xref: /freebsd/sys/geom/raid/g_raid.c (revision 66df505066f51e6d8411b966765d828817f88971)
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 | CTLFLAG_MPSAFE, 0,
58     "GEOM_RAID stuff");
59 int g_raid_enable = 1;
60 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN,
61     &g_raid_enable, 0, "Enable on-disk metadata taste");
62 u_int g_raid_aggressive_spare = 0;
63 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN,
64     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
65 u_int g_raid_debug = 0;
66 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0,
67     "Debug level");
68 int g_raid_read_err_thresh = 10;
69 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN,
70     &g_raid_read_err_thresh, 0,
71     "Number of read errors equated to disk failure");
72 u_int g_raid_start_timeout = 30;
73 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN,
74     &g_raid_start_timeout, 0,
75     "Time to wait for all array components");
76 static u_int g_raid_clean_time = 5;
77 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN,
78     &g_raid_clean_time, 0, "Mark volume as clean when idling");
79 static u_int g_raid_disconnect_on_failure = 1;
80 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN,
81     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
82 static u_int g_raid_name_format = 0;
83 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN,
84     &g_raid_name_format, 0, "Providers name format.");
85 static u_int g_raid_idle_threshold = 1000000;
86 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN,
87     &g_raid_idle_threshold, 1000000,
88     "Time in microseconds to consider a volume idle.");
89 
90 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
91 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
92 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
93 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
94 } while (0)
95 
96 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
97     LIST_HEAD_INITIALIZER(g_raid_md_classes);
98 
99 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
100     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
101 
102 LIST_HEAD(, g_raid_volume) g_raid_volumes =
103     LIST_HEAD_INITIALIZER(g_raid_volumes);
104 
105 static eventhandler_tag g_raid_post_sync = NULL;
106 static int g_raid_started = 0;
107 static int g_raid_shutdown = 0;
108 
109 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
110     struct g_geom *gp);
111 static g_taste_t g_raid_taste;
112 static void g_raid_init(struct g_class *mp);
113 static void g_raid_fini(struct g_class *mp);
114 
115 struct g_class g_raid_class = {
116 	.name = G_RAID_CLASS_NAME,
117 	.version = G_VERSION,
118 	.ctlreq = g_raid_ctl,
119 	.taste = g_raid_taste,
120 	.destroy_geom = g_raid_destroy_geom,
121 	.init = g_raid_init,
122 	.fini = g_raid_fini
123 };
124 
125 static void g_raid_destroy_provider(struct g_raid_volume *vol);
126 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
127 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
128 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
129 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
130 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
131     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
132 static void g_raid_start(struct bio *bp);
133 static void g_raid_start_request(struct bio *bp);
134 static void g_raid_disk_done(struct bio *bp);
135 static void g_raid_poll(struct g_raid_softc *sc);
136 
137 static const char *
138 g_raid_node_event2str(int event)
139 {
140 
141 	switch (event) {
142 	case G_RAID_NODE_E_WAKE:
143 		return ("WAKE");
144 	case G_RAID_NODE_E_START:
145 		return ("START");
146 	default:
147 		return ("INVALID");
148 	}
149 }
150 
151 const char *
152 g_raid_disk_state2str(int state)
153 {
154 
155 	switch (state) {
156 	case G_RAID_DISK_S_NONE:
157 		return ("NONE");
158 	case G_RAID_DISK_S_OFFLINE:
159 		return ("OFFLINE");
160 	case G_RAID_DISK_S_DISABLED:
161 		return ("DISABLED");
162 	case G_RAID_DISK_S_FAILED:
163 		return ("FAILED");
164 	case G_RAID_DISK_S_STALE_FAILED:
165 		return ("STALE_FAILED");
166 	case G_RAID_DISK_S_SPARE:
167 		return ("SPARE");
168 	case G_RAID_DISK_S_STALE:
169 		return ("STALE");
170 	case G_RAID_DISK_S_ACTIVE:
171 		return ("ACTIVE");
172 	default:
173 		return ("INVALID");
174 	}
175 }
176 
177 static const char *
178 g_raid_disk_event2str(int event)
179 {
180 
181 	switch (event) {
182 	case G_RAID_DISK_E_DISCONNECTED:
183 		return ("DISCONNECTED");
184 	default:
185 		return ("INVALID");
186 	}
187 }
188 
189 const char *
190 g_raid_subdisk_state2str(int state)
191 {
192 
193 	switch (state) {
194 	case G_RAID_SUBDISK_S_NONE:
195 		return ("NONE");
196 	case G_RAID_SUBDISK_S_FAILED:
197 		return ("FAILED");
198 	case G_RAID_SUBDISK_S_NEW:
199 		return ("NEW");
200 	case G_RAID_SUBDISK_S_REBUILD:
201 		return ("REBUILD");
202 	case G_RAID_SUBDISK_S_UNINITIALIZED:
203 		return ("UNINITIALIZED");
204 	case G_RAID_SUBDISK_S_STALE:
205 		return ("STALE");
206 	case G_RAID_SUBDISK_S_RESYNC:
207 		return ("RESYNC");
208 	case G_RAID_SUBDISK_S_ACTIVE:
209 		return ("ACTIVE");
210 	default:
211 		return ("INVALID");
212 	}
213 }
214 
215 static const char *
216 g_raid_subdisk_event2str(int event)
217 {
218 
219 	switch (event) {
220 	case G_RAID_SUBDISK_E_NEW:
221 		return ("NEW");
222 	case G_RAID_SUBDISK_E_FAILED:
223 		return ("FAILED");
224 	case G_RAID_SUBDISK_E_DISCONNECTED:
225 		return ("DISCONNECTED");
226 	default:
227 		return ("INVALID");
228 	}
229 }
230 
231 const char *
232 g_raid_volume_state2str(int state)
233 {
234 
235 	switch (state) {
236 	case G_RAID_VOLUME_S_STARTING:
237 		return ("STARTING");
238 	case G_RAID_VOLUME_S_BROKEN:
239 		return ("BROKEN");
240 	case G_RAID_VOLUME_S_DEGRADED:
241 		return ("DEGRADED");
242 	case G_RAID_VOLUME_S_SUBOPTIMAL:
243 		return ("SUBOPTIMAL");
244 	case G_RAID_VOLUME_S_OPTIMAL:
245 		return ("OPTIMAL");
246 	case G_RAID_VOLUME_S_UNSUPPORTED:
247 		return ("UNSUPPORTED");
248 	case G_RAID_VOLUME_S_STOPPED:
249 		return ("STOPPED");
250 	default:
251 		return ("INVALID");
252 	}
253 }
254 
255 static const char *
256 g_raid_volume_event2str(int event)
257 {
258 
259 	switch (event) {
260 	case G_RAID_VOLUME_E_UP:
261 		return ("UP");
262 	case G_RAID_VOLUME_E_DOWN:
263 		return ("DOWN");
264 	case G_RAID_VOLUME_E_START:
265 		return ("START");
266 	case G_RAID_VOLUME_E_STARTMD:
267 		return ("STARTMD");
268 	default:
269 		return ("INVALID");
270 	}
271 }
272 
273 const char *
274 g_raid_volume_level2str(int level, int qual)
275 {
276 
277 	switch (level) {
278 	case G_RAID_VOLUME_RL_RAID0:
279 		return ("RAID0");
280 	case G_RAID_VOLUME_RL_RAID1:
281 		return ("RAID1");
282 	case G_RAID_VOLUME_RL_RAID3:
283 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
284 			return ("RAID3-P0");
285 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
286 			return ("RAID3-PN");
287 		return ("RAID3");
288 	case G_RAID_VOLUME_RL_RAID4:
289 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
290 			return ("RAID4-P0");
291 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
292 			return ("RAID4-PN");
293 		return ("RAID4");
294 	case G_RAID_VOLUME_RL_RAID5:
295 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
296 			return ("RAID5-RA");
297 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
298 			return ("RAID5-RS");
299 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
300 			return ("RAID5-LA");
301 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
302 			return ("RAID5-LS");
303 		return ("RAID5");
304 	case G_RAID_VOLUME_RL_RAID6:
305 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
306 			return ("RAID6-RA");
307 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
308 			return ("RAID6-RS");
309 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
310 			return ("RAID6-LA");
311 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
312 			return ("RAID6-LS");
313 		return ("RAID6");
314 	case G_RAID_VOLUME_RL_RAIDMDF:
315 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
316 			return ("RAIDMDF-RA");
317 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
318 			return ("RAIDMDF-RS");
319 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
320 			return ("RAIDMDF-LA");
321 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
322 			return ("RAIDMDF-LS");
323 		return ("RAIDMDF");
324 	case G_RAID_VOLUME_RL_RAID1E:
325 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
326 			return ("RAID1E-A");
327 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
328 			return ("RAID1E-O");
329 		return ("RAID1E");
330 	case G_RAID_VOLUME_RL_SINGLE:
331 		return ("SINGLE");
332 	case G_RAID_VOLUME_RL_CONCAT:
333 		return ("CONCAT");
334 	case G_RAID_VOLUME_RL_RAID5E:
335 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
336 			return ("RAID5E-RA");
337 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
338 			return ("RAID5E-RS");
339 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
340 			return ("RAID5E-LA");
341 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
342 			return ("RAID5E-LS");
343 		return ("RAID5E");
344 	case G_RAID_VOLUME_RL_RAID5EE:
345 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
346 			return ("RAID5EE-RA");
347 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
348 			return ("RAID5EE-RS");
349 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
350 			return ("RAID5EE-LA");
351 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
352 			return ("RAID5EE-LS");
353 		return ("RAID5EE");
354 	case G_RAID_VOLUME_RL_RAID5R:
355 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
356 			return ("RAID5R-RA");
357 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
358 			return ("RAID5R-RS");
359 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
360 			return ("RAID5R-LA");
361 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
362 			return ("RAID5R-LS");
363 		return ("RAID5E");
364 	default:
365 		return ("UNKNOWN");
366 	}
367 }
368 
369 int
370 g_raid_volume_str2level(const char *str, int *level, int *qual)
371 {
372 
373 	*level = G_RAID_VOLUME_RL_UNKNOWN;
374 	*qual = G_RAID_VOLUME_RLQ_NONE;
375 	if (strcasecmp(str, "RAID0") == 0)
376 		*level = G_RAID_VOLUME_RL_RAID0;
377 	else if (strcasecmp(str, "RAID1") == 0)
378 		*level = G_RAID_VOLUME_RL_RAID1;
379 	else if (strcasecmp(str, "RAID3-P0") == 0) {
380 		*level = G_RAID_VOLUME_RL_RAID3;
381 		*qual = G_RAID_VOLUME_RLQ_R3P0;
382 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
383 		   strcasecmp(str, "RAID3") == 0) {
384 		*level = G_RAID_VOLUME_RL_RAID3;
385 		*qual = G_RAID_VOLUME_RLQ_R3PN;
386 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
387 		*level = G_RAID_VOLUME_RL_RAID4;
388 		*qual = G_RAID_VOLUME_RLQ_R4P0;
389 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
390 		   strcasecmp(str, "RAID4") == 0) {
391 		*level = G_RAID_VOLUME_RL_RAID4;
392 		*qual = G_RAID_VOLUME_RLQ_R4PN;
393 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
394 		*level = G_RAID_VOLUME_RL_RAID5;
395 		*qual = G_RAID_VOLUME_RLQ_R5RA;
396 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
397 		*level = G_RAID_VOLUME_RL_RAID5;
398 		*qual = G_RAID_VOLUME_RLQ_R5RS;
399 	} else if (strcasecmp(str, "RAID5") == 0 ||
400 		   strcasecmp(str, "RAID5-LA") == 0) {
401 		*level = G_RAID_VOLUME_RL_RAID5;
402 		*qual = G_RAID_VOLUME_RLQ_R5LA;
403 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
404 		*level = G_RAID_VOLUME_RL_RAID5;
405 		*qual = G_RAID_VOLUME_RLQ_R5LS;
406 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
407 		*level = G_RAID_VOLUME_RL_RAID6;
408 		*qual = G_RAID_VOLUME_RLQ_R6RA;
409 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
410 		*level = G_RAID_VOLUME_RL_RAID6;
411 		*qual = G_RAID_VOLUME_RLQ_R6RS;
412 	} else if (strcasecmp(str, "RAID6") == 0 ||
413 		   strcasecmp(str, "RAID6-LA") == 0) {
414 		*level = G_RAID_VOLUME_RL_RAID6;
415 		*qual = G_RAID_VOLUME_RLQ_R6LA;
416 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
417 		*level = G_RAID_VOLUME_RL_RAID6;
418 		*qual = G_RAID_VOLUME_RLQ_R6LS;
419 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
420 		*level = G_RAID_VOLUME_RL_RAIDMDF;
421 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
422 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
423 		*level = G_RAID_VOLUME_RL_RAIDMDF;
424 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
425 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
426 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
427 		*level = G_RAID_VOLUME_RL_RAIDMDF;
428 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
429 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
430 		*level = G_RAID_VOLUME_RL_RAIDMDF;
431 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
432 	} else if (strcasecmp(str, "RAID10") == 0 ||
433 		   strcasecmp(str, "RAID1E") == 0 ||
434 		   strcasecmp(str, "RAID1E-A") == 0) {
435 		*level = G_RAID_VOLUME_RL_RAID1E;
436 		*qual = G_RAID_VOLUME_RLQ_R1EA;
437 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
438 		*level = G_RAID_VOLUME_RL_RAID1E;
439 		*qual = G_RAID_VOLUME_RLQ_R1EO;
440 	} else if (strcasecmp(str, "SINGLE") == 0)
441 		*level = G_RAID_VOLUME_RL_SINGLE;
442 	else if (strcasecmp(str, "CONCAT") == 0)
443 		*level = G_RAID_VOLUME_RL_CONCAT;
444 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
445 		*level = G_RAID_VOLUME_RL_RAID5E;
446 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
447 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
448 		*level = G_RAID_VOLUME_RL_RAID5E;
449 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
450 	} else if (strcasecmp(str, "RAID5E") == 0 ||
451 		   strcasecmp(str, "RAID5E-LA") == 0) {
452 		*level = G_RAID_VOLUME_RL_RAID5E;
453 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
454 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
455 		*level = G_RAID_VOLUME_RL_RAID5E;
456 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
457 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
458 		*level = G_RAID_VOLUME_RL_RAID5EE;
459 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
460 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
461 		*level = G_RAID_VOLUME_RL_RAID5EE;
462 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
463 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
464 		   strcasecmp(str, "RAID5EE-LA") == 0) {
465 		*level = G_RAID_VOLUME_RL_RAID5EE;
466 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
467 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
468 		*level = G_RAID_VOLUME_RL_RAID5EE;
469 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
470 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
471 		*level = G_RAID_VOLUME_RL_RAID5R;
472 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
473 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
474 		*level = G_RAID_VOLUME_RL_RAID5R;
475 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
476 	} else if (strcasecmp(str, "RAID5R") == 0 ||
477 		   strcasecmp(str, "RAID5R-LA") == 0) {
478 		*level = G_RAID_VOLUME_RL_RAID5R;
479 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
480 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
481 		*level = G_RAID_VOLUME_RL_RAID5R;
482 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
483 	} else
484 		return (-1);
485 	return (0);
486 }
487 
488 const char *
489 g_raid_get_diskname(struct g_raid_disk *disk)
490 {
491 
492 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
493 		return ("[unknown]");
494 	return (disk->d_consumer->provider->name);
495 }
496 
497 void
498 g_raid_get_disk_info(struct g_raid_disk *disk)
499 {
500 	struct g_consumer *cp = disk->d_consumer;
501 	int error, len;
502 
503 	/* Read kernel dumping information. */
504 	disk->d_kd.offset = 0;
505 	disk->d_kd.length = OFF_MAX;
506 	len = sizeof(disk->d_kd);
507 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
508 	if (error)
509 		disk->d_kd.di.dumper = NULL;
510 	if (disk->d_kd.di.dumper == NULL)
511 		G_RAID_DEBUG1(2, disk->d_softc,
512 		    "Dumping not supported by %s: %d.",
513 		    cp->provider->name, error);
514 
515 	/* Read BIO_DELETE support. */
516 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
517 	if (error)
518 		disk->d_candelete = 0;
519 	if (!disk->d_candelete)
520 		G_RAID_DEBUG1(2, disk->d_softc,
521 		    "BIO_DELETE not supported by %s: %d.",
522 		    cp->provider->name, error);
523 }
524 
525 void
526 g_raid_report_disk_state(struct g_raid_disk *disk)
527 {
528 	struct g_raid_subdisk *sd;
529 	int len, state;
530 	uint32_t s;
531 
532 	if (disk->d_consumer == NULL)
533 		return;
534 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
535 		s = G_STATE_ACTIVE; /* XXX */
536 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
537 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
538 		s = G_STATE_FAILED;
539 	} else {
540 		state = G_RAID_SUBDISK_S_ACTIVE;
541 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
542 			if (sd->sd_state < state)
543 				state = sd->sd_state;
544 		}
545 		if (state == G_RAID_SUBDISK_S_FAILED)
546 			s = G_STATE_FAILED;
547 		else if (state == G_RAID_SUBDISK_S_NEW ||
548 		    state == G_RAID_SUBDISK_S_REBUILD)
549 			s = G_STATE_REBUILD;
550 		else if (state == G_RAID_SUBDISK_S_STALE ||
551 		    state == G_RAID_SUBDISK_S_RESYNC)
552 			s = G_STATE_RESYNC;
553 		else
554 			s = G_STATE_ACTIVE;
555 	}
556 	len = sizeof(s);
557 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
558 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
559 	    g_raid_get_diskname(disk), s);
560 }
561 
562 void
563 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
564 {
565 
566 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
567 	    g_raid_get_diskname(disk),
568 	    g_raid_disk_state2str(disk->d_state),
569 	    g_raid_disk_state2str(state));
570 	disk->d_state = state;
571 	g_raid_report_disk_state(disk);
572 }
573 
574 void
575 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
576 {
577 
578 	G_RAID_DEBUG1(0, sd->sd_softc,
579 	    "Subdisk %s:%d-%s state changed from %s to %s.",
580 	    sd->sd_volume->v_name, sd->sd_pos,
581 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
582 	    g_raid_subdisk_state2str(sd->sd_state),
583 	    g_raid_subdisk_state2str(state));
584 	sd->sd_state = state;
585 	if (sd->sd_disk)
586 		g_raid_report_disk_state(sd->sd_disk);
587 }
588 
589 void
590 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
591 {
592 
593 	G_RAID_DEBUG1(0, vol->v_softc,
594 	    "Volume %s state changed from %s to %s.",
595 	    vol->v_name,
596 	    g_raid_volume_state2str(vol->v_state),
597 	    g_raid_volume_state2str(state));
598 	vol->v_state = state;
599 }
600 
601 /*
602  * --- Events handling functions ---
603  * Events in geom_raid are used to maintain subdisks and volumes status
604  * from one thread to simplify locking.
605  */
606 static void
607 g_raid_event_free(struct g_raid_event *ep)
608 {
609 
610 	free(ep, M_RAID);
611 }
612 
613 int
614 g_raid_event_send(void *arg, int event, int flags)
615 {
616 	struct g_raid_softc *sc;
617 	struct g_raid_event *ep;
618 	int error;
619 
620 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
621 		sc = ((struct g_raid_volume *)arg)->v_softc;
622 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
623 		sc = ((struct g_raid_disk *)arg)->d_softc;
624 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
625 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
626 	} else {
627 		sc = arg;
628 	}
629 	ep = malloc(sizeof(*ep), M_RAID,
630 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
631 	if (ep == NULL)
632 		return (ENOMEM);
633 	ep->e_tgt = arg;
634 	ep->e_event = event;
635 	ep->e_flags = flags;
636 	ep->e_error = 0;
637 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
638 	mtx_lock(&sc->sc_queue_mtx);
639 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
640 	mtx_unlock(&sc->sc_queue_mtx);
641 	wakeup(sc);
642 
643 	if ((flags & G_RAID_EVENT_WAIT) == 0)
644 		return (0);
645 
646 	sx_assert(&sc->sc_lock, SX_XLOCKED);
647 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
648 	sx_xunlock(&sc->sc_lock);
649 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
650 		mtx_lock(&sc->sc_queue_mtx);
651 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
652 		    hz * 5);
653 	}
654 	error = ep->e_error;
655 	g_raid_event_free(ep);
656 	sx_xlock(&sc->sc_lock);
657 	return (error);
658 }
659 
660 static void
661 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
662 {
663 	struct g_raid_event *ep, *tmpep;
664 
665 	sx_assert(&sc->sc_lock, SX_XLOCKED);
666 
667 	mtx_lock(&sc->sc_queue_mtx);
668 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
669 		if (ep->e_tgt != tgt)
670 			continue;
671 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
672 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
673 			g_raid_event_free(ep);
674 		else {
675 			ep->e_error = ECANCELED;
676 			wakeup(ep);
677 		}
678 	}
679 	mtx_unlock(&sc->sc_queue_mtx);
680 }
681 
682 static int
683 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
684 {
685 	struct g_raid_event *ep;
686 	int	res = 0;
687 
688 	sx_assert(&sc->sc_lock, SX_XLOCKED);
689 
690 	mtx_lock(&sc->sc_queue_mtx);
691 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
692 		if (ep->e_tgt != tgt)
693 			continue;
694 		res = 1;
695 		break;
696 	}
697 	mtx_unlock(&sc->sc_queue_mtx);
698 	return (res);
699 }
700 
701 /*
702  * Return the number of disks in given state.
703  * If state is equal to -1, count all connected disks.
704  */
705 u_int
706 g_raid_ndisks(struct g_raid_softc *sc, int state)
707 {
708 	struct g_raid_disk *disk;
709 	u_int n;
710 
711 	sx_assert(&sc->sc_lock, SX_LOCKED);
712 
713 	n = 0;
714 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
715 		if (disk->d_state == state || state == -1)
716 			n++;
717 	}
718 	return (n);
719 }
720 
721 /*
722  * Return the number of subdisks in given state.
723  * If state is equal to -1, count all connected disks.
724  */
725 u_int
726 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
727 {
728 	struct g_raid_subdisk *subdisk;
729 	struct g_raid_softc *sc __diagused;
730 	u_int i, n ;
731 
732 	sc = vol->v_softc;
733 	sx_assert(&sc->sc_lock, SX_LOCKED);
734 
735 	n = 0;
736 	for (i = 0; i < vol->v_disks_count; i++) {
737 		subdisk = &vol->v_subdisks[i];
738 		if ((state == -1 &&
739 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
740 		    subdisk->sd_state == state)
741 			n++;
742 	}
743 	return (n);
744 }
745 
746 /*
747  * Return the first subdisk in given state.
748  * If state is equal to -1, then the first connected disks.
749  */
750 struct g_raid_subdisk *
751 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
752 {
753 	struct g_raid_subdisk *sd;
754 	struct g_raid_softc *sc __diagused;
755 	u_int i;
756 
757 	sc = vol->v_softc;
758 	sx_assert(&sc->sc_lock, SX_LOCKED);
759 
760 	for (i = 0; i < vol->v_disks_count; i++) {
761 		sd = &vol->v_subdisks[i];
762 		if ((state == -1 &&
763 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
764 		    sd->sd_state == state)
765 			return (sd);
766 	}
767 	return (NULL);
768 }
769 
770 struct g_consumer *
771 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
772 {
773 	struct g_consumer *cp;
774 	struct g_provider *pp;
775 
776 	g_topology_assert();
777 
778 	if (strncmp(name, _PATH_DEV, 5) == 0)
779 		name += 5;
780 	pp = g_provider_by_name(name);
781 	if (pp == NULL)
782 		return (NULL);
783 	cp = g_new_consumer(sc->sc_geom);
784 	cp->flags |= G_CF_DIRECT_RECEIVE;
785 	if (g_attach(cp, pp) != 0) {
786 		g_destroy_consumer(cp);
787 		return (NULL);
788 	}
789 	if (g_access(cp, 1, 1, 1) != 0) {
790 		g_detach(cp);
791 		g_destroy_consumer(cp);
792 		return (NULL);
793 	}
794 	return (cp);
795 }
796 
797 static u_int
798 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
799 {
800 	struct bio *bp;
801 	u_int nreqs = 0;
802 
803 	mtx_lock(&sc->sc_queue_mtx);
804 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
805 		if (bp->bio_from == cp)
806 			nreqs++;
807 	}
808 	mtx_unlock(&sc->sc_queue_mtx);
809 	return (nreqs);
810 }
811 
812 u_int
813 g_raid_nopens(struct g_raid_softc *sc)
814 {
815 	struct g_raid_volume *vol;
816 	u_int opens;
817 
818 	opens = 0;
819 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
820 		if (vol->v_provider_open != 0)
821 			opens++;
822 	}
823 	return (opens);
824 }
825 
826 static int
827 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
828 {
829 
830 	if (cp->index > 0) {
831 		G_RAID_DEBUG1(2, sc,
832 		    "I/O requests for %s exist, can't destroy it now.",
833 		    cp->provider->name);
834 		return (1);
835 	}
836 	if (g_raid_nrequests(sc, cp) > 0) {
837 		G_RAID_DEBUG1(2, sc,
838 		    "I/O requests for %s in queue, can't destroy it now.",
839 		    cp->provider->name);
840 		return (1);
841 	}
842 	return (0);
843 }
844 
845 static void
846 g_raid_destroy_consumer(void *arg, int flags __unused)
847 {
848 	struct g_consumer *cp;
849 
850 	g_topology_assert();
851 
852 	cp = arg;
853 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
854 	g_detach(cp);
855 	g_destroy_consumer(cp);
856 }
857 
858 void
859 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
860 {
861 	struct g_provider *pp;
862 	int retaste_wait;
863 
864 	g_topology_assert_not();
865 
866 	g_topology_lock();
867 	cp->private = NULL;
868 	if (g_raid_consumer_is_busy(sc, cp))
869 		goto out;
870 	pp = cp->provider;
871 	retaste_wait = 0;
872 	if (cp->acw == 1) {
873 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
874 			retaste_wait = 1;
875 	}
876 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
877 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
878 	if (retaste_wait) {
879 		/*
880 		 * After retaste event was send (inside g_access()), we can send
881 		 * event to detach and destroy consumer.
882 		 * A class, which has consumer to the given provider connected
883 		 * will not receive retaste event for the provider.
884 		 * This is the way how I ignore retaste events when I close
885 		 * consumers opened for write: I detach and destroy consumer
886 		 * after retaste event is sent.
887 		 */
888 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
889 		goto out;
890 	}
891 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
892 	g_detach(cp);
893 	g_destroy_consumer(cp);
894 out:
895 	g_topology_unlock();
896 }
897 
898 static void
899 g_raid_orphan(struct g_consumer *cp)
900 {
901 	struct g_raid_disk *disk;
902 
903 	g_topology_assert();
904 
905 	disk = cp->private;
906 	if (disk == NULL)
907 		return;
908 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
909 	    G_RAID_EVENT_DISK);
910 }
911 
912 static void
913 g_raid_clean(struct g_raid_volume *vol, int acw)
914 {
915 	struct g_raid_softc *sc;
916 	int timeout;
917 
918 	sc = vol->v_softc;
919 	g_topology_assert_not();
920 	sx_assert(&sc->sc_lock, SX_XLOCKED);
921 
922 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
923 //		return;
924 	if (!vol->v_dirty)
925 		return;
926 	if (vol->v_writes > 0)
927 		return;
928 	if (acw > 0 || (acw == -1 &&
929 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
930 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
931 		if (!g_raid_shutdown && timeout > 0)
932 			return;
933 	}
934 	vol->v_dirty = 0;
935 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
936 	    vol->v_name);
937 	g_raid_write_metadata(sc, vol, NULL, NULL);
938 }
939 
940 static void
941 g_raid_dirty(struct g_raid_volume *vol)
942 {
943 	struct g_raid_softc *sc;
944 
945 	sc = vol->v_softc;
946 	g_topology_assert_not();
947 	sx_assert(&sc->sc_lock, SX_XLOCKED);
948 
949 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
950 //		return;
951 	vol->v_dirty = 1;
952 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
953 	    vol->v_name);
954 	g_raid_write_metadata(sc, vol, NULL, NULL);
955 }
956 
957 void
958 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
959 {
960 	struct g_raid_volume *vol;
961 	struct g_raid_subdisk *sd;
962 	struct bio_queue_head queue;
963 	struct bio *cbp;
964 	int i;
965 
966 	vol = tr->tro_volume;
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 i, val;
1081 
1082 	pp = bp->bio_to;
1083 	vol = pp->private;
1084 	for (i = 0; i < vol->v_disks_count; i++) {
1085 		sd = &vol->v_subdisks[i];
1086 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1087 			continue;
1088 		if (sd->sd_disk->d_candelete)
1089 			break;
1090 	}
1091 	val = i < vol->v_disks_count;
1092 	g_handleattr(bp, "GEOM::candelete", &val, sizeof(val));
1093 }
1094 
1095 static void
1096 g_raid_start(struct bio *bp)
1097 {
1098 	struct g_raid_softc *sc;
1099 
1100 	sc = bp->bio_to->geom->softc;
1101 	/*
1102 	 * If sc == NULL or there are no valid disks, provider's error
1103 	 * should be set and g_raid_start() should not be called at all.
1104 	 */
1105 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1106 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1107 //	    bp->bio_to->error, bp->bio_to->name));
1108 	G_RAID_LOGREQ(3, bp, "Request received.");
1109 
1110 	switch (bp->bio_cmd) {
1111 	case BIO_READ:
1112 	case BIO_WRITE:
1113 	case BIO_DELETE:
1114 	case BIO_FLUSH:
1115 	case BIO_SPEEDUP:
1116 		break;
1117 	case BIO_GETATTR:
1118 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1119 			g_raid_candelete(sc, bp);
1120 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1121 			g_raid_kerneldump(sc, bp);
1122 		else
1123 			g_io_deliver(bp, EOPNOTSUPP);
1124 		return;
1125 	default:
1126 		g_io_deliver(bp, EOPNOTSUPP);
1127 		return;
1128 	}
1129 	mtx_lock(&sc->sc_queue_mtx);
1130 	bioq_insert_tail(&sc->sc_queue, bp);
1131 	mtx_unlock(&sc->sc_queue_mtx);
1132 	if (!dumping) {
1133 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1134 		wakeup(sc);
1135 	}
1136 }
1137 
1138 static int
1139 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1140 {
1141 	/*
1142 	 * 5 cases:
1143 	 * (1) bp entirely below NO
1144 	 * (2) bp entirely above NO
1145 	 * (3) bp start below, but end in range YES
1146 	 * (4) bp entirely within YES
1147 	 * (5) bp starts within, ends above YES
1148 	 *
1149 	 * lock range 10-19 (offset 10 length 10)
1150 	 * (1) 1-5: first if kicks it out
1151 	 * (2) 30-35: second if kicks it out
1152 	 * (3) 5-15: passes both ifs
1153 	 * (4) 12-14: passes both ifs
1154 	 * (5) 19-20: passes both
1155 	 */
1156 	off_t lend = lstart + len - 1;
1157 	off_t bstart = bp->bio_offset;
1158 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1159 
1160 	if (bend < lstart)
1161 		return (0);
1162 	if (lend < bstart)
1163 		return (0);
1164 	return (1);
1165 }
1166 
1167 static int
1168 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1169 {
1170 	struct g_raid_lock *lp;
1171 
1172 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1173 
1174 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1175 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1176 			return (1);
1177 	}
1178 	return (0);
1179 }
1180 
1181 static void
1182 g_raid_start_request(struct bio *bp)
1183 {
1184 	struct g_raid_softc *sc __diagused;
1185 	struct g_raid_volume *vol;
1186 
1187 	sc = bp->bio_to->geom->softc;
1188 	sx_assert(&sc->sc_lock, SX_LOCKED);
1189 	vol = bp->bio_to->private;
1190 
1191 	/*
1192 	 * Check to see if this item is in a locked range.  If so,
1193 	 * queue it to our locked queue and return.  We'll requeue
1194 	 * it when the range is unlocked.  Internal I/O for the
1195 	 * rebuild/rescan/recovery process is excluded from this
1196 	 * check so we can actually do the recovery.
1197 	 */
1198 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1199 	    g_raid_is_in_locked_range(vol, bp)) {
1200 		G_RAID_LOGREQ(3, bp, "Defer request.");
1201 		bioq_insert_tail(&vol->v_locked, bp);
1202 		return;
1203 	}
1204 
1205 	/*
1206 	 * If we're actually going to do the write/delete, then
1207 	 * update the idle stats for the volume.
1208 	 */
1209 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1210 		if (!vol->v_dirty)
1211 			g_raid_dirty(vol);
1212 		vol->v_writes++;
1213 	}
1214 
1215 	/*
1216 	 * Put request onto inflight queue, so we can check if new
1217 	 * synchronization requests don't collide with it.  Then tell
1218 	 * the transformation layer to start the I/O.
1219 	 */
1220 	bioq_insert_tail(&vol->v_inflight, bp);
1221 	G_RAID_LOGREQ(4, bp, "Request started");
1222 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1223 }
1224 
1225 static void
1226 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1227 {
1228 	off_t off, len;
1229 	struct bio *nbp;
1230 	struct g_raid_lock *lp;
1231 
1232 	vol->v_pending_lock = 0;
1233 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1234 		if (lp->l_pending) {
1235 			off = lp->l_offset;
1236 			len = lp->l_length;
1237 			lp->l_pending = 0;
1238 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1239 				if (g_raid_bio_overlaps(nbp, off, len))
1240 					lp->l_pending++;
1241 			}
1242 			if (lp->l_pending) {
1243 				vol->v_pending_lock = 1;
1244 				G_RAID_DEBUG1(4, vol->v_softc,
1245 				    "Deferred lock(%jd, %jd) has %d pending",
1246 				    (intmax_t)off, (intmax_t)(off + len),
1247 				    lp->l_pending);
1248 				continue;
1249 			}
1250 			G_RAID_DEBUG1(4, vol->v_softc,
1251 			    "Deferred lock of %jd to %jd completed",
1252 			    (intmax_t)off, (intmax_t)(off + len));
1253 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1254 		}
1255 	}
1256 }
1257 
1258 void
1259 g_raid_iodone(struct bio *bp, int error)
1260 {
1261 	struct g_raid_softc *sc __diagused;
1262 	struct g_raid_volume *vol;
1263 
1264 	sc = bp->bio_to->geom->softc;
1265 	sx_assert(&sc->sc_lock, SX_LOCKED);
1266 	vol = bp->bio_to->private;
1267 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1268 
1269 	/* Update stats if we done write/delete. */
1270 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1271 		vol->v_writes--;
1272 		vol->v_last_write = time_uptime;
1273 	}
1274 
1275 	bioq_remove(&vol->v_inflight, bp);
1276 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1277 		g_raid_finish_with_locked_ranges(vol, bp);
1278 	getmicrouptime(&vol->v_last_done);
1279 	g_io_deliver(bp, error);
1280 }
1281 
1282 int
1283 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1284     struct bio *ignore, void *argp)
1285 {
1286 	struct g_raid_softc *sc;
1287 	struct g_raid_lock *lp;
1288 	struct bio *bp;
1289 
1290 	sc = vol->v_softc;
1291 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1292 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1293 	lp->l_offset = off;
1294 	lp->l_length = len;
1295 	lp->l_callback_arg = argp;
1296 
1297 	lp->l_pending = 0;
1298 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1299 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1300 			lp->l_pending++;
1301 	}
1302 
1303 	/*
1304 	 * If there are any writes that are pending, we return EBUSY.  All
1305 	 * callers will have to wait until all pending writes clear.
1306 	 */
1307 	if (lp->l_pending > 0) {
1308 		vol->v_pending_lock = 1;
1309 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1310 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1311 		return (EBUSY);
1312 	}
1313 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1314 	    (intmax_t)off, (intmax_t)(off+len));
1315 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1316 	return (0);
1317 }
1318 
1319 int
1320 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1321 {
1322 	struct g_raid_lock *lp;
1323 	struct g_raid_softc *sc;
1324 	struct bio *bp;
1325 
1326 	sc = vol->v_softc;
1327 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1328 		if (lp->l_offset == off && lp->l_length == len) {
1329 			LIST_REMOVE(lp, l_next);
1330 			/* XXX
1331 			 * Right now we just put them all back on the queue
1332 			 * and hope for the best.  We hope this because any
1333 			 * locked ranges will go right back on this list
1334 			 * when the worker thread runs.
1335 			 * XXX
1336 			 */
1337 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1338 			    (intmax_t)lp->l_offset,
1339 			    (intmax_t)(lp->l_offset+lp->l_length));
1340 			mtx_lock(&sc->sc_queue_mtx);
1341 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1342 				bioq_insert_tail(&sc->sc_queue, bp);
1343 			mtx_unlock(&sc->sc_queue_mtx);
1344 			free(lp, M_RAID);
1345 			return (0);
1346 		}
1347 	}
1348 	return (EINVAL);
1349 }
1350 
1351 void
1352 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1353 {
1354 	struct g_consumer *cp;
1355 	struct g_raid_disk *disk, *tdisk;
1356 
1357 	bp->bio_caller1 = sd;
1358 
1359 	/*
1360 	 * Make sure that the disk is present. Generally it is a task of
1361 	 * transformation layers to not send requests to absent disks, but
1362 	 * it is better to be safe and report situation then sorry.
1363 	 */
1364 	if (sd->sd_disk == NULL) {
1365 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1366 nodisk:
1367 		bp->bio_from = NULL;
1368 		bp->bio_to = NULL;
1369 		bp->bio_error = ENXIO;
1370 		g_raid_disk_done(bp);
1371 		return;
1372 	}
1373 	disk = sd->sd_disk;
1374 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1375 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1376 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1377 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1378 		goto nodisk;
1379 	}
1380 
1381 	cp = disk->d_consumer;
1382 	bp->bio_from = cp;
1383 	bp->bio_to = cp->provider;
1384 	cp->index++;
1385 
1386 	/* Update average disks load. */
1387 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1388 		if (tdisk->d_consumer == NULL)
1389 			tdisk->d_load = 0;
1390 		else
1391 			tdisk->d_load = (tdisk->d_consumer->index *
1392 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1393 	}
1394 
1395 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1396 	if (dumping) {
1397 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1398 		if (bp->bio_cmd == BIO_WRITE) {
1399 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1400 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1401 		} else
1402 			bp->bio_error = EOPNOTSUPP;
1403 		g_raid_disk_done(bp);
1404 	} else {
1405 		bp->bio_done = g_raid_disk_done;
1406 		bp->bio_offset += sd->sd_offset;
1407 		G_RAID_LOGREQ(3, bp, "Sending request.");
1408 		g_io_request(bp, cp);
1409 	}
1410 }
1411 
1412 int
1413 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1414     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1415 {
1416 
1417 	if (sd->sd_disk == NULL)
1418 		return (ENXIO);
1419 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1420 		return (EOPNOTSUPP);
1421 	return (dump_write(&sd->sd_disk->d_kd.di,
1422 	    virtual, physical,
1423 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1424 	    length));
1425 }
1426 
1427 static void
1428 g_raid_disk_done(struct bio *bp)
1429 {
1430 	struct g_raid_softc *sc;
1431 	struct g_raid_subdisk *sd;
1432 
1433 	sd = bp->bio_caller1;
1434 	sc = sd->sd_softc;
1435 	mtx_lock(&sc->sc_queue_mtx);
1436 	bioq_insert_tail(&sc->sc_queue, bp);
1437 	mtx_unlock(&sc->sc_queue_mtx);
1438 	if (!dumping)
1439 		wakeup(sc);
1440 }
1441 
1442 static void
1443 g_raid_disk_done_request(struct bio *bp)
1444 {
1445 	struct g_raid_softc *sc;
1446 	struct g_raid_disk *disk;
1447 	struct g_raid_subdisk *sd;
1448 	struct g_raid_volume *vol;
1449 
1450 	g_topology_assert_not();
1451 
1452 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1453 	sd = bp->bio_caller1;
1454 	sc = sd->sd_softc;
1455 	vol = sd->sd_volume;
1456 	if (bp->bio_from != NULL) {
1457 		bp->bio_from->index--;
1458 		disk = bp->bio_from->private;
1459 		if (disk == NULL)
1460 			g_raid_kill_consumer(sc, bp->bio_from);
1461 	}
1462 	bp->bio_offset -= sd->sd_offset;
1463 
1464 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1465 }
1466 
1467 static void
1468 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1469 {
1470 
1471 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1472 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1473 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1474 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1475 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1476 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1477 	else
1478 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1479 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1480 		KASSERT(ep->e_error == 0,
1481 		    ("Error cannot be handled."));
1482 		g_raid_event_free(ep);
1483 	} else {
1484 		ep->e_flags |= G_RAID_EVENT_DONE;
1485 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1486 		mtx_lock(&sc->sc_queue_mtx);
1487 		wakeup(ep);
1488 		mtx_unlock(&sc->sc_queue_mtx);
1489 	}
1490 }
1491 
1492 /*
1493  * Worker thread.
1494  */
1495 static void
1496 g_raid_worker(void *arg)
1497 {
1498 	struct g_raid_softc *sc;
1499 	struct g_raid_event *ep;
1500 	struct g_raid_volume *vol;
1501 	struct bio *bp;
1502 	struct timeval now, t;
1503 	int timeout, rv;
1504 
1505 	sc = arg;
1506 	thread_lock(curthread);
1507 	sched_prio(curthread, PRIBIO);
1508 	thread_unlock(curthread);
1509 
1510 	sx_xlock(&sc->sc_lock);
1511 	for (;;) {
1512 		mtx_lock(&sc->sc_queue_mtx);
1513 		/*
1514 		 * First take a look at events.
1515 		 * This is important to handle events before any I/O requests.
1516 		 */
1517 		bp = NULL;
1518 		vol = NULL;
1519 		rv = 0;
1520 		ep = TAILQ_FIRST(&sc->sc_events);
1521 		if (ep != NULL)
1522 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1523 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1524 			;
1525 		else {
1526 			getmicrouptime(&now);
1527 			t = now;
1528 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1529 				if (bioq_first(&vol->v_inflight) == NULL &&
1530 				    vol->v_tr &&
1531 				    timevalcmp(&vol->v_last_done, &t, < ))
1532 					t = vol->v_last_done;
1533 			}
1534 			timevalsub(&t, &now);
1535 			timeout = g_raid_idle_threshold +
1536 			    t.tv_sec * 1000000 + t.tv_usec;
1537 			if (timeout > 0) {
1538 				/*
1539 				 * Two steps to avoid overflows at HZ=1000
1540 				 * and idle timeouts > 2.1s.  Some rounding
1541 				 * errors can occur, but they are < 1tick,
1542 				 * which is deemed to be close enough for
1543 				 * this purpose.
1544 				 */
1545 				int micpertic = 1000000 / hz;
1546 				timeout = (timeout + micpertic - 1) / micpertic;
1547 				sx_xunlock(&sc->sc_lock);
1548 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1549 				    PRIBIO | PDROP, "-", timeout);
1550 				sx_xlock(&sc->sc_lock);
1551 				goto process;
1552 			} else
1553 				rv = EWOULDBLOCK;
1554 		}
1555 		mtx_unlock(&sc->sc_queue_mtx);
1556 process:
1557 		if (ep != NULL) {
1558 			g_raid_handle_event(sc, ep);
1559 		} else if (bp != NULL) {
1560 			if (bp->bio_to != NULL &&
1561 			    bp->bio_to->geom == sc->sc_geom)
1562 				g_raid_start_request(bp);
1563 			else
1564 				g_raid_disk_done_request(bp);
1565 		} else if (rv == EWOULDBLOCK) {
1566 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1567 				g_raid_clean(vol, -1);
1568 				if (bioq_first(&vol->v_inflight) == NULL &&
1569 				    vol->v_tr) {
1570 					t.tv_sec = g_raid_idle_threshold / 1000000;
1571 					t.tv_usec = g_raid_idle_threshold % 1000000;
1572 					timevaladd(&t, &vol->v_last_done);
1573 					getmicrouptime(&now);
1574 					if (timevalcmp(&t, &now, <= )) {
1575 						G_RAID_TR_IDLE(vol->v_tr);
1576 						vol->v_last_done = now;
1577 					}
1578 				}
1579 			}
1580 		}
1581 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1582 			g_raid_destroy_node(sc, 1);	/* May not return. */
1583 	}
1584 }
1585 
1586 static void
1587 g_raid_poll(struct g_raid_softc *sc)
1588 {
1589 	struct g_raid_event *ep;
1590 	struct bio *bp;
1591 
1592 	sx_xlock(&sc->sc_lock);
1593 	mtx_lock(&sc->sc_queue_mtx);
1594 	/*
1595 	 * First take a look at events.
1596 	 * This is important to handle events before any I/O requests.
1597 	 */
1598 	ep = TAILQ_FIRST(&sc->sc_events);
1599 	if (ep != NULL) {
1600 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1601 		mtx_unlock(&sc->sc_queue_mtx);
1602 		g_raid_handle_event(sc, ep);
1603 		goto out;
1604 	}
1605 	bp = bioq_takefirst(&sc->sc_queue);
1606 	if (bp != NULL) {
1607 		mtx_unlock(&sc->sc_queue_mtx);
1608 		if (bp->bio_from == NULL ||
1609 		    bp->bio_from->geom != sc->sc_geom)
1610 			g_raid_start_request(bp);
1611 		else
1612 			g_raid_disk_done_request(bp);
1613 	}
1614 out:
1615 	sx_xunlock(&sc->sc_lock);
1616 }
1617 
1618 static void
1619 g_raid_launch_provider(struct g_raid_volume *vol)
1620 {
1621 	struct g_raid_disk *disk;
1622 	struct g_raid_subdisk *sd;
1623 	struct g_raid_softc *sc;
1624 	struct g_provider *pp;
1625 	char name[G_RAID_MAX_VOLUMENAME];
1626 	off_t off;
1627 	int i;
1628 
1629 	sc = vol->v_softc;
1630 	sx_assert(&sc->sc_lock, SX_LOCKED);
1631 
1632 	g_topology_lock();
1633 	/* Try to name provider with volume name. */
1634 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1635 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1636 	    g_provider_by_name(name) != NULL) {
1637 		/* Otherwise use sequential volume number. */
1638 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1639 	}
1640 
1641 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1642 	pp->flags |= G_PF_DIRECT_RECEIVE;
1643 	if (vol->v_tr->tro_class->trc_accept_unmapped) {
1644 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1645 		for (i = 0; i < vol->v_disks_count; i++) {
1646 			sd = &vol->v_subdisks[i];
1647 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1648 				continue;
1649 			if ((sd->sd_disk->d_consumer->provider->flags &
1650 			    G_PF_ACCEPT_UNMAPPED) == 0)
1651 				pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
1652 		}
1653 	}
1654 	pp->private = vol;
1655 	pp->mediasize = vol->v_mediasize;
1656 	pp->sectorsize = vol->v_sectorsize;
1657 	pp->stripesize = 0;
1658 	pp->stripeoffset = 0;
1659 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1660 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1661 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1662 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1663 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1664 		    disk->d_consumer != NULL &&
1665 		    disk->d_consumer->provider != NULL) {
1666 			pp->stripesize = disk->d_consumer->provider->stripesize;
1667 			off = disk->d_consumer->provider->stripeoffset;
1668 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1669 			if (off > 0)
1670 				pp->stripeoffset %= off;
1671 		}
1672 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1673 			pp->stripesize *= (vol->v_disks_count - 1);
1674 			pp->stripeoffset *= (vol->v_disks_count - 1);
1675 		}
1676 	} else
1677 		pp->stripesize = vol->v_strip_size;
1678 	vol->v_provider = pp;
1679 	g_error_provider(pp, 0);
1680 	g_topology_unlock();
1681 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1682 	    pp->name, vol->v_name);
1683 }
1684 
1685 static void
1686 g_raid_destroy_provider(struct g_raid_volume *vol)
1687 {
1688 	struct g_raid_softc *sc;
1689 	struct g_provider *pp;
1690 	struct bio *bp, *tmp;
1691 
1692 	g_topology_assert_not();
1693 	sc = vol->v_softc;
1694 	pp = vol->v_provider;
1695 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1696 
1697 	g_topology_lock();
1698 	g_error_provider(pp, ENXIO);
1699 	mtx_lock(&sc->sc_queue_mtx);
1700 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1701 		if (bp->bio_to != pp)
1702 			continue;
1703 		bioq_remove(&sc->sc_queue, bp);
1704 		g_io_deliver(bp, ENXIO);
1705 	}
1706 	mtx_unlock(&sc->sc_queue_mtx);
1707 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1708 	    pp->name, vol->v_name);
1709 	g_wither_provider(pp, ENXIO);
1710 	g_topology_unlock();
1711 	vol->v_provider = NULL;
1712 }
1713 
1714 /*
1715  * Update device state.
1716  */
1717 static int
1718 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1719 {
1720 	struct g_raid_softc *sc;
1721 
1722 	sc = vol->v_softc;
1723 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1724 
1725 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1726 	    g_raid_volume_event2str(event),
1727 	    vol->v_name);
1728 	switch (event) {
1729 	case G_RAID_VOLUME_E_DOWN:
1730 		if (vol->v_provider != NULL)
1731 			g_raid_destroy_provider(vol);
1732 		break;
1733 	case G_RAID_VOLUME_E_UP:
1734 		if (vol->v_provider == NULL)
1735 			g_raid_launch_provider(vol);
1736 		break;
1737 	case G_RAID_VOLUME_E_START:
1738 		if (vol->v_tr)
1739 			G_RAID_TR_START(vol->v_tr);
1740 		return (0);
1741 	default:
1742 		if (sc->sc_md)
1743 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1744 		return (0);
1745 	}
1746 
1747 	/* Manage root mount release. */
1748 	if (vol->v_starting) {
1749 		vol->v_starting = 0;
1750 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1751 		root_mount_rel(vol->v_rootmount);
1752 		vol->v_rootmount = NULL;
1753 	}
1754 	if (vol->v_stopping && vol->v_provider_open == 0)
1755 		g_raid_destroy_volume(vol);
1756 	return (0);
1757 }
1758 
1759 /*
1760  * Update subdisk state.
1761  */
1762 static int
1763 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1764 {
1765 	struct g_raid_softc *sc;
1766 	struct g_raid_volume *vol;
1767 
1768 	sc = sd->sd_softc;
1769 	vol = sd->sd_volume;
1770 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1771 
1772 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1773 	    g_raid_subdisk_event2str(event),
1774 	    vol->v_name, sd->sd_pos,
1775 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1776 	if (vol->v_tr)
1777 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1778 
1779 	return (0);
1780 }
1781 
1782 /*
1783  * Update disk state.
1784  */
1785 static int
1786 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1787 {
1788 	struct g_raid_softc *sc;
1789 
1790 	sc = disk->d_softc;
1791 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1792 
1793 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1794 	    g_raid_disk_event2str(event),
1795 	    g_raid_get_diskname(disk));
1796 
1797 	if (sc->sc_md)
1798 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1799 	return (0);
1800 }
1801 
1802 /*
1803  * Node event.
1804  */
1805 static int
1806 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1807 {
1808 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1809 
1810 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1811 	    g_raid_node_event2str(event));
1812 
1813 	if (event == G_RAID_NODE_E_WAKE)
1814 		return (0);
1815 	if (sc->sc_md)
1816 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1817 	return (0);
1818 }
1819 
1820 static int
1821 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1822 {
1823 	struct g_raid_volume *vol;
1824 	struct g_raid_softc *sc;
1825 	int dcw, opens, error = 0;
1826 
1827 	g_topology_assert();
1828 	sc = pp->geom->softc;
1829 	vol = pp->private;
1830 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1831 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1832 
1833 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1834 	    acr, acw, ace);
1835 	dcw = pp->acw + acw;
1836 
1837 	g_topology_unlock();
1838 	sx_xlock(&sc->sc_lock);
1839 	/* Deny new opens while dying. */
1840 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1841 		error = ENXIO;
1842 		goto out;
1843 	}
1844 	/* Deny write opens for read-only volumes. */
1845 	if (vol->v_read_only && acw > 0) {
1846 		error = EROFS;
1847 		goto out;
1848 	}
1849 	if (dcw == 0)
1850 		g_raid_clean(vol, dcw);
1851 	vol->v_provider_open += acr + acw + ace;
1852 	/* Handle delayed node destruction. */
1853 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1854 	    vol->v_provider_open == 0) {
1855 		/* Count open volumes. */
1856 		opens = g_raid_nopens(sc);
1857 		if (opens == 0) {
1858 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1859 			/* Wake up worker to make it selfdestruct. */
1860 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1861 		}
1862 	}
1863 	/* Handle open volume destruction. */
1864 	if (vol->v_stopping && vol->v_provider_open == 0)
1865 		g_raid_destroy_volume(vol);
1866 out:
1867 	sx_xunlock(&sc->sc_lock);
1868 	g_topology_lock();
1869 	return (error);
1870 }
1871 
1872 struct g_raid_softc *
1873 g_raid_create_node(struct g_class *mp,
1874     const char *name, struct g_raid_md_object *md)
1875 {
1876 	struct g_raid_softc *sc;
1877 	struct g_geom *gp;
1878 	int error;
1879 
1880 	g_topology_assert();
1881 	G_RAID_DEBUG(1, "Creating array %s.", name);
1882 
1883 	gp = g_new_geomf(mp, "%s", name);
1884 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1885 	gp->start = g_raid_start;
1886 	gp->orphan = g_raid_orphan;
1887 	gp->access = g_raid_access;
1888 	gp->dumpconf = g_raid_dumpconf;
1889 
1890 	sc->sc_md = md;
1891 	sc->sc_geom = gp;
1892 	sc->sc_flags = 0;
1893 	TAILQ_INIT(&sc->sc_volumes);
1894 	TAILQ_INIT(&sc->sc_disks);
1895 	sx_init(&sc->sc_lock, "graid:lock");
1896 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1897 	TAILQ_INIT(&sc->sc_events);
1898 	bioq_init(&sc->sc_queue);
1899 	gp->softc = sc;
1900 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1901 	    "g_raid %s", name);
1902 	if (error != 0) {
1903 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1904 		mtx_destroy(&sc->sc_queue_mtx);
1905 		sx_destroy(&sc->sc_lock);
1906 		g_destroy_geom(sc->sc_geom);
1907 		free(sc, M_RAID);
1908 		return (NULL);
1909 	}
1910 
1911 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1912 	return (sc);
1913 }
1914 
1915 struct g_raid_volume *
1916 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1917 {
1918 	struct g_raid_volume	*vol, *vol1;
1919 	int i;
1920 
1921 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1922 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1923 	vol->v_softc = sc;
1924 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1925 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1926 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1927 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1928 	vol->v_rotate_parity = 1;
1929 	bioq_init(&vol->v_inflight);
1930 	bioq_init(&vol->v_locked);
1931 	LIST_INIT(&vol->v_locks);
1932 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1933 		vol->v_subdisks[i].sd_softc = sc;
1934 		vol->v_subdisks[i].sd_volume = vol;
1935 		vol->v_subdisks[i].sd_pos = i;
1936 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1937 	}
1938 
1939 	/* Find free ID for this volume. */
1940 	g_topology_lock();
1941 	vol1 = vol;
1942 	if (id >= 0) {
1943 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1944 			if (vol1->v_global_id == id)
1945 				break;
1946 		}
1947 	}
1948 	if (vol1 != NULL) {
1949 		for (id = 0; ; id++) {
1950 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1951 				if (vol1->v_global_id == id)
1952 					break;
1953 			}
1954 			if (vol1 == NULL)
1955 				break;
1956 		}
1957 	}
1958 	vol->v_global_id = id;
1959 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1960 	g_topology_unlock();
1961 
1962 	/* Delay root mounting. */
1963 	vol->v_rootmount = root_mount_hold("GRAID");
1964 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1965 	vol->v_starting = 1;
1966 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1967 	return (vol);
1968 }
1969 
1970 struct g_raid_disk *
1971 g_raid_create_disk(struct g_raid_softc *sc)
1972 {
1973 	struct g_raid_disk	*disk;
1974 
1975 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1976 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1977 	disk->d_softc = sc;
1978 	disk->d_state = G_RAID_DISK_S_NONE;
1979 	TAILQ_INIT(&disk->d_subdisks);
1980 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1981 	return (disk);
1982 }
1983 
1984 int g_raid_start_volume(struct g_raid_volume *vol)
1985 {
1986 	struct g_raid_tr_class *class;
1987 	struct g_raid_tr_object *obj;
1988 	int status;
1989 
1990 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1991 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1992 		if (!class->trc_enable)
1993 			continue;
1994 		G_RAID_DEBUG1(2, vol->v_softc,
1995 		    "Tasting volume %s for %s transformation.",
1996 		    vol->v_name, class->name);
1997 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1998 		    M_WAITOK);
1999 		obj->tro_class = class;
2000 		obj->tro_volume = vol;
2001 		status = G_RAID_TR_TASTE(obj, vol);
2002 		if (status != G_RAID_TR_TASTE_FAIL)
2003 			break;
2004 		kobj_delete((kobj_t)obj, M_RAID);
2005 	}
2006 	if (class == NULL) {
2007 		G_RAID_DEBUG1(0, vol->v_softc,
2008 		    "No transformation module found for %s.",
2009 		    vol->v_name);
2010 		vol->v_tr = NULL;
2011 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2012 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2013 		    G_RAID_EVENT_VOLUME);
2014 		return (-1);
2015 	}
2016 	G_RAID_DEBUG1(2, vol->v_softc,
2017 	    "Transformation module %s chosen for %s.",
2018 	    class->name, vol->v_name);
2019 	vol->v_tr = obj;
2020 	return (0);
2021 }
2022 
2023 int
2024 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2025 {
2026 	struct g_raid_volume *vol, *tmpv;
2027 	struct g_raid_disk *disk, *tmpd;
2028 	int error = 0;
2029 
2030 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2031 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2032 		if (g_raid_destroy_volume(vol))
2033 			error = EBUSY;
2034 	}
2035 	if (error)
2036 		return (error);
2037 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2038 		if (g_raid_destroy_disk(disk))
2039 			error = EBUSY;
2040 	}
2041 	if (error)
2042 		return (error);
2043 	if (sc->sc_md) {
2044 		G_RAID_MD_FREE(sc->sc_md);
2045 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2046 		sc->sc_md = NULL;
2047 	}
2048 	if (sc->sc_geom != NULL) {
2049 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2050 		g_topology_lock();
2051 		sc->sc_geom->softc = NULL;
2052 		g_wither_geom(sc->sc_geom, ENXIO);
2053 		g_topology_unlock();
2054 		sc->sc_geom = NULL;
2055 	} else
2056 		G_RAID_DEBUG(1, "Array destroyed.");
2057 	if (worker) {
2058 		g_raid_event_cancel(sc, sc);
2059 		mtx_destroy(&sc->sc_queue_mtx);
2060 		sx_xunlock(&sc->sc_lock);
2061 		sx_destroy(&sc->sc_lock);
2062 		wakeup(&sc->sc_stopping);
2063 		free(sc, M_RAID);
2064 		curthread->td_pflags &= ~TDP_GEOM;
2065 		G_RAID_DEBUG(1, "Thread exiting.");
2066 		kproc_exit(0);
2067 	} else {
2068 		/* Wake up worker to make it selfdestruct. */
2069 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2070 	}
2071 	return (0);
2072 }
2073 
2074 int
2075 g_raid_destroy_volume(struct g_raid_volume *vol)
2076 {
2077 	struct g_raid_softc *sc;
2078 	struct g_raid_disk *disk;
2079 	int i;
2080 
2081 	sc = vol->v_softc;
2082 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2083 	vol->v_stopping = 1;
2084 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2085 		if (vol->v_tr) {
2086 			G_RAID_TR_STOP(vol->v_tr);
2087 			return (EBUSY);
2088 		} else
2089 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2090 	}
2091 	if (g_raid_event_check(sc, vol) != 0)
2092 		return (EBUSY);
2093 	if (vol->v_provider != NULL)
2094 		return (EBUSY);
2095 	if (vol->v_provider_open != 0)
2096 		return (EBUSY);
2097 	if (vol->v_tr) {
2098 		G_RAID_TR_FREE(vol->v_tr);
2099 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2100 		vol->v_tr = NULL;
2101 	}
2102 	if (vol->v_rootmount)
2103 		root_mount_rel(vol->v_rootmount);
2104 	g_topology_lock();
2105 	LIST_REMOVE(vol, v_global_next);
2106 	g_topology_unlock();
2107 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2108 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2109 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2110 		disk = vol->v_subdisks[i].sd_disk;
2111 		if (disk == NULL)
2112 			continue;
2113 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2114 	}
2115 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2116 	if (sc->sc_md)
2117 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2118 	g_raid_event_cancel(sc, vol);
2119 	free(vol, M_RAID);
2120 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2121 		/* Wake up worker to let it selfdestruct. */
2122 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2123 	}
2124 	return (0);
2125 }
2126 
2127 int
2128 g_raid_destroy_disk(struct g_raid_disk *disk)
2129 {
2130 	struct g_raid_softc *sc;
2131 	struct g_raid_subdisk *sd, *tmp;
2132 
2133 	sc = disk->d_softc;
2134 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2135 	if (disk->d_consumer) {
2136 		g_raid_kill_consumer(sc, disk->d_consumer);
2137 		disk->d_consumer = NULL;
2138 	}
2139 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2140 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2141 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2142 		    G_RAID_EVENT_SUBDISK);
2143 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2144 		sd->sd_disk = NULL;
2145 	}
2146 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2147 	if (sc->sc_md)
2148 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2149 	g_raid_event_cancel(sc, disk);
2150 	free(disk, M_RAID);
2151 	return (0);
2152 }
2153 
2154 int
2155 g_raid_destroy(struct g_raid_softc *sc, int how)
2156 {
2157 	int error, opens;
2158 
2159 	g_topology_assert_not();
2160 	if (sc == NULL)
2161 		return (ENXIO);
2162 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2163 
2164 	/* Count open volumes. */
2165 	opens = g_raid_nopens(sc);
2166 
2167 	/* React on some opened volumes. */
2168 	if (opens > 0) {
2169 		switch (how) {
2170 		case G_RAID_DESTROY_SOFT:
2171 			G_RAID_DEBUG1(1, sc,
2172 			    "%d volumes are still open.",
2173 			    opens);
2174 			sx_xunlock(&sc->sc_lock);
2175 			return (EBUSY);
2176 		case G_RAID_DESTROY_DELAYED:
2177 			G_RAID_DEBUG1(1, sc,
2178 			    "Array will be destroyed on last close.");
2179 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2180 			sx_xunlock(&sc->sc_lock);
2181 			return (EBUSY);
2182 		case G_RAID_DESTROY_HARD:
2183 			G_RAID_DEBUG1(1, sc,
2184 			    "%d volumes are still open.",
2185 			    opens);
2186 		}
2187 	}
2188 
2189 	/* Mark node for destruction. */
2190 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2191 	/* Wake up worker to let it selfdestruct. */
2192 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2193 	/* Sleep until node destroyed. */
2194 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2195 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2196 	return (error == EWOULDBLOCK ? EBUSY : 0);
2197 }
2198 
2199 static void
2200 g_raid_taste_orphan(struct g_consumer *cp)
2201 {
2202 
2203 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2204 	    cp->provider->name));
2205 }
2206 
2207 static struct g_geom *
2208 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2209 {
2210 	struct g_consumer *cp;
2211 	struct g_geom *gp, *geom;
2212 	struct g_raid_md_class *class;
2213 	struct g_raid_md_object *obj;
2214 	int status;
2215 
2216 	g_topology_assert();
2217 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2218 	if (!g_raid_enable)
2219 		return (NULL);
2220 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2221 
2222 	geom = NULL;
2223 	status = G_RAID_MD_TASTE_FAIL;
2224 	gp = g_new_geomf(mp, "raid:taste");
2225 	/*
2226 	 * This orphan function should be never called.
2227 	 */
2228 	gp->orphan = g_raid_taste_orphan;
2229 	cp = g_new_consumer(gp);
2230 	cp->flags |= G_CF_DIRECT_RECEIVE;
2231 	if (g_attach(cp, pp) != 0)
2232 		goto ofail2;
2233 	if (g_access(cp, 1, 0, 0) != 0)
2234 		goto ofail;
2235 
2236 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2237 		if (!class->mdc_enable)
2238 			continue;
2239 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2240 		    pp->name, class->name);
2241 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2242 		    M_WAITOK);
2243 		obj->mdo_class = class;
2244 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2245 		if (status != G_RAID_MD_TASTE_NEW)
2246 			kobj_delete((kobj_t)obj, M_RAID);
2247 		if (status != G_RAID_MD_TASTE_FAIL)
2248 			break;
2249 	}
2250 
2251 	if (status == G_RAID_MD_TASTE_FAIL)
2252 		(void)g_access(cp, -1, 0, 0);
2253 ofail:
2254 	g_detach(cp);
2255 ofail2:
2256 	g_destroy_consumer(cp);
2257 	g_destroy_geom(gp);
2258 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2259 	return (geom);
2260 }
2261 
2262 int
2263 g_raid_create_node_format(const char *format, struct gctl_req *req,
2264     struct g_geom **gp)
2265 {
2266 	struct g_raid_md_class *class;
2267 	struct g_raid_md_object *obj;
2268 	int status;
2269 
2270 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2271 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2272 		if (strcasecmp(class->name, format) == 0)
2273 			break;
2274 	}
2275 	if (class == NULL) {
2276 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2277 		return (G_RAID_MD_TASTE_FAIL);
2278 	}
2279 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2280 	    M_WAITOK);
2281 	obj->mdo_class = class;
2282 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2283 	if (status != G_RAID_MD_TASTE_NEW)
2284 		kobj_delete((kobj_t)obj, M_RAID);
2285 	return (status);
2286 }
2287 
2288 static int
2289 g_raid_destroy_geom(struct gctl_req *req __unused,
2290     struct g_class *mp __unused, struct g_geom *gp)
2291 {
2292 	struct g_raid_softc *sc;
2293 	int error;
2294 
2295 	g_topology_unlock();
2296 	sc = gp->softc;
2297 	sx_xlock(&sc->sc_lock);
2298 	g_cancel_event(sc);
2299 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2300 	g_topology_lock();
2301 	return (error);
2302 }
2303 
2304 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2305     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2306 {
2307 
2308 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2309 		return;
2310 	if (sc->sc_md)
2311 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2312 }
2313 
2314 void g_raid_fail_disk(struct g_raid_softc *sc,
2315     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2316 {
2317 
2318 	if (disk == NULL)
2319 		disk = sd->sd_disk;
2320 	if (disk == NULL) {
2321 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2322 		return;
2323 	}
2324 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2325 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2326 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2327 		return;
2328 	}
2329 	if (sc->sc_md)
2330 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2331 }
2332 
2333 static void
2334 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2335     struct g_consumer *cp, struct g_provider *pp)
2336 {
2337 	struct g_raid_softc *sc;
2338 	struct g_raid_volume *vol;
2339 	struct g_raid_subdisk *sd;
2340 	struct g_raid_disk *disk;
2341 	int i, s;
2342 
2343 	g_topology_assert();
2344 
2345 	sc = gp->softc;
2346 	if (sc == NULL)
2347 		return;
2348 	if (pp != NULL) {
2349 		vol = pp->private;
2350 		g_topology_unlock();
2351 		sx_xlock(&sc->sc_lock);
2352 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2353 		    sc->sc_md->mdo_class->name,
2354 		    g_raid_volume_level2str(vol->v_raid_level,
2355 		    vol->v_raid_level_qualifier));
2356 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2357 		    vol->v_name);
2358 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2359 		    g_raid_volume_level2str(vol->v_raid_level,
2360 		    vol->v_raid_level_qualifier));
2361 		sbuf_printf(sb,
2362 		    "%s<Transformation>%s</Transformation>\n", indent,
2363 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2364 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2365 		    vol->v_disks_count);
2366 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2367 		    vol->v_strip_size);
2368 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2369 		    g_raid_volume_state2str(vol->v_state));
2370 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2371 		    vol->v_dirty ? "Yes" : "No");
2372 		sbuf_printf(sb, "%s<Subdisks>", indent);
2373 		for (i = 0; i < vol->v_disks_count; i++) {
2374 			sd = &vol->v_subdisks[i];
2375 			if (sd->sd_disk != NULL &&
2376 			    sd->sd_disk->d_consumer != NULL) {
2377 				sbuf_printf(sb, "%s ",
2378 				    g_raid_get_diskname(sd->sd_disk));
2379 			} else {
2380 				sbuf_cat(sb, "NONE ");
2381 			}
2382 			sbuf_printf(sb, "(%s",
2383 			    g_raid_subdisk_state2str(sd->sd_state));
2384 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2385 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2386 				sbuf_printf(sb, " %d%%",
2387 				    (int)(sd->sd_rebuild_pos * 100 /
2388 				     sd->sd_size));
2389 			}
2390 			sbuf_cat(sb, ")");
2391 			if (i + 1 < vol->v_disks_count)
2392 				sbuf_cat(sb, ", ");
2393 		}
2394 		sbuf_cat(sb, "</Subdisks>\n");
2395 		sx_xunlock(&sc->sc_lock);
2396 		g_topology_lock();
2397 	} else if (cp != NULL) {
2398 		disk = cp->private;
2399 		if (disk == NULL)
2400 			return;
2401 		g_topology_unlock();
2402 		sx_xlock(&sc->sc_lock);
2403 		sbuf_printf(sb, "%s<State>%s", indent,
2404 		    g_raid_disk_state2str(disk->d_state));
2405 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2406 			sbuf_cat(sb, " (");
2407 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2408 				sbuf_printf(sb, "%s",
2409 				    g_raid_subdisk_state2str(sd->sd_state));
2410 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2411 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2412 					sbuf_printf(sb, " %d%%",
2413 					    (int)(sd->sd_rebuild_pos * 100 /
2414 					     sd->sd_size));
2415 				}
2416 				if (TAILQ_NEXT(sd, sd_next))
2417 					sbuf_cat(sb, ", ");
2418 			}
2419 			sbuf_cat(sb, ")");
2420 		}
2421 		sbuf_cat(sb, "</State>\n");
2422 		sbuf_printf(sb, "%s<Subdisks>", indent);
2423 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2424 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2425 			    sd->sd_volume->v_global_id,
2426 			    sd->sd_volume->v_name,
2427 			    sd->sd_pos, (uintmax_t)sd->sd_offset);
2428 			if (TAILQ_NEXT(sd, sd_next))
2429 				sbuf_cat(sb, ", ");
2430 		}
2431 		sbuf_cat(sb, "</Subdisks>\n");
2432 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2433 		    disk->d_read_errs);
2434 		sx_xunlock(&sc->sc_lock);
2435 		g_topology_lock();
2436 	} else {
2437 		g_topology_unlock();
2438 		sx_xlock(&sc->sc_lock);
2439 		if (sc->sc_md) {
2440 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2441 			    sc->sc_md->mdo_class->name);
2442 		}
2443 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2444 			s = 0xff;
2445 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2446 				if (vol->v_state < s)
2447 					s = vol->v_state;
2448 			}
2449 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2450 			    g_raid_volume_state2str(s));
2451 		}
2452 		sx_xunlock(&sc->sc_lock);
2453 		g_topology_lock();
2454 	}
2455 }
2456 
2457 static void
2458 g_raid_shutdown_post_sync(void *arg, int howto)
2459 {
2460 	struct g_class *mp;
2461 	struct g_geom *gp, *gp2;
2462 	struct g_raid_softc *sc;
2463 	struct g_raid_volume *vol;
2464 
2465 	mp = arg;
2466 	g_topology_lock();
2467 	g_raid_shutdown = 1;
2468 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2469 		if ((sc = gp->softc) == NULL)
2470 			continue;
2471 		g_topology_unlock();
2472 		sx_xlock(&sc->sc_lock);
2473 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2474 			g_raid_clean(vol, -1);
2475 		g_cancel_event(sc);
2476 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2477 		g_topology_lock();
2478 	}
2479 	g_topology_unlock();
2480 }
2481 
2482 static void
2483 g_raid_init(struct g_class *mp)
2484 {
2485 
2486 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2487 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2488 	if (g_raid_post_sync == NULL)
2489 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2490 	g_raid_started = 1;
2491 }
2492 
2493 static void
2494 g_raid_fini(struct g_class *mp)
2495 {
2496 
2497 	if (g_raid_post_sync != NULL)
2498 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2499 	g_raid_started = 0;
2500 }
2501 
2502 int
2503 g_raid_md_modevent(module_t mod, int type, void *arg)
2504 {
2505 	struct g_raid_md_class *class, *c, *nc;
2506 	int error;
2507 
2508 	error = 0;
2509 	class = arg;
2510 	switch (type) {
2511 	case MOD_LOAD:
2512 		c = LIST_FIRST(&g_raid_md_classes);
2513 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2514 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2515 		else {
2516 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2517 			    nc->mdc_priority < class->mdc_priority)
2518 				c = nc;
2519 			LIST_INSERT_AFTER(c, class, mdc_list);
2520 		}
2521 		if (g_raid_started)
2522 			g_retaste(&g_raid_class);
2523 		break;
2524 	case MOD_UNLOAD:
2525 		LIST_REMOVE(class, mdc_list);
2526 		break;
2527 	default:
2528 		error = EOPNOTSUPP;
2529 		break;
2530 	}
2531 
2532 	return (error);
2533 }
2534 
2535 int
2536 g_raid_tr_modevent(module_t mod, int type, void *arg)
2537 {
2538 	struct g_raid_tr_class *class, *c, *nc;
2539 	int error;
2540 
2541 	error = 0;
2542 	class = arg;
2543 	switch (type) {
2544 	case MOD_LOAD:
2545 		c = LIST_FIRST(&g_raid_tr_classes);
2546 		if (c == NULL || c->trc_priority > class->trc_priority)
2547 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2548 		else {
2549 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2550 			    nc->trc_priority < class->trc_priority)
2551 				c = nc;
2552 			LIST_INSERT_AFTER(c, class, trc_list);
2553 		}
2554 		break;
2555 	case MOD_UNLOAD:
2556 		LIST_REMOVE(class, trc_list);
2557 		break;
2558 	default:
2559 		error = EOPNOTSUPP;
2560 		break;
2561 	}
2562 
2563 	return (error);
2564 }
2565 
2566 /*
2567  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2568  * to reduce module priority, allowing submodules to register them first.
2569  */
2570 static moduledata_t g_raid_mod = {
2571 	"g_raid",
2572 	g_modevent,
2573 	&g_raid_class
2574 };
2575 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2576 MODULE_VERSION(geom_raid, 0);
2577