {
258 	case G_RAID_VOLUME_E_UP:
259 		return ("UP");
260 	case G_RAID_VOLUME_E_DOWN:
261 		return ("DOWN");
262 	case G_RAID_VOLUME_E_START:
263 		return ("START");
264 	case G_RAID_VOLUME_E_STARTMD:
265 		return ("STARTMD");
266 	default:
267 		return ("INVALID");
268 	}
269 }
270 
271 const char *
272 g_raid_volume_level2str(int level, int qual)
273 {
274 
275 	switch (level) {
276 	case G_RAID_VOLUME_RL_RAID0:
277 		return ("RAID0");
278 	case G_RAID_VOLUME_RL_RAID1:
279 		return ("RAID1");
280 	case G_RAID_VOLUME_RL_RAID3:
281 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
282 			return ("RAID3-P0");
283 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
284 			return ("RAID3-PN");
285 		return ("RAID3");
286 	case G_RAID_VOLUME_RL_RAID4:
287 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
288 			return ("RAID4-P0");
289 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
290 			return ("RAID4-PN");
291 		return ("RAID4");
292 	case G_RAID_VOLUME_RL_RAID5:
293 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
294 			return ("RAID5-RA");
295 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
296 			return ("RAID5-RS");
297 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
298 			return ("RAID5-LA");
299 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
300 			return ("RAID5-LS");
301 		return ("RAID5");
302 	case G_RAID_VOLUME_RL_RAID6:
303 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
304 			return ("RAID6-RA");
305 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
306 			return ("RAID6-RS");
307 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
308 			return ("RAID6-LA");
309 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
310 			return ("RAID6-LS");
311 		return ("RAID6");
312 	case G_RAID_VOLUME_RL_RAIDMDF:
313 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
314 			return ("RAIDMDF-RA");
315 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
316 			return ("RAIDMDF-RS");
317 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
318 			return ("RAIDMDF-LA");
319 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
320 			return ("RAIDMDF-LS");
321 		return ("RAIDMDF");
322 	case G_RAID_VOLUME_RL_RAID1E:
323 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
324 			return ("RAID1E-A");
325 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
326 			return ("RAID1E-O");
327 		return ("RAID1E");
328 	case G_RAID_VOLUME_RL_SINGLE:
329 		return ("SINGLE");
330 	case G_RAID_VOLUME_RL_CONCAT:
331 		return ("CONCAT");
332 	case G_RAID_VOLUME_RL_RAID5E:
333 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
334 			return ("RAID5E-RA");
335 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
336 			return ("RAID5E-RS");
337 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
338 			return ("RAID5E-LA");
339 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
340 			return ("RAID5E-LS");
341 		return ("RAID5E");
342 	case G_RAID_VOLUME_RL_RAID5EE:
343 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
344 			return ("RAID5EE-RA");
345 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
346 			return ("RAID5EE-RS");
347 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
348 			return ("RAID5EE-LA");
349 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
350 			return ("RAID5EE-LS");
351 		return ("RAID5EE");
352 	case G_RAID_VOLUME_RL_RAID5R:
353 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
354 			return ("RAID5R-RA");
355 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
356 			return ("RAID5R-RS");
357 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
358 			return ("RAID5R-LA");
359 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
360 			return ("RAID5R-LS");
361 		return ("RAID5E");
362 	default:
363 		return ("UNKNOWN");
364 	}
365 }
366 
367 int
368 g_raid_volume_str2level(const char *str, int *level, int *qual)
369 {
370 
371 	*level = G_RAID_VOLUME_RL_UNKNOWN;
372 	*qual = G_RAID_VOLUME_RLQ_NONE;
373 	if (strcasecmp(str, "RAID0") == 0)
374 		*level = G_RAID_VOLUME_RL_RAID0;
375 	else if (strcasecmp(str, "RAID1") == 0)
376 		*level = G_RAID_VOLUME_RL_RAID1;
377 	else if (strcasecmp(str, "RAID3-P0") == 0) {
378 		*level = G_RAID_VOLUME_RL_RAID3;
379 		*qual = G_RAID_VOLUME_RLQ_R3P0;
380 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
381 		   strcasecmp(str, "RAID3") == 0) {
382 		*level = G_RAID_VOLUME_RL_RAID3;
383 		*qual = G_RAID_VOLUME_RLQ_R3PN;
384 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
385 		*level = G_RAID_VOLUME_RL_RAID4;
386 		*qual = G_RAID_VOLUME_RLQ_R4P0;
387 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
388 		   strcasecmp(str, "RAID4") == 0) {
389 		*level = G_RAID_VOLUME_RL_RAID4;
390 		*qual = G_RAID_VOLUME_RLQ_R4PN;
391 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
392 		*level = G_RAID_VOLUME_RL_RAID5;
393 		*qual = G_RAID_VOLUME_RLQ_R5RA;
394 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
395 		*level = G_RAID_VOLUME_RL_RAID5;
396 		*qual = G_RAID_VOLUME_RLQ_R5RS;
397 	} else if (strcasecmp(str, "RAID5") == 0 ||
398 		   strcasecmp(str, "RAID5-LA") == 0) {
399 		*level = G_RAID_VOLUME_RL_RAID5;
400 		*qual = G_RAID_VOLUME_RLQ_R5LA;
401 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
402 		*level = G_RAID_VOLUME_RL_RAID5;
403 		*qual = G_RAID_VOLUME_RLQ_R5LS;
404 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
405 		*level = G_RAID_VOLUME_RL_RAID6;
406 		*qual = G_RAID_VOLUME_RLQ_R6RA;
407 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
408 		*level = G_RAID_VOLUME_RL_RAID6;
409 		*qual = G_RAID_VOLUME_RLQ_R6RS;
410 	} else if (strcasecmp(str, "RAID6") == 0 ||
411 		   strcasecmp(str, "RAID6-LA") == 0) {
412 		*level = G_RAID_VOLUME_RL_RAID6;
413 		*qual = G_RAID_VOLUME_RLQ_R6LA;
414 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
415 		*level = G_RAID_VOLUME_RL_RAID6;
416 		*qual = G_RAID_VOLUME_RLQ_R6LS;
417 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
418 		*level = G_RAID_VOLUME_RL_RAIDMDF;
419 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
420 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
421 		*level = G_RAID_VOLUME_RL_RAIDMDF;
422 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
423 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
424 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
425 		*level = G_RAID_VOLUME_RL_RAIDMDF;
426 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
427 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
428 		*level = G_RAID_VOLUME_RL_RAIDMDF;
429 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
430 	} else if (strcasecmp(str, "RAID10") == 0 ||
431 		   strcasecmp(str, "RAID1E") == 0 ||
432 		   strcasecmp(str, "RAID1E-A") == 0) {
433 		*level = G_RAID_VOLUME_RL_RAID1E;
434 		*qual = G_RAID_VOLUME_RLQ_R1EA;
435 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
436 		*level = G_RAID_VOLUME_RL_RAID1E;
437 		*qual = G_RAID_VOLUME_RLQ_R1EO;
438 	} else if (strcasecmp(str, "SINGLE") == 0)
439 		*level = G_RAID_VOLUME_RL_SINGLE;
440 	else if (strcasecmp(str, "CONCAT") == 0)
441 		*level = G_RAID_VOLUME_RL_CONCAT;
442 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
443 		*level = G_RAID_VOLUME_RL_RAID5E;
444 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
445 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
446 		*level = G_RAID_VOLUME_RL_RAID5E;
447 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
448 	} else if (strcasecmp(str, "RAID5E") == 0 ||
449 		   strcasecmp(str, "RAID5E-LA") == 0) {
450 		*level = G_RAID_VOLUME_RL_RAID5E;
451 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
452 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
453 		*level = G_RAID_VOLUME_RL_RAID5E;
454 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
455 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
456 		*level = G_RAID_VOLUME_RL_RAID5EE;
457 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
458 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
459 		*level = G_RAID_VOLUME_RL_RAID5EE;
460 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
461 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
462 		   strcasecmp(str, "RAID5EE-LA") == 0) {
463 		*level = G_RAID_VOLUME_RL_RAID5EE;
464 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
465 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
466 		*level = G_RAID_VOLUME_RL_RAID5EE;
467 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
468 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
469 		*level = G_RAID_VOLUME_RL_RAID5R;
470 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
471 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
472 		*level = G_RAID_VOLUME_RL_RAID5R;
473 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
474 	} else if (strcasecmp(str, "RAID5R") == 0 ||
475 		   strcasecmp(str, "RAID5R-LA") == 0) {
476 		*level = G_RAID_VOLUME_RL_RAID5R;
477 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
478 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
479 		*level = G_RAID_VOLUME_RL_RAID5R;
480 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
481 	} else
482 		return (-1);
483 	return (0);
484 }
485 
486 const char *
487 g_raid_get_diskname(struct g_raid_disk *disk)
488 {
489 
490 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
491 		return ("[unknown]");
492 	return (disk->d_consumer->provider->name);
493 }
494 
495 void
496 g_raid_get_disk_info(struct g_raid_disk *disk)
497 {
498 	struct g_consumer *cp = disk->d_consumer;
499 	int error, len;
500 
501 	/* Read kernel dumping information. */
502 	disk->d_kd.offset = 0;
503 	disk->d_kd.length = OFF_MAX;
504 	len = sizeof(disk->d_kd);
505 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
506 	if (error)
507 		disk->d_kd.di.dumper = NULL;
508 	if (disk->d_kd.di.dumper == NULL)
509 		G_RAID_DEBUG1(2, disk->d_softc,
510 		    "Dumping not supported by %s: %d.",
511 		    cp->provider->name, error);
512 
513 	/* Read BIO_DELETE support. */
514 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
515 	if (error)
516 		disk->d_candelete = 0;
517 	if (!disk->d_candelete)
518 		G_RAID_DEBUG1(2, disk->d_softc,
519 		    "BIO_DELETE not supported by %s: %d.",
520 		    cp->provider->name, error);
521 }
522 
523 void
524 g_raid_report_disk_state(struct g_raid_disk *disk)
525 {
526 	struct g_raid_subdisk *sd;
527 	int len, state;
528 	uint32_t s;
529 
530 	if (disk->d_consumer == NULL)
531 		return;
532 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
533 		s = G_STATE_ACTIVE; /* XXX */
534 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
535 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
536 		s = G_STATE_FAILED;
537 	} else {
538 		state = G_RAID_SUBDISK_S_ACTIVE;
539 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
540 			if (sd->sd_state < state)
541 				state = sd->sd_state;
542 		}
543 		if (state == G_RAID_SUBDISK_S_FAILED)
544 			s = G_STATE_FAILED;
545 		else if (state == G_RAID_SUBDISK_S_NEW ||
546 		    state == G_RAID_SUBDISK_S_REBUILD)
547 			s = G_STATE_REBUILD;
548 		else if (state == G_RAID_SUBDISK_S_STALE ||
549 		    state == G_RAID_SUBDISK_S_RESYNC)
550 			s = G_STATE_RESYNC;
551 		else
552 			s = G_STATE_ACTIVE;
553 	}
554 	len = sizeof(s);
555 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
556 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
557 	    g_raid_get_diskname(disk), s);
558 }
559 
560 void
561 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
562 {
563 
564 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
565 	    g_raid_get_diskname(disk),
566 	    g_raid_disk_state2str(disk->d_state),
567 	    g_raid_disk_state2str(state));
568 	disk->d_state = state;
569 	g_raid_report_disk_state(disk);
570 }
571 
572 void
573 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
574 {
575 
576 	G_RAID_DEBUG1(0, sd->sd_softc,
577 	    "Subdisk %s:%d-%s state changed from %s to %s.",
578 	    sd->sd_volume->v_name, sd->sd_pos,
579 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
580 	    g_raid_subdisk_state2str(sd->sd_state),
581 	    g_raid_subdisk_state2str(state));
582 	sd->sd_state = state;
583 	if (sd->sd_disk)
584 		g_raid_report_disk_state(sd->sd_disk);
585 }
586 
587 void
588 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
589 {
590 
591 	G_RAID_DEBUG1(0, vol->v_softc,
592 	    "Volume %s state changed from %s to %s.",
593 	    vol->v_name,
594 	    g_raid_volume_state2str(vol->v_state),
595 	    g_raid_volume_state2str(state));
596 	vol->v_state = state;
597 }
598 
599 /*
600  * --- Events handling functions ---
601  * Events in geom_raid are used to maintain subdisks and volumes status
602  * from one thread to simplify locking.
603  */
604 static void
605 g_raid_event_free(struct g_raid_event *ep)
606 {
607 
608 	free(ep, M_RAID);
609 }
610 
611 int
612 g_raid_event_send(void *arg, int event, int flags)
613 {
614 	struct g_raid_softc *sc;
615 	struct g_raid_event *ep;
616 	int error;
617 
618 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
619 		sc = ((struct g_raid_volume *)arg)->v_softc;
620 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
621 		sc = ((struct g_raid_disk *)arg)->d_softc;
622 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
623 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
624 	} else {
625 		sc = arg;
626 	}
627 	ep = malloc(sizeof(*ep), M_RAID,
628 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
629 	if (ep == NULL)
630 		return (ENOMEM);
631 	ep->e_tgt = arg;
632 	ep->e_event = event;
633 	ep->e_flags = flags;
634 	ep->e_error = 0;
635 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
636 	mtx_lock(&sc->sc_queue_mtx);
637 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
638 	mtx_unlock(&sc->sc_queue_mtx);
639 	wakeup(sc);
640 
641 	if ((flags & G_RAID_EVENT_WAIT) == 0)
642 		return (0);
643 
644 	sx_assert(&sc->sc_lock, SX_XLOCKED);
645 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
646 	sx_xunlock(&sc->sc_lock);
647 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
648 		mtx_lock(&sc->sc_queue_mtx);
649 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
650 		    hz * 5);
651 	}
652 	error = ep->e_error;
653 	g_raid_event_free(ep);
654 	sx_xlock(&sc->sc_lock);
655 	return (error);
656 }
657 
658 static void
659 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
660 {
661 	struct g_raid_event *ep, *tmpep;
662 
663 	sx_assert(&sc->sc_lock, SX_XLOCKED);
664 
665 	mtx_lock(&sc->sc_queue_mtx);
666 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
667 		if (ep->e_tgt != tgt)
668 			continue;
669 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
670 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
671 			g_raid_event_free(ep);
672 		else {
673 			ep->e_error = ECANCELED;
674 			wakeup(ep);
675 		}
676 	}
677 	mtx_unlock(&sc->sc_queue_mtx);
678 }
679 
680 static int
681 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
682 {
683 	struct g_raid_event *ep;
684 	int	res = 0;
685 
686 	sx_assert(&sc->sc_lock, SX_XLOCKED);
687 
688 	mtx_lock(&sc->sc_queue_mtx);
689 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
690 		if (ep->e_tgt != tgt)
691 			continue;
692 		res = 1;
693 		break;
694 	}
695 	mtx_unlock(&sc->sc_queue_mtx);
696 	return (res);
697 }
698 
699 /*
700  * Return the number of disks in given state.
701  * If state is equal to -1, count all connected disks.
702  */
703 u_int
704 g_raid_ndisks(struct g_raid_softc *sc, int state)
705 {
706 	struct g_raid_disk *disk;
707 	u_int n;
708 
709 	sx_assert(&sc->sc_lock, SX_LOCKED);
710 
711 	n = 0;
712 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
713 		if (disk->d_state == state || state == -1)
714 			n++;
715 	}
716 	return (n);
717 }
718 
719 /*
720  * Return the number of subdisks in given state.
721  * If state is equal to -1, count all connected disks.
722  */
723 u_int
724 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
725 {
726 	struct g_raid_subdisk *subdisk;
727 	struct g_raid_softc *sc;
728 	u_int i, n ;
729 
730 	sc = vol->v_softc;
731 	sx_assert(&sc->sc_lock, SX_LOCKED);
732 
733 	n = 0;
734 	for (i = 0; i < vol->v_disks_count; i++) {
735 		subdisk = &vol->v_subdisks[i];
736 		if ((state == -1 &&
737 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
738 		    subdisk->sd_state == state)
739 			n++;
740 	}
741 	return (n);
742 }
743 
744 /*
745  * Return the first subdisk in given state.
746  * If state is equal to -1, then the first connected disks.
747  */
748 struct g_raid_subdisk *
749 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
750 {
751 	struct g_raid_subdisk *sd;
752 	struct g_raid_softc *sc;
753 	u_int i;
754 
755 	sc = vol->v_softc;
756 	sx_assert(&sc->sc_lock, SX_LOCKED);
757 
758 	for (i = 0; i < vol->v_disks_count; i++) {
759 		sd = &vol->v_subdisks[i];
760 		if ((state == -1 &&
761 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
762 		    sd->sd_state == state)
763 			return (sd);
764 	}
765 	return (NULL);
766 }
767 
768 struct g_consumer *
769 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
770 {
771 	struct g_consumer *cp;
772 	struct g_provider *pp;
773 
774 	g_topology_assert();
775 
776 	if (strncmp(name, "/dev/", 5) == 0)
777 		name += 5;
778 	pp = g_provider_by_name(name);
779 	if (pp == NULL)
780 		return (NULL);
781 	cp = g_new_consumer(sc->sc_geom);
782 	cp->flags |= G_CF_DIRECT_RECEIVE;
783 	if (g_attach(cp, pp) != 0) {
784 		g_destroy_consumer(cp);
785 		return (NULL);
786 	}
787 	if (g_access(cp, 1, 1, 1) != 0) {
788 		g_detach(cp);
789 		g_destroy_consumer(cp);
790 		return (NULL);
791 	}
792 	return (cp);
793 }
794 
795 static u_int
796 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
797 {
798 	struct bio *bp;
799 	u_int nreqs = 0;
800 
801 	mtx_lock(&sc->sc_queue_mtx);
802 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
803 		if (bp->bio_from == cp)
804 			nreqs++;
805 	}
806 	mtx_unlock(&sc->sc_queue_mtx);
807 	return (nreqs);
808 }
809 
810 u_int
811 g_raid_nopens(struct g_raid_softc *sc)
812 {
813 	struct g_raid_volume *vol;
814 	u_int opens;
815 
816 	opens = 0;
817 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
818 		if (vol->v_provider_open != 0)
819 			opens++;
820 	}
821 	return (opens);
822 }
823 
824 static int
825 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
826 {
827 
828 	if (cp->index > 0) {
829 		G_RAID_DEBUG1(2, sc,
830 		    "I/O requests for %s exist, can't destroy it now.",
831 		    cp->provider->name);
832 		return (1);
833 	}
834 	if (g_raid_nrequests(sc, cp) > 0) {
835 		G_RAID_DEBUG1(2, sc,
836 		    "I/O requests for %s in queue, can't destroy it now.",
837 		    cp->provider->name);
838 		return (1);
839 	}
840 	return (0);
841 }
842 
843 static void
844 g_raid_destroy_consumer(void *arg, int flags __unused)
845 {
846 	struct g_consumer *cp;
847 
848 	g_topology_assert();
849 
850 	cp = arg;
851 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
852 	g_detach(cp);
853 	g_destroy_consumer(cp);
854 }
855 
856 void
857 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
858 {
859 	struct g_provider *pp;
860 	int retaste_wait;
861 
862 	g_topology_assert_not();
863 
864 	g_topology_lock();
865 	cp->private = NULL;
866 	if (g_raid_consumer_is_busy(sc, cp))
867 		goto out;
868 	pp = cp->provider;
869 	retaste_wait = 0;
870 	if (cp->acw == 1) {
871 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
872 			retaste_wait = 1;
873 	}
874 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
875 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
876 	if (retaste_wait) {
877 		/*
878 		 * After retaste event was send (inside g_access()), we can send
879 		 * event to detach and destroy consumer.
880 		 * A class, which has consumer to the given provider connected
881 		 * will not receive retaste event for the provider.
882 		 * This is the way how I ignore retaste events when I close
883 		 * consumers opened for write: I detach and destroy consumer
884 		 * after retaste event is sent.
885 		 */
886 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
887 		goto out;
888 	}
889 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
890 	g_detach(cp);
891 	g_destroy_consumer(cp);
892 out:
893 	g_topology_unlock();
894 }
895 
896 static void
897 g_raid_orphan(struct g_consumer *cp)
898 {
899 	struct g_raid_disk *disk;
900 
901 	g_topology_assert();
902 
903 	disk = cp->private;
904 	if (disk == NULL)
905 		return;
906 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
907 	    G_RAID_EVENT_DISK);
908 }
909 
910 static void
911 g_raid_clean(struct g_raid_volume *vol, int acw)
912 {
913 	struct g_raid_softc *sc;
914 	int timeout;
915 
916 	sc = vol->v_softc;
917 	g_topology_assert_not();
918 	sx_assert(&sc->sc_lock, SX_XLOCKED);
919 
920 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
921 //		return;
922 	if (!vol->v_dirty)
923 		return;
924 	if (vol->v_writes > 0)
925 		return;
926 	if (acw > 0 || (acw == -1 &&
927 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
928 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
929 		if (!g_raid_shutdown && timeout > 0)
930 			return;
931 	}
932 	vol->v_dirty = 0;
933 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
934 	    vol->v_name);
935 	g_raid_write_metadata(sc, vol, NULL, NULL);
936 }
937 
938 static void
939 g_raid_dirty(struct g_raid_volume *vol)
940 {
941 	struct g_raid_softc *sc;
942 
943 	sc = vol->v_softc;
944 	g_topology_assert_not();
945 	sx_assert(&sc->sc_lock, SX_XLOCKED);
946 
947 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
948 //		return;
949 	vol->v_dirty = 1;
950 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
951 	    vol->v_name);
952 	g_raid_write_metadata(sc, vol, NULL, NULL);
953 }
954 
955 void
956 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
957 {
958 	struct g_raid_softc *sc;
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 	sc = vol->v_softc;
967 
968 	/*
969 	 * Allocate all bios before sending any request, so we can return
970 	 * ENOMEM in nice and clean way.
971 	 */
972 	bioq_init(&queue);
973 	for (i = 0; i < vol->v_disks_count; i++) {
974 		sd = &vol->v_subdisks[i];
975 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
976 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
977 			continue;
978 		cbp = g_clone_bio(bp);
979 		if (cbp == NULL)
980 			goto failure;
981 		cbp->bio_caller1 = sd;
982 		bioq_insert_tail(&queue, cbp);
983 	}
984 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
985 		sd = cbp->bio_caller1;
986 		cbp->bio_caller1 = NULL;
987 		g_raid_subdisk_iostart(sd, cbp);
988 	}
989 	return;
990 failure:
991 	while ((cbp = bioq_takefirst(&queue)) != NULL)
992 		g_destroy_bio(cbp);
993 	if (bp->bio_error == 0)
994 		bp->bio_error = ENOMEM;
995 	g_raid_iodone(bp, bp->bio_error);
996 }
997 
998 static void
999 g_raid_tr_kerneldump_common_done(struct bio *bp)
1000 {
1001 
1002 	bp->bio_flags |= BIO_DONE;
1003 }
1004 
1005 int
1006 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1007     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1008 {
1009 	struct g_raid_softc *sc;
1010 	struct g_raid_volume *vol;
1011 	struct bio bp;
1012 
1013 	vol = tr->tro_volume;
1014 	sc = vol->v_softc;
1015 
1016 	g_reset_bio(&bp);
1017 	bp.bio_cmd = BIO_WRITE;
1018 	bp.bio_done = g_raid_tr_kerneldump_common_done;
1019 	bp.bio_attribute = NULL;
1020 	bp.bio_offset = offset;
1021 	bp.bio_length = length;
1022 	bp.bio_data = virtual;
1023 	bp.bio_to = vol->v_provider;
1024 
1025 	g_raid_start(&bp);
1026 	while (!(bp.bio_flags & BIO_DONE)) {
1027 		G_RAID_DEBUG1(4, sc, "Poll...");
1028 		g_raid_poll(sc);
1029 		DELAY(10);
1030 	}
1031 
1032 	return (bp.bio_error != 0 ? EIO : 0);
1033 }
1034 
1035 static int
1036 g_raid_dump(void *arg,
1037     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1038 {
1039 	struct g_raid_volume *vol;
1040 	int error;
1041 
1042 	vol = (struct g_raid_volume *)arg;
1043 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1044 	    (long long unsigned)offset, (long long unsigned)length);
1045 
1046 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1047 	    virtual, physical, offset, length);
1048 	return (error);
1049 }
1050 
1051 static void
1052 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1053 {
1054 	struct g_kerneldump *gkd;
1055 	struct g_provider *pp;
1056 	struct g_raid_volume *vol;
1057 
1058 	gkd = (struct g_kerneldump*)bp->bio_data;
1059 	pp = bp->bio_to;
1060 	vol = pp->private;
1061 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1062 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1063 	gkd->di.dumper = g_raid_dump;
1064 	gkd->di.priv = vol;
1065 	gkd->di.blocksize = vol->v_sectorsize;
1066 	gkd->di.maxiosize = DFLTPHYS;
1067 	gkd->di.mediaoffset = gkd->offset;
1068 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1069 		gkd->length = vol->v_mediasize - gkd->offset;
1070 	gkd->di.mediasize = gkd->length;
1071 	g_io_deliver(bp, 0);
1072 }
1073 
1074 static void
1075 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1076 {
1077 	struct g_provider *pp;
1078 	struct g_raid_volume *vol;
1079 	struct g_raid_subdisk *sd;
1080 	int *val;
1081 	int i;
1082 
1083 	val = (int *)bp->bio_data;
1084 	pp = bp->bio_to;
1085 	vol = pp->private;
1086 	*val = 0;
1087 	for (i = 0; i < vol->v_disks_count; i++) {
1088 		sd = &vol->v_subdisks[i];
1089 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1090 			continue;
1091 		if (sd->sd_disk->d_candelete) {
1092 			*val = 1;
1093 			break;
1094 		}
1095 	}
1096 	g_io_deliver(bp, 0);
1097 }
1098 
1099 static void
1100 g_raid_start(struct bio *bp)
1101 {
1102 	struct g_raid_softc *sc;
1103 
1104 	sc = bp->bio_to->geom->softc;
1105 	/*
1106 	 * If sc == NULL or there are no valid disks, provider's error
1107 	 * should be set and g_raid_start() should not be called at all.
1108 	 */
1109 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1110 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1111 //	    bp->bio_to->error, bp->bio_to->name));
1112 	G_RAID_LOGREQ(3, bp, "Request received.");
1113 
1114 	switch (bp->bio_cmd) {
1115 	case BIO_READ:
1116 	case BIO_WRITE:
1117 	case BIO_DELETE:
1118 	case BIO_FLUSH:
1119 		break;
1120 	case BIO_GETATTR:
1121 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1122 			g_raid_candelete(sc, bp);
1123 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1124 			g_raid_kerneldump(sc, bp);
1125 		else
1126 			g_io_deliver(bp, EOPNOTSUPP);
1127 		return;
1128 	default:
1129 		g_io_deliver(bp, EOPNOTSUPP);
1130 		return;
1131 	}
1132 	mtx_lock(&sc->sc_queue_mtx);
1133 	bioq_insert_tail(&sc->sc_queue, bp);
1134 	mtx_unlock(&sc->sc_queue_mtx);
1135 	if (!dumping) {
1136 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1137 		wakeup(sc);
1138 	}
1139 }
1140 
1141 static int
1142 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1143 {
1144 	/*
1145 	 * 5 cases:
1146 	 * (1) bp entirely below NO
1147 	 * (2) bp entirely above NO
1148 	 * (3) bp start below, but end in range YES
1149 	 * (4) bp entirely within YES
1150 	 * (5) bp starts within, ends above YES
1151 	 *
1152 	 * lock range 10-19 (offset 10 length 10)
1153 	 * (1) 1-5: first if kicks it out
1154 	 * (2) 30-35: second if kicks it out
1155 	 * (3) 5-15: passes both ifs
1156 	 * (4) 12-14: passes both ifs
1157 	 * (5) 19-20: passes both
1158 	 */
1159 	off_t lend = lstart + len - 1;
1160 	off_t bstart = bp->bio_offset;
1161 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1162 
1163 	if (bend < lstart)
1164 		return (0);
1165 	if (lend < bstart)
1166 		return (0);
1167 	return (1);
1168 }
1169 
1170 static int
1171 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1172 {
1173 	struct g_raid_lock *lp;
1174 
1175 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1176 
1177 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1178 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1179 			return (1);
1180 	}
1181 	return (0);
1182 }
1183 
1184 static void
1185 g_raid_start_request(struct bio *bp)
1186 {
1187 	struct g_raid_softc *sc;
1188 	struct g_raid_volume *vol;
1189 
1190 	sc = bp->bio_to->geom->softc;
1191 	sx_assert(&sc->sc_lock, SX_LOCKED);
1192 	vol = bp->bio_to->private;
1193 
1194 	/*
1195 	 * Check to see if this item is in a locked range.  If so,
1196 	 * queue it to our locked queue and return.  We'll requeue
1197 	 * it when the range is unlocked.  Internal I/O for the
1198 	 * rebuild/rescan/recovery process is excluded from this
1199 	 * check so we can actually do the recovery.
1200 	 */
1201 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1202 	    g_raid_is_in_locked_range(vol, bp)) {
1203 		G_RAID_LOGREQ(3, bp, "Defer request.");
1204 		bioq_insert_tail(&vol->v_locked, bp);
1205 		return;
1206 	}
1207 
1208 	/*
1209 	 * If we're actually going to do the write/delete, then
1210 	 * update the idle stats for the volume.
1211 	 */
1212 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1213 		if (!vol->v_dirty)
1214 			g_raid_dirty(vol);
1215 		vol->v_writes++;
1216 	}
1217 
1218 	/*
1219 	 * Put request onto inflight queue, so we can check if new
1220 	 * synchronization requests don't collide with it.  Then tell
1221 	 * the transformation layer to start the I/O.
1222 	 */
1223 	bioq_insert_tail(&vol->v_inflight, bp);
1224 	G_RAID_LOGREQ(4, bp, "Request started");
1225 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1226 }
1227 
1228 static void
1229 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1230 {
1231 	off_t off, len;
1232 	struct bio *nbp;
1233 	struct g_raid_lock *lp;
1234 
1235 	vol->v_pending_lock = 0;
1236 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1237 		if (lp->l_pending) {
1238 			off = lp->l_offset;
1239 			len = lp->l_length;
1240 			lp->l_pending = 0;
1241 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1242 				if (g_raid_bio_overlaps(nbp, off, len))
1243 					lp->l_pending++;
1244 			}
1245 			if (lp->l_pending) {
1246 				vol->v_pending_lock = 1;
1247 				G_RAID_DEBUG1(4, vol->v_softc,
1248 				    "Deferred lock(%jd, %jd) has %d pending",
1249 				    (intmax_t)off, (intmax_t)(off + len),
1250 				    lp->l_pending);
1251 				continue;
1252 			}
1253 			G_RAID_DEBUG1(4, vol->v_softc,
1254 			    "Deferred lock of %jd to %jd completed",
1255 			    (intmax_t)off, (intmax_t)(off + len));
1256 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1257 		}
1258 	}
1259 }
1260 
1261 void
1262 g_raid_iodone(struct bio *bp, int error)
1263 {
1264 	struct g_raid_softc *sc;
1265 	struct g_raid_volume *vol;
1266 
1267 	sc = bp->bio_to->geom->softc;
1268 	sx_assert(&sc->sc_lock, SX_LOCKED);
1269 	vol = bp->bio_to->private;
1270 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1271 
1272 	/* Update stats if we done write/delete. */
1273 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1274 		vol->v_writes--;
1275 		vol->v_last_write = time_uptime;
1276 	}
1277 
1278 	bioq_remove(&vol->v_inflight, bp);
1279 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1280 		g_raid_finish_with_locked_ranges(vol, bp);
1281 	getmicrouptime(&vol->v_last_done);
1282 	g_io_deliver(bp, error);
1283 }
1284 
1285 int
1286 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1287     struct bio *ignore, void *argp)
1288 {
1289 	struct g_raid_softc *sc;
1290 	struct g_raid_lock *lp;
1291 	struct bio *bp;
1292 
1293 	sc = vol->v_softc;
1294 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1295 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1296 	lp->l_offset = off;
1297 	lp->l_length = len;
1298 	lp->l_callback_arg = argp;
1299 
1300 	lp->l_pending = 0;
1301 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1302 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1303 			lp->l_pending++;
1304 	}
1305 
1306 	/*
1307 	 * If there are any writes that are pending, we return EBUSY.  All
1308 	 * callers will have to wait until all pending writes clear.
1309 	 */
1310 	if (lp->l_pending > 0) {
1311 		vol->v_pending_lock = 1;
1312 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1313 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1314 		return (EBUSY);
1315 	}
1316 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1317 	    (intmax_t)off, (intmax_t)(off+len));
1318 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1319 	return (0);
1320 }
1321 
1322 int
1323 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1324 {
1325 	struct g_raid_lock *lp;
1326 	struct g_raid_softc *sc;
1327 	struct bio *bp;
1328 
1329 	sc = vol->v_softc;
1330 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1331 		if (lp->l_offset == off && lp->l_length == len) {
1332 			LIST_REMOVE(lp, l_next);
1333 			/* XXX
1334 			 * Right now we just put them all back on the queue
1335 			 * and hope for the best.  We hope this because any
1336 			 * locked ranges will go right back on this list
1337 			 * when the worker thread runs.
1338 			 * XXX
1339 			 */
1340 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1341 			    (intmax_t)lp->l_offset,
1342 			    (intmax_t)(lp->l_offset+lp->l_length));
1343 			mtx_lock(&sc->sc_queue_mtx);
1344 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1345 				bioq_insert_tail(&sc->sc_queue, bp);
1346 			mtx_unlock(&sc->sc_queue_mtx);
1347 			free(lp, M_RAID);
1348 			return (0);
1349 		}
1350 	}
1351 	return (EINVAL);
1352 }
1353 
1354 void
1355 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1356 {
1357 	struct g_consumer *cp;
1358 	struct g_raid_disk *disk, *tdisk;
1359 
1360 	bp->bio_caller1 = sd;
1361 
1362 	/*
1363 	 * Make sure that the disk is present. Generally it is a task of
1364 	 * transformation layers to not send requests to absent disks, but
1365 	 * it is better to be safe and report situation then sorry.
1366 	 */
1367 	if (sd->sd_disk == NULL) {
1368 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1369 nodisk:
1370 		bp->bio_from = NULL;
1371 		bp->bio_to = NULL;
1372 		bp->bio_error = ENXIO;
1373 		g_raid_disk_done(bp);
1374 		return;
1375 	}
1376 	disk = sd->sd_disk;
1377 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1378 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1379 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1380 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1381 		goto nodisk;
1382 	}
1383 
1384 	cp = disk->d_consumer;
1385 	bp->bio_from = cp;
1386 	bp->bio_to = cp->provider;
1387 	cp->index++;
1388 
1389 	/* Update average disks load. */
1390 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1391 		if (tdisk->d_consumer == NULL)
1392 			tdisk->d_load = 0;
1393 		else
1394 			tdisk->d_load = (tdisk->d_consumer->index *
1395 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1396 	}
1397 
1398 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1399 	if (dumping) {
1400 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1401 		if (bp->bio_cmd == BIO_WRITE) {
1402 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1403 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1404 		} else
1405 			bp->bio_error = EOPNOTSUPP;
1406 		g_raid_disk_done(bp);
1407 	} else {
1408 		bp->bio_done = g_raid_disk_done;
1409 		bp->bio_offset += sd->sd_offset;
1410 		G_RAID_LOGREQ(3, bp, "Sending request.");
1411 		g_io_request(bp, cp);
1412 	}
1413 }
1414 
1415 int
1416 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1417     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1418 {
1419 
1420 	if (sd->sd_disk == NULL)
1421 		return (ENXIO);
1422 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1423 		return (EOPNOTSUPP);
1424 	return (dump_write(&sd->sd_disk->d_kd.di,
1425 	    virtual, physical,
1426 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1427 	    length));
1428 }
1429 
1430 static void
1431 g_raid_disk_done(struct bio *bp)
1432 {
1433 	struct g_raid_softc *sc;
1434 	struct g_raid_subdisk *sd;
1435 
1436 	sd = bp->bio_caller1;
1437 	sc = sd->sd_softc;
1438 	mtx_lock(&sc->sc_queue_mtx);
1439 	bioq_insert_tail(&sc->sc_queue, bp);
1440 	mtx_unlock(&sc->sc_queue_mtx);
1441 	if (!dumping)
1442 		wakeup(sc);
1443 }
1444 
1445 static void
1446 g_raid_disk_done_request(struct bio *bp)
1447 {
1448 	struct g_raid_softc *sc;
1449 	struct g_raid_disk *disk;
1450 	struct g_raid_subdisk *sd;
1451 	struct g_raid_volume *vol;
1452 
1453 	g_topology_assert_not();
1454 
1455 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1456 	sd = bp->bio_caller1;
1457 	sc = sd->sd_softc;
1458 	vol = sd->sd_volume;
1459 	if (bp->bio_from != NULL) {
1460 		bp->bio_from->index--;
1461 		disk = bp->bio_from->private;
1462 		if (disk == NULL)
1463 			g_raid_kill_consumer(sc, bp->bio_from);
1464 	}
1465 	bp->bio_offset -= sd->sd_offset;
1466 
1467 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1468 }
1469 
1470 static void
1471 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1472 {
1473 
1474 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1475 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1476 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1477 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1478 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1479 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1480 	else
1481 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1482 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1483 		KASSERT(ep->e_error == 0,
1484 		    ("Error cannot be handled."));
1485 		g_raid_event_free(ep);
1486 	} else {
1487 		ep->e_flags |= G_RAID_EVENT_DONE;
1488 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1489 		mtx_lock(&sc->sc_queue_mtx);
1490 		wakeup(ep);
1491 		mtx_unlock(&sc->sc_queue_mtx);
1492 	}
1493 }
1494 
1495 /*
1496  * Worker thread.
1497  */
1498 static void
1499 g_raid_worker(void *arg)
1500 {
1501 	struct g_raid_softc *sc;
1502 	struct g_raid_event *ep;
1503 	struct g_raid_volume *vol;
1504 	struct bio *bp;
1505 	struct timeval now, t;
1506 	int timeout, rv;
1507 
1508 	sc = arg;
1509 	thread_lock(curthread);
1510 	sched_prio(curthread, PRIBIO);
1511 	thread_unlock(curthread);
1512 
1513 	sx_xlock(&sc->sc_lock);
1514 	for (;;) {
1515 		mtx_lock(&sc->sc_queue_mtx);
1516 		/*
1517 		 * First take a look at events.
1518 		 * This is important to handle events before any I/O requests.
1519 		 */
1520 		bp = NULL;
1521 		vol = NULL;
1522 		rv = 0;
1523 		ep = TAILQ_FIRST(&sc->sc_events);
1524 		if (ep != NULL)
1525 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1526 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1527 			;
1528 		else {
1529 			getmicrouptime(&now);
1530 			t = now;
1531 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1532 				if (bioq_first(&vol->v_inflight) == NULL &&
1533 				    vol->v_tr &&
1534 				    timevalcmp(&vol->v_last_done, &t, < ))
1535 					t = vol->v_last_done;
1536 			}
1537 			timevalsub(&t, &now);
1538 			timeout = g_raid_idle_threshold +
1539 			    t.tv_sec * 1000000 + t.tv_usec;
1540 			if (timeout > 0) {
1541 				/*
1542 				 * Two steps to avoid overflows at HZ=1000
1543 				 * and idle timeouts > 2.1s.  Some rounding
1544 				 * errors can occur, but they are < 1tick,
1545 				 * which is deemed to be close enough for
1546 				 * this purpose.
1547 				 */
1548 				int micpertic = 1000000 / hz;
1549 				timeout = (timeout + micpertic - 1) / micpertic;
1550 				sx_xunlock(&sc->sc_lock);
1551 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1552 				    PRIBIO | PDROP, "-", timeout);
1553 				sx_xlock(&sc->sc_lock);
1554 				goto process;
1555 			} else
1556 				rv = EWOULDBLOCK;
1557 		}
1558 		mtx_unlock(&sc->sc_queue_mtx);
1559 process:
1560 		if (ep != NULL) {
1561 			g_raid_handle_event(sc, ep);
1562 		} else if (bp != NULL) {
1563 			if (bp->bio_to != NULL &&
1564 			    bp->bio_to->geom == sc->sc_geom)
1565 				g_raid_start_request(bp);
1566 			else
1567 				g_raid_disk_done_request(bp);
1568 		} else if (rv == EWOULDBLOCK) {
1569 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1570 				g_raid_clean(vol, -1);
1571 				if (bioq_first(&vol->v_inflight) == NULL &&
1572 				    vol->v_tr) {
1573 					t.tv_sec = g_raid_idle_threshold / 1000000;
1574 					t.tv_usec = g_raid_idle_threshold % 1000000;
1575 					timevaladd(&t, &vol->v_last_done);
1576 					getmicrouptime(&now);
1577 					if (timevalcmp(&t, &now, <= )) {
1578 						G_RAID_TR_IDLE(vol->v_tr);
1579 						vol->v_last_done = now;
1580 					}
1581 				}
1582 			}
1583 		}
1584 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1585 			g_raid_destroy_node(sc, 1);	/* May not return. */
1586 	}
1587 }
1588 
1589 static void
1590 g_raid_poll(struct g_raid_softc *sc)
1591 {
1592 	struct g_raid_event *ep;
1593 	struct bio *bp;
1594 
1595 	sx_xlock(&sc->sc_lock);
1596 	mtx_lock(&sc->sc_queue_mtx);
1597 	/*
1598 	 * First take a look at events.
1599 	 * This is important to handle events before any I/O requests.
1600 	 */
1601 	ep = TAILQ_FIRST(&sc->sc_events);
1602 	if (ep != NULL) {
1603 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1604 		mtx_unlock(&sc->sc_queue_mtx);
1605 		g_raid_handle_event(sc, ep);
1606 		goto out;
1607 	}
1608 	bp = bioq_takefirst(&sc->sc_queue);
1609 	if (bp != NULL) {
1610 		mtx_unlock(&sc->sc_queue_mtx);
1611 		if (bp->bio_from == NULL ||
1612 		    bp->bio_from->geom != sc->sc_geom)
1613 			g_raid_start_request(bp);
1614 		else
1615 			g_raid_disk_done_request(bp);
1616 	}
1617 out:
1618 	sx_xunlock(&sc->sc_lock);
1619 }
1620 
1621 static void
1622 g_raid_launch_provider(struct g_raid_volume *vol)
1623 {
1624 	struct g_raid_disk *disk;
1625 	struct g_raid_subdisk *sd;
1626 	struct g_raid_softc *sc;
1627 	struct g_provider *pp;
1628 	char name[G_RAID_MAX_VOLUMENAME];
1629 	off_t off;
1630 	int i;
1631 
1632 	sc = vol->v_softc;
1633 	sx_assert(&sc->sc_lock, SX_LOCKED);
1634 
1635 	g_topology_lock();
1636 	/* Try to name provider with volume name. */
1637 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1638 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1639 	    g_provider_by_name(name) != NULL) {
1640 		/* Otherwise use sequential volume number. */
1641 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1642 	}
1643 
1644 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1645 	pp->flags |= G_PF_DIRECT_RECEIVE;
1646 	if (vol->v_tr->tro_class->trc_accept_unmapped) {
1647 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1648 		for (i = 0; i < vol->v_disks_count; i++) {
1649 			sd = &vol->v_subdisks[i];
1650 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1651 				continue;
1652 			if ((sd->sd_disk->d_consumer->provider->flags &
1653 			    G_PF_ACCEPT_UNMAPPED) == 0)
1654 				pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
1655 		}
1656 	}
1657 	pp->private = vol;
1658 	pp->mediasize = vol->v_mediasize;
1659 	pp->sectorsize = vol->v_sectorsize;
1660 	pp->stripesize = 0;
1661 	pp->stripeoffset = 0;
1662 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1663 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1664 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1665 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1666 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1667 		    disk->d_consumer != NULL &&
1668 		    disk->d_consumer->provider != NULL) {
1669 			pp->stripesize = disk->d_consumer->provider->stripesize;
1670 			off = disk->d_consumer->provider->stripeoffset;
1671 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1672 			if (off > 0)
1673 				pp->stripeoffset %= off;
1674 		}
1675 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1676 			pp->stripesize *= (vol->v_disks_count - 1);
1677 			pp->stripeoffset *= (vol->v_disks_count - 1);
1678 		}
1679 	} else
1680 		pp->stripesize = vol->v_strip_size;
1681 	vol->v_provider = pp;
1682 	g_error_provider(pp, 0);
1683 	g_topology_unlock();
1684 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1685 	    pp->name, vol->v_name);
1686 }
1687 
1688 static void
1689 g_raid_destroy_provider(struct g_raid_volume *vol)
1690 {
1691 	struct g_raid_softc *sc;
1692 	struct g_provider *pp;
1693 	struct bio *bp, *tmp;
1694 
1695 	g_topology_assert_not();
1696 	sc = vol->v_softc;
1697 	pp = vol->v_provider;
1698 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1699 
1700 	g_topology_lock();
1701 	g_error_provider(pp, ENXIO);
1702 	mtx_lock(&sc->sc_queue_mtx);
1703 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1704 		if (bp->bio_to != pp)
1705 			continue;
1706 		bioq_remove(&sc->sc_queue, bp);
1707 		g_io_deliver(bp, ENXIO);
1708 	}
1709 	mtx_unlock(&sc->sc_queue_mtx);
1710 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1711 	    pp->name, vol->v_name);
1712 	g_wither_provider(pp, ENXIO);
1713 	g_topology_unlock();
1714 	vol->v_provider = NULL;
1715 }
1716 
1717 /*
1718  * Update device state.
1719  */
1720 static int
1721 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1722 {
1723 	struct g_raid_softc *sc;
1724 
1725 	sc = vol->v_softc;
1726 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1727 
1728 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1729 	    g_raid_volume_event2str(event),
1730 	    vol->v_name);
1731 	switch (event) {
1732 	case G_RAID_VOLUME_E_DOWN:
1733 		if (vol->v_provider != NULL)
1734 			g_raid_destroy_provider(vol);
1735 		break;
1736 	case G_RAID_VOLUME_E_UP:
1737 		if (vol->v_provider == NULL)
1738 			g_raid_launch_provider(vol);
1739 		break;
1740 	case G_RAID_VOLUME_E_START:
1741 		if (vol->v_tr)
1742 			G_RAID_TR_START(vol->v_tr);
1743 		return (0);
1744 	default:
1745 		if (sc->sc_md)
1746 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1747 		return (0);
1748 	}
1749 
1750 	/* Manage root mount release. */
1751 	if (vol->v_starting) {
1752 		vol->v_starting = 0;
1753 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1754 		root_mount_rel(vol->v_rootmount);
1755 		vol->v_rootmount = NULL;
1756 	}
1757 	if (vol->v_stopping && vol->v_provider_open == 0)
1758 		g_raid_destroy_volume(vol);
1759 	return (0);
1760 }
1761 
1762 /*
1763  * Update subdisk state.
1764  */
1765 static int
1766 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1767 {
1768 	struct g_raid_softc *sc;
1769 	struct g_raid_volume *vol;
1770 
1771 	sc = sd->sd_softc;
1772 	vol = sd->sd_volume;
1773 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1774 
1775 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1776 	    g_raid_subdisk_event2str(event),
1777 	    vol->v_name, sd->sd_pos,
1778 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1779 	if (vol->v_tr)
1780 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1781 
1782 	return (0);
1783 }
1784 
1785 /*
1786  * Update disk state.
1787  */
1788 static int
1789 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1790 {
1791 	struct g_raid_softc *sc;
1792 
1793 	sc = disk->d_softc;
1794 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1795 
1796 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1797 	    g_raid_disk_event2str(event),
1798 	    g_raid_get_diskname(disk));
1799 
1800 	if (sc->sc_md)
1801 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1802 	return (0);
1803 }
1804 
1805 /*
1806  * Node event.
1807  */
1808 static int
1809 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1810 {
1811 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1812 
1813 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1814 	    g_raid_node_event2str(event));
1815 
1816 	if (event == G_RAID_NODE_E_WAKE)
1817 		return (0);
1818 	if (sc->sc_md)
1819 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1820 	return (0);
1821 }
1822 
1823 static int
1824 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1825 {
1826 	struct g_raid_volume *vol;
1827 	struct g_raid_softc *sc;
1828 	int dcw, opens, error = 0;
1829 
1830 	g_topology_assert();
1831 	sc = pp->geom->softc;
1832 	vol = pp->private;
1833 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1834 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1835 
1836 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1837 	    acr, acw, ace);
1838 	dcw = pp->acw + acw;
1839 
1840 	g_topology_unlock();
1841 	sx_xlock(&sc->sc_lock);
1842 	/* Deny new opens while dying. */
1843 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1844 		error = ENXIO;
1845 		goto out;
1846 	}
1847 	/* Deny write opens for read-only volumes. */
1848 	if (vol->v_read_only && acw > 0) {
1849 		error = EROFS;
1850 		goto out;
1851 	}
1852 	if (dcw == 0)
1853 		g_raid_clean(vol, dcw);
1854 	vol->v_provider_open += acr + acw + ace;
1855 	/* Handle delayed node destruction. */
1856 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1857 	    vol->v_provider_open == 0) {
1858 		/* Count open volumes. */
1859 		opens = g_raid_nopens(sc);
1860 		if (opens == 0) {
1861 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1862 			/* Wake up worker to make it selfdestruct. */
1863 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1864 		}
1865 	}
1866 	/* Handle open volume destruction. */
1867 	if (vol->v_stopping && vol->v_provider_open == 0)
1868 		g_raid_destroy_volume(vol);
1869 out:
1870 	sx_xunlock(&sc->sc_lock);
1871 	g_topology_lock();
1872 	return (error);
1873 }
1874 
1875 struct g_raid_softc *
1876 g_raid_create_node(struct g_class *mp,
1877     const char *name, struct g_raid_md_object *md)
1878 {
1879 	struct g_raid_softc *sc;
1880 	struct g_geom *gp;
1881 	int error;
1882 
1883 	g_topology_assert();
1884 	G_RAID_DEBUG(1, "Creating array %s.", name);
1885 
1886 	gp = g_new_geomf(mp, "%s", name);
1887 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1888 	gp->start = g_raid_start;
1889 	gp->orphan = g_raid_orphan;
1890 	gp->access = g_raid_access;
1891 	gp->dumpconf = g_raid_dumpconf;
1892 
1893 	sc->sc_md = md;
1894 	sc->sc_geom = gp;
1895 	sc->sc_flags = 0;
1896 	TAILQ_INIT(&sc->sc_volumes);
1897 	TAILQ_INIT(&sc->sc_disks);
1898 	sx_init(&sc->sc_lock, "graid:lock");
1899 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1900 	TAILQ_INIT(&sc->sc_events);
1901 	bioq_init(&sc->sc_queue);
1902 	gp->softc = sc;
1903 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1904 	    "g_raid %s", name);
1905 	if (error != 0) {
1906 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1907 		mtx_destroy(&sc->sc_queue_mtx);
1908 		sx_destroy(&sc->sc_lock);
1909 		g_destroy_geom(sc->sc_geom);
1910 		free(sc, M_RAID);
1911 		return (NULL);
1912 	}
1913 
1914 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1915 	return (sc);
1916 }
1917 
1918 struct g_raid_volume *
1919 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1920 {
1921 	struct g_raid_volume	*vol, *vol1;
1922 	int i;
1923 
1924 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1925 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1926 	vol->v_softc = sc;
1927 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1928 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1929 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1930 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1931 	vol->v_rotate_parity = 1;
1932 	bioq_init(&vol->v_inflight);
1933 	bioq_init(&vol->v_locked);
1934 	LIST_INIT(&vol->v_locks);
1935 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1936 		vol->v_subdisks[i].sd_softc = sc;
1937 		vol->v_subdisks[i].sd_volume = vol;
1938 		vol->v_subdisks[i].sd_pos = i;
1939 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1940 	}
1941 
1942 	/* Find free ID for this volume. */
1943 	g_topology_lock();
1944 	vol1 = vol;
1945 	if (id >= 0) {
1946 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1947 			if (vol1->v_global_id == id)
1948 				break;
1949 		}
1950 	}
1951 	if (vol1 != NULL) {
1952 		for (id = 0; ; id++) {
1953 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1954 				if (vol1->v_global_id == id)
1955 					break;
1956 			}
1957 			if (vol1 == NULL)
1958 				break;
1959 		}
1960 	}
1961 	vol->v_global_id = id;
1962 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1963 	g_topology_unlock();
1964 
1965 	/* Delay root mounting. */
1966 	vol->v_rootmount = root_mount_hold("GRAID");
1967 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1968 	vol->v_starting = 1;
1969 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1970 	return (vol);
1971 }
1972 
1973 struct g_raid_disk *
1974 g_raid_create_disk(struct g_raid_softc *sc)
1975 {
1976 	struct g_raid_disk	*disk;
1977 
1978 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1979 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1980 	disk->d_softc = sc;
1981 	disk->d_state = G_RAID_DISK_S_NONE;
1982 	TAILQ_INIT(&disk->d_subdisks);
1983 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1984 	return (disk);
1985 }
1986 
1987 int g_raid_start_volume(struct g_raid_volume *vol)
1988 {
1989 	struct g_raid_tr_class *class;
1990 	struct g_raid_tr_object *obj;
1991 	int status;
1992 
1993 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1994 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1995 		if (!class->trc_enable)
1996 			continue;
1997 		G_RAID_DEBUG1(2, vol->v_softc,
1998 		    "Tasting volume %s for %s transformation.",
1999 		    vol->v_name, class->name);
2000 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2001 		    M_WAITOK);
2002 		obj->tro_class = class;
2003 		obj->tro_volume = vol;
2004 		status = G_RAID_TR_TASTE(obj, vol);
2005 		if (status != G_RAID_TR_TASTE_FAIL)
2006 			break;
2007 		kobj_delete((kobj_t)obj, M_RAID);
2008 	}
2009 	if (class == NULL) {
2010 		G_RAID_DEBUG1(0, vol->v_softc,
2011 		    "No transformation module found for %s.",
2012 		    vol->v_name);
2013 		vol->v_tr = NULL;
2014 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2015 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2016 		    G_RAID_EVENT_VOLUME);
2017 		return (-1);
2018 	}
2019 	G_RAID_DEBUG1(2, vol->v_softc,
2020 	    "Transformation module %s chosen for %s.",
2021 	    class->name, vol->v_name);
2022 	vol->v_tr = obj;
2023 	return (0);
2024 }
2025 
2026 int
2027 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2028 {
2029 	struct g_raid_volume *vol, *tmpv;
2030 	struct g_raid_disk *disk, *tmpd;
2031 	int error = 0;
2032 
2033 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2034 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2035 		if (g_raid_destroy_volume(vol))
2036 			error = EBUSY;
2037 	}
2038 	if (error)
2039 		return (error);
2040 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2041 		if (g_raid_destroy_disk(disk))
2042 			error = EBUSY;
2043 	}
2044 	if (error)
2045 		return (error);
2046 	if (sc->sc_md) {
2047 		G_RAID_MD_FREE(sc->sc_md);
2048 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2049 		sc->sc_md = NULL;
2050 	}
2051 	if (sc->sc_geom != NULL) {
2052 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2053 		g_topology_lock();
2054 		sc->sc_geom->softc = NULL;
2055 		g_wither_geom(sc->sc_geom, ENXIO);
2056 		g_topology_unlock();
2057 		sc->sc_geom = NULL;
2058 	} else
2059 		G_RAID_DEBUG(1, "Array destroyed.");
2060 	if (worker) {
2061 		g_raid_event_cancel(sc, sc);
2062 		mtx_destroy(&sc->sc_queue_mtx);
2063 		sx_xunlock(&sc->sc_lock);
2064 		sx_destroy(&sc->sc_lock);
2065 		wakeup(&sc->sc_stopping);
2066 		free(sc, M_RAID);
2067 		curthread->td_pflags &= ~TDP_GEOM;
2068 		G_RAID_DEBUG(1, "Thread exiting.");
2069 		kproc_exit(0);
2070 	} else {
2071 		/* Wake up worker to make it selfdestruct. */
2072 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2073 	}
2074 	return (0);
2075 }
2076 
2077 int
2078 g_raid_destroy_volume(struct g_raid_volume *vol)
2079 {
2080 	struct g_raid_softc *sc;
2081 	struct g_raid_disk *disk;
2082 	int i;
2083 
2084 	sc = vol->v_softc;
2085 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2086 	vol->v_stopping = 1;
2087 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2088 		if (vol->v_tr) {
2089 			G_RAID_TR_STOP(vol->v_tr);
2090 			return (EBUSY);
2091 		} else
2092 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2093 	}
2094 	if (g_raid_event_check(sc, vol) != 0)
2095 		return (EBUSY);
2096 	if (vol->v_provider != NULL)
2097 		return (EBUSY);
2098 	if (vol->v_provider_open != 0)
2099 		return (EBUSY);
2100 	if (vol->v_tr) {
2101 		G_RAID_TR_FREE(vol->v_tr);
2102 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2103 		vol->v_tr = NULL;
2104 	}
2105 	if (vol->v_rootmount)
2106 		root_mount_rel(vol->v_rootmount);
2107 	g_topology_lock();
2108 	LIST_REMOVE(vol, v_global_next);
2109 	g_topology_unlock();
2110 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2111 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2112 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2113 		disk = vol->v_subdisks[i].sd_disk;
2114 		if (disk == NULL)
2115 			continue;
2116 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2117 	}
2118 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2119 	if (sc->sc_md)
2120 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2121 	g_raid_event_cancel(sc, vol);
2122 	free(vol, M_RAID);
2123 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2124 		/* Wake up worker to let it selfdestruct. */
2125 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2126 	}
2127 	return (0);
2128 }
2129 
2130 int
2131 g_raid_destroy_disk(struct g_raid_disk *disk)
2132 {
2133 	struct g_raid_softc *sc;
2134 	struct g_raid_subdisk *sd, *tmp;
2135 
2136 	sc = disk->d_softc;
2137 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2138 	if (disk->d_consumer) {
2139 		g_raid_kill_consumer(sc, disk->d_consumer);
2140 		disk->d_consumer = NULL;
2141 	}
2142 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2143 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2144 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2145 		    G_RAID_EVENT_SUBDISK);
2146 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2147 		sd->sd_disk = NULL;
2148 	}
2149 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2150 	if (sc->sc_md)
2151 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2152 	g_raid_event_cancel(sc, disk);
2153 	free(disk, M_RAID);
2154 	return (0);
2155 }
2156 
2157 int
2158 g_raid_destroy(struct g_raid_softc *sc, int how)
2159 {
2160 	int error, opens;
2161 
2162 	g_topology_assert_not();
2163 	if (sc == NULL)
2164 		return (ENXIO);
2165 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2166 
2167 	/* Count open volumes. */
2168 	opens = g_raid_nopens(sc);
2169 
2170 	/* React on some opened volumes. */
2171 	if (opens > 0) {
2172 		switch (how) {
2173 		case G_RAID_DESTROY_SOFT:
2174 			G_RAID_DEBUG1(1, sc,
2175 			    "%d volumes are still open.",
2176 			    opens);
2177 			sx_xunlock(&sc->sc_lock);
2178 			return (EBUSY);
2179 		case G_RAID_DESTROY_DELAYED:
2180 			G_RAID_DEBUG1(1, sc,
2181 			    "Array will be destroyed on last close.");
2182 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2183 			sx_xunlock(&sc->sc_lock);
2184 			return (EBUSY);
2185 		case G_RAID_DESTROY_HARD:
2186 			G_RAID_DEBUG1(1, sc,
2187 			    "%d volumes are still open.",
2188 			    opens);
2189 		}
2190 	}
2191 
2192 	/* Mark node for destruction. */
2193 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2194 	/* Wake up worker to let it selfdestruct. */
2195 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2196 	/* Sleep until node destroyed. */
2197 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2198 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2199 	return (error == EWOULDBLOCK ? EBUSY : 0);
2200 }
2201 
2202 static void
2203 g_raid_taste_orphan(struct g_consumer *cp)
2204 {
2205 
2206 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2207 	    cp->provider->name));
2208 }
2209 
2210 static struct g_geom *
2211 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2212 {
2213 	struct g_consumer *cp;
2214 	struct g_geom *gp, *geom;
2215 	struct g_raid_md_class *class;
2216 	struct g_raid_md_object *obj;
2217 	int status;
2218 
2219 	g_topology_assert();
2220 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2221 	if (!g_raid_enable)
2222 		return (NULL);
2223 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2224 
2225 	geom = NULL;
2226 	status = G_RAID_MD_TASTE_FAIL;
2227 	gp = g_new_geomf(mp, "raid:taste");
2228 	/*
2229 	 * This orphan function should be never called.
2230 	 */
2231 	gp->orphan = g_raid_taste_orphan;
2232 	cp = g_new_consumer(gp);
2233 	cp->flags |= G_CF_DIRECT_RECEIVE;
2234 	g_attach(cp, pp);
2235 	if (g_access(cp, 1, 0, 0) != 0)
2236 		goto ofail;
2237 
2238 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2239 		if (!class->mdc_enable)
2240 			continue;
2241 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2242 		    pp->name, class->name);
2243 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2244 		    M_WAITOK);
2245 		obj->mdo_class = class;
2246 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2247 		if (status != G_RAID_MD_TASTE_NEW)
2248 			kobj_delete((kobj_t)obj, M_RAID);
2249 		if (status != G_RAID_MD_TASTE_FAIL)
2250 			break;
2251 	}
2252 
2253 	if (status == G_RAID_MD_TASTE_FAIL)
2254 		(void)g_access(cp, -1, 0, 0);
2255 ofail:
2256 	g_detach(cp);
2257 	g_destroy_consumer(cp);
2258 	g_destroy_geom(gp);
2259 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2260 	return (geom);
2261 }
2262 
2263 int
2264 g_raid_create_node_format(const char *format, struct gctl_req *req,
2265     struct g_geom **gp)
2266 {
2267 	struct g_raid_md_class *class;
2268 	struct g_raid_md_object *obj;
2269 	int status;
2270 
2271 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2272 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2273 		if (strcasecmp(class->name, format) == 0)
2274 			break;
2275 	}
2276 	if (class == NULL) {
2277 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2278 		return (G_RAID_MD_TASTE_FAIL);
2279 	}
2280 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2281 	    M_WAITOK);
2282 	obj->mdo_class = class;
2283 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2284 	if (status != G_RAID_MD_TASTE_NEW)
2285 		kobj_delete((kobj_t)obj, M_RAID);
2286 	return (status);
2287 }
2288 
2289 static int
2290 g_raid_destroy_geom(struct gctl_req *req __unused,
2291     struct g_class *mp __unused, struct g_geom *gp)
2292 {
2293 	struct g_raid_softc *sc;
2294 	int error;
2295 
2296 	g_topology_unlock();
2297 	sc = gp->softc;
2298 	sx_xlock(&sc->sc_lock);
2299 	g_cancel_event(sc);
2300 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2301 	g_topology_lock();
2302 	return (error);
2303 }
2304 
2305 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2306     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2307 {
2308 
2309 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2310 		return;
2311 	if (sc->sc_md)
2312 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2313 }
2314 
2315 void g_raid_fail_disk(struct g_raid_softc *sc,
2316     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2317 {
2318 
2319 	if (disk == NULL)
2320 		disk = sd->sd_disk;
2321 	if (disk == NULL) {
2322 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2323 		return;
2324 	}
2325 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2326 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2327 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2328 		return;
2329 	}
2330 	if (sc->sc_md)
2331 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2332 }
2333 
2334 static void
2335 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2336     struct g_consumer *cp, struct g_provider *pp)
2337 {
2338 	struct g_raid_softc *sc;
2339 	struct g_raid_volume *vol;
2340 	struct g_raid_subdisk *sd;
2341 	struct g_raid_disk *disk;
2342 	int i, s;
2343 
2344 	g_topology_assert();
2345 
2346 	sc = gp->softc;
2347 	if (sc == NULL)
2348 		return;
2349 	if (pp != NULL) {
2350 		vol = pp->private;
2351 		g_topology_unlock();
2352 		sx_xlock(&sc->sc_lock);
2353 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2354 		    sc->sc_md->mdo_class->name,
2355 		    g_raid_volume_level2str(vol->v_raid_level,
2356 		    vol->v_raid_level_qualifier));
2357 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2358 		    vol->v_name);
2359 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2360 		    g_raid_volume_level2str(vol->v_raid_level,
2361 		    vol->v_raid_level_qualifier));
2362 		sbuf_printf(sb,
2363 		    "%s<Transformation>%s</Transformation>\n", indent,
2364 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2365 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2366 		    vol->v_disks_count);
2367 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2368 		    vol->v_strip_size);
2369 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2370 		    g_raid_volume_state2str(vol->v_state));
2371 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2372 		    vol->v_dirty ? "Yes" : "No");
2373 		sbuf_printf(sb, "%s<Subdisks>", indent);
2374 		for (i = 0; i < vol->v_disks_count; i++) {
2375 			sd = &vol->v_subdisks[i];
2376 			if (sd->sd_disk != NULL &&
2377 			    sd->sd_disk->d_consumer != NULL) {
2378 				sbuf_printf(sb, "%s ",
2379 				    g_raid_get_diskname(sd->sd_disk));
2380 			} else {
2381 				sbuf_printf(sb, "NONE ");
2382 			}
2383 			sbuf_printf(sb, "(%s",
2384 			    g_raid_subdisk_state2str(sd->sd_state));
2385 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2386 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2387 				sbuf_printf(sb, " %d%%",
2388 				    (int)(sd->sd_rebuild_pos * 100 /
2389 				     sd->sd_size));
2390 			}
2391 			sbuf_printf(sb, ")");
2392 			if (i + 1 < vol->v_disks_count)
2393 				sbuf_printf(sb, ", ");
2394 		}
2395 		sbuf_printf(sb, "</Subdisks>\n");
2396 		sx_xunlock(&sc->sc_lock);
2397 		g_topology_lock();
2398 	} else if (cp != NULL) {
2399 		disk = cp->private;
2400 		if (disk == NULL)
2401 			return;
2402 		g_topology_unlock();
2403 		sx_xlock(&sc->sc_lock);
2404 		sbuf_printf(sb, "%s<State>%s", indent,
2405 		    g_raid_disk_state2str(disk->d_state));
2406 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2407 			sbuf_printf(sb, " (");
2408 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2409 				sbuf_printf(sb, "%s",
2410 				    g_raid_subdisk_state2str(sd->sd_state));
2411 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2412 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2413 					sbuf_printf(sb, " %d%%",
2414 					    (int)(sd->sd_rebuild_pos * 100 /
2415 					     sd->sd_size));
2416 				}
2417 				if (TAILQ_NEXT(sd, sd_next))
2418 					sbuf_printf(sb, ", ");
2419 			}
2420 			sbuf_printf(sb, ")");
2421 		}
2422 		sbuf_printf(sb, "</State>\n");
2423 		sbuf_printf(sb, "%s<Subdisks>", indent);
2424 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2425 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2426 			    sd->sd_volume->v_global_id,
2427 			    sd->sd_volume->v_name,
2428 			    sd->sd_pos, sd->sd_offset);
2429 			if (TAILQ_NEXT(sd, sd_next))
2430 				sbuf_printf(sb, ", ");
2431 		}
2432 		sbuf_printf(sb, "</Subdisks>\n");
2433 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2434 		    disk->d_read_errs);
2435 		sx_xunlock(&sc->sc_lock);
2436 		g_topology_lock();
2437 	} else {
2438 		g_topology_unlock();
2439 		sx_xlock(&sc->sc_lock);
2440 		if (sc->sc_md) {
2441 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2442 			    sc->sc_md->mdo_class->name);
2443 		}
2444 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2445 			s = 0xff;
2446 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2447 				if (vol->v_state < s)
2448 					s = vol->v_state;
2449 			}
2450 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2451 			    g_raid_volume_state2str(s));
2452 		}
2453 		sx_xunlock(&sc->sc_lock);
2454 		g_topology_lock();
2455 	}
2456 }
2457 
2458 static void
2459 g_raid_shutdown_post_sync(void *arg, int howto)
2460 {
2461 	struct g_class *mp;
2462 	struct g_geom *gp, *gp2;
2463 	struct g_raid_softc *sc;
2464 	struct g_raid_volume *vol;
2465 
2466 	mp = arg;
2467 	g_topology_lock();
2468 	g_raid_shutdown = 1;
2469 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2470 		if ((sc = gp->softc) == NULL)
2471 			continue;
2472 		g_topology_unlock();
2473 		sx_xlock(&sc->sc_lock);
2474 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2475 			g_raid_clean(vol, -1);
2476 		g_cancel_event(sc);
2477 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2478 		g_topology_lock();
2479 	}
2480 	g_topology_unlock();
2481 }
2482 
2483 static void
2484 g_raid_init(struct g_class *mp)
2485 {
2486 
2487 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2488 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2489 	if (g_raid_post_sync == NULL)
2490 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2491 	g_raid_started = 1;
2492 }
2493 
2494 static void
2495 g_raid_fini(struct g_class *mp)
2496 {
2497 
2498 	if (g_raid_post_sync != NULL)
2499 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2500 	g_raid_started = 0;
2501 }
2502 
2503 int
2504 g_raid_md_modevent(module_t mod, int type, void *arg)
2505 {
2506 	struct g_raid_md_class *class, *c, *nc;
2507 	int error;
2508 
2509 	error = 0;
2510 	class = arg;
2511 	switch (type) {
2512 	case MOD_LOAD:
2513 		c = LIST_FIRST(&g_raid_md_classes);
2514 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2515 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2516 		else {
2517 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2518 			    nc->mdc_priority < class->mdc_priority)
2519 				c = nc;
2520 			LIST_INSERT_AFTER(c, class, mdc_list);
2521 		}
2522 		if (g_raid_started)
2523 			g_retaste(&g_raid_class);
2524 		break;
2525 	case MOD_UNLOAD:
2526 		LIST_REMOVE(class, mdc_list);
2527 		break;
2528 	default:
2529 		error = EOPNOTSUPP;
2530 		break;
2531 	}
2532 
2533 	return (error);
2534 }
2535 
2536 int
2537 g_raid_tr_modevent(module_t mod, int type, void *arg)
2538 {
2539 	struct g_raid_tr_class *class, *c, *nc;
2540 	int error;
2541 
2542 	error = 0;
2543 	class = arg;
2544 	switch (type) {
2545 	case MOD_LOAD:
2546 		c = LIST_FIRST(&g_raid_tr_classes);
2547 		if (c == NULL || c->trc_priority > class->trc_priority)
2548 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2549 		else {
2550 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2551 			    nc->trc_priority < class->trc_priority)
2552 				c = nc;
2553 			LIST_INSERT_AFTER(c, class, trc_list);
2554 		}
2555 		break;
2556 	case MOD_UNLOAD:
2557 		LIST_REMOVE(class, trc_list);
2558 		break;
2559 	default:
2560 		error = EOPNOTSUPP;
2561 		break;
2562 	}
2563 
2564 	return (error);
2565 }
2566 
2567 /*
2568  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2569  * to reduce module priority, allowing submodules to register them first.
2570  */
2571 static moduledata_t g_raid_mod = {
2572 	"g_raid",
2573 	g_modevent,
2574 	&g_raid_class
2575 };
2576 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2577 MODULE_VERSION(geom_raid, 0);
2578