xref: /freebsd/sys/geom/raid/g_raid.c (revision 0a10f22a30d61a6f32777a236a82d461129538cc)
1 /*-
2  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/limits.h>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/bio.h>
38 #include <sys/sbuf.h>
39 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
41 #include <sys/eventhandler.h>
42 #include <vm/uma.h>
43 #include <geom/geom.h>
44 #include <sys/proc.h>
45 #include <sys/kthread.h>
46 #include <sys/sched.h>
47 #include <geom/raid/g_raid.h>
48 #include "g_raid_md_if.h"
49 #include "g_raid_tr_if.h"
50 
51 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
52 
53 SYSCTL_DECL(_kern_geom);
54 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
55 int g_raid_enable = 1;
56 TUNABLE_INT("kern.geom.raid.enable", &g_raid_enable);
57 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RW,
58     &g_raid_enable, 0, "Enable on-disk metadata taste");
59 u_int g_raid_aggressive_spare = 0;
60 TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare);
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW,
62     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug);
65 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0,
66     "Debug level");
67 int g_raid_read_err_thresh = 10;
68 TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh);
69 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW,
70     &g_raid_read_err_thresh, 0,
71     "Number of read errors equated to disk failure");
72 u_int g_raid_start_timeout = 30;
73 TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout);
74 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW,
75     &g_raid_start_timeout, 0,
76     "Time to wait for all array components");
77 static u_int g_raid_clean_time = 5;
78 TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time);
79 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW,
80     &g_raid_clean_time, 0, "Mark volume as clean when idling");
81 static u_int g_raid_disconnect_on_failure = 1;
82 TUNABLE_INT("kern.geom.raid.disconnect_on_failure",
83     &g_raid_disconnect_on_failure);
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW,
85     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
86 static u_int g_raid_name_format = 0;
87 TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format);
88 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW,
89     &g_raid_name_format, 0, "Providers name format.");
90 static u_int g_raid_idle_threshold = 1000000;
91 TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold);
92 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW,
93     &g_raid_idle_threshold, 1000000,
94     "Time in microseconds to consider a volume idle.");
95 static u_int ar_legacy_aliases = 1;
96 SYSCTL_INT(_kern_geom_raid, OID_AUTO, legacy_aliases, CTLFLAG_RW,
97            &ar_legacy_aliases, 0, "Create aliases named as the legacy ataraid style.");
98 TUNABLE_INT("kern.geom_raid.legacy_aliases", &ar_legacy_aliases);
99 
100 
101 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
102 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
103 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
104 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
105 } while (0)
106 
107 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
108     LIST_HEAD_INITIALIZER(g_raid_md_classes);
109 
110 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
111     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
112 
113 LIST_HEAD(, g_raid_volume) g_raid_volumes =
114     LIST_HEAD_INITIALIZER(g_raid_volumes);
115 
116 static eventhandler_tag g_raid_post_sync = NULL;
117 static int g_raid_started = 0;
118 static int g_raid_shutdown = 0;
119 
120 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
121     struct g_geom *gp);
122 static g_taste_t g_raid_taste;
123 static void g_raid_init(struct g_class *mp);
124 static void g_raid_fini(struct g_class *mp);
125 
126 struct g_class g_raid_class = {
127 	.name = G_RAID_CLASS_NAME,
128 	.version = G_VERSION,
129 	.ctlreq = g_raid_ctl,
130 	.taste = g_raid_taste,
131 	.destroy_geom = g_raid_destroy_geom,
132 	.init = g_raid_init,
133 	.fini = g_raid_fini
134 };
135 
136 static void g_raid_destroy_provider(struct g_raid_volume *vol);
137 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
138 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
139 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
140 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
141 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
142     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
143 static void g_raid_start(struct bio *bp);
144 static void g_raid_start_request(struct bio *bp);
145 static void g_raid_disk_done(struct bio *bp);
146 static void g_raid_poll(struct g_raid_softc *sc);
147 
148 static const char *
149 g_raid_node_event2str(int event)
150 {
151 
152 	switch (event) {
153 	case G_RAID_NODE_E_WAKE:
154 		return ("WAKE");
155 	case G_RAID_NODE_E_START:
156 		return ("START");
157 	default:
158 		return ("INVALID");
159 	}
160 }
161 
162 const char *
163 g_raid_disk_state2str(int state)
164 {
165 
166 	switch (state) {
167 	case G_RAID_DISK_S_NONE:
168 		return ("NONE");
169 	case G_RAID_DISK_S_OFFLINE:
170 		return ("OFFLINE");
171 	case G_RAID_DISK_S_DISABLED:
172 		return ("DISABLED");
173 	case G_RAID_DISK_S_FAILED:
174 		return ("FAILED");
175 	case G_RAID_DISK_S_STALE_FAILED:
176 		return ("STALE_FAILED");
177 	case G_RAID_DISK_S_SPARE:
178 		return ("SPARE");
179 	case G_RAID_DISK_S_STALE:
180 		return ("STALE");
181 	case G_RAID_DISK_S_ACTIVE:
182 		return ("ACTIVE");
183 	default:
184 		return ("INVALID");
185 	}
186 }
187 
188 static const char *
189 g_raid_disk_event2str(int event)
190 {
191 
192 	switch (event) {
193 	case G_RAID_DISK_E_DISCONNECTED:
194 		return ("DISCONNECTED");
195 	default:
196 		return ("INVALID");
197 	}
198 }
199 
200 const char *
201 g_raid_subdisk_state2str(int state)
202 {
203 
204 	switch (state) {
205 	case G_RAID_SUBDISK_S_NONE:
206 		return ("NONE");
207 	case G_RAID_SUBDISK_S_FAILED:
208 		return ("FAILED");
209 	case G_RAID_SUBDISK_S_NEW:
210 		return ("NEW");
211 	case G_RAID_SUBDISK_S_REBUILD:
212 		return ("REBUILD");
213 	case G_RAID_SUBDISK_S_UNINITIALIZED:
214 		return ("UNINITIALIZED");
215 	case G_RAID_SUBDISK_S_STALE:
216 		return ("STALE");
217 	case G_RAID_SUBDISK_S_RESYNC:
218 		return ("RESYNC");
219 	case G_RAID_SUBDISK_S_ACTIVE:
220 		return ("ACTIVE");
221 	default:
222 		return ("INVALID");
223 	}
224 }
225 
226 static const char *
227 g_raid_subdisk_event2str(int event)
228 {
229 
230 	switch (event) {
231 	case G_RAID_SUBDISK_E_NEW:
232 		return ("NEW");
233 	case G_RAID_SUBDISK_E_FAILED:
234 		return ("FAILED");
235 	case G_RAID_SUBDISK_E_DISCONNECTED:
236 		return ("DISCONNECTED");
237 	default:
238 		return ("INVALID");
239 	}
240 }
241 
242 const char *
243 g_raid_volume_state2str(int state)
244 {
245 
246 	switch (state) {
247 	case G_RAID_VOLUME_S_STARTING:
248 		return ("STARTING");
249 	case G_RAID_VOLUME_S_BROKEN:
250 		return ("BROKEN");
251 	case G_RAID_VOLUME_S_DEGRADED:
252 		return ("DEGRADED");
253 	case G_RAID_VOLUME_S_SUBOPTIMAL:
254 		return ("SUBOPTIMAL");
255 	case G_RAID_VOLUME_S_OPTIMAL:
256 		return ("OPTIMAL");
257 	case G_RAID_VOLUME_S_UNSUPPORTED:
258 		return ("UNSUPPORTED");
259 	case G_RAID_VOLUME_S_STOPPED:
260 		return ("STOPPED");
261 	default:
262 		return ("INVALID");
263 	}
264 }
265 
266 static const char *
267 g_raid_volume_event2str(int event)
268 {
269 
270 	switch (event) {
271 	case G_RAID_VOLUME_E_UP:
272 		return ("UP");
273 	case G_RAID_VOLUME_E_DOWN:
274 		return ("DOWN");
275 	case G_RAID_VOLUME_E_START:
276 		return ("START");
277 	case G_RAID_VOLUME_E_STARTMD:
278 		return ("STARTMD");
279 	default:
280 		return ("INVALID");
281 	}
282 }
283 
284 const char *
285 g_raid_volume_level2str(int level, int qual)
286 {
287 
288 	switch (level) {
289 	case G_RAID_VOLUME_RL_RAID0:
290 		return ("RAID0");
291 	case G_RAID_VOLUME_RL_RAID1:
292 		return ("RAID1");
293 	case G_RAID_VOLUME_RL_RAID3:
294 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
295 			return ("RAID3-P0");
296 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
297 			return ("RAID3-PN");
298 		return ("RAID3");
299 	case G_RAID_VOLUME_RL_RAID4:
300 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
301 			return ("RAID4-P0");
302 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
303 			return ("RAID4-PN");
304 		return ("RAID4");
305 	case G_RAID_VOLUME_RL_RAID5:
306 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
307 			return ("RAID5-RA");
308 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
309 			return ("RAID5-RS");
310 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
311 			return ("RAID5-LA");
312 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
313 			return ("RAID5-LS");
314 		return ("RAID5");
315 	case G_RAID_VOLUME_RL_RAID6:
316 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
317 			return ("RAID6-RA");
318 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
319 			return ("RAID6-RS");
320 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
321 			return ("RAID6-LA");
322 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
323 			return ("RAID6-LS");
324 		return ("RAID6");
325 	case G_RAID_VOLUME_RL_RAIDMDF:
326 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
327 			return ("RAIDMDF-RA");
328 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
329 			return ("RAIDMDF-RS");
330 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
331 			return ("RAIDMDF-LA");
332 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
333 			return ("RAIDMDF-LS");
334 		return ("RAIDMDF");
335 	case G_RAID_VOLUME_RL_RAID1E:
336 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
337 			return ("RAID1E-A");
338 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
339 			return ("RAID1E-O");
340 		return ("RAID1E");
341 	case G_RAID_VOLUME_RL_SINGLE:
342 		return ("SINGLE");
343 	case G_RAID_VOLUME_RL_CONCAT:
344 		return ("CONCAT");
345 	case G_RAID_VOLUME_RL_RAID5E:
346 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
347 			return ("RAID5E-RA");
348 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
349 			return ("RAID5E-RS");
350 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
351 			return ("RAID5E-LA");
352 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
353 			return ("RAID5E-LS");
354 		return ("RAID5E");
355 	case G_RAID_VOLUME_RL_RAID5EE:
356 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
357 			return ("RAID5EE-RA");
358 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
359 			return ("RAID5EE-RS");
360 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
361 			return ("RAID5EE-LA");
362 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
363 			return ("RAID5EE-LS");
364 		return ("RAID5EE");
365 	case G_RAID_VOLUME_RL_RAID5R:
366 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
367 			return ("RAID5R-RA");
368 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
369 			return ("RAID5R-RS");
370 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
371 			return ("RAID5R-LA");
372 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
373 			return ("RAID5R-LS");
374 		return ("RAID5E");
375 	default:
376 		return ("UNKNOWN");
377 	}
378 }
379 
380 int
381 g_raid_volume_str2level(const char *str, int *level, int *qual)
382 {
383 
384 	*level = G_RAID_VOLUME_RL_UNKNOWN;
385 	*qual = G_RAID_VOLUME_RLQ_NONE;
386 	if (strcasecmp(str, "RAID0") == 0)
387 		*level = G_RAID_VOLUME_RL_RAID0;
388 	else if (strcasecmp(str, "RAID1") == 0)
389 		*level = G_RAID_VOLUME_RL_RAID1;
390 	else if (strcasecmp(str, "RAID3-P0") == 0) {
391 		*level = G_RAID_VOLUME_RL_RAID3;
392 		*qual = G_RAID_VOLUME_RLQ_R3P0;
393 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
394 		   strcasecmp(str, "RAID3") == 0) {
395 		*level = G_RAID_VOLUME_RL_RAID3;
396 		*qual = G_RAID_VOLUME_RLQ_R3PN;
397 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
398 		*level = G_RAID_VOLUME_RL_RAID4;
399 		*qual = G_RAID_VOLUME_RLQ_R4P0;
400 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
401 		   strcasecmp(str, "RAID4") == 0) {
402 		*level = G_RAID_VOLUME_RL_RAID4;
403 		*qual = G_RAID_VOLUME_RLQ_R4PN;
404 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
405 		*level = G_RAID_VOLUME_RL_RAID5;
406 		*qual = G_RAID_VOLUME_RLQ_R5RA;
407 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
408 		*level = G_RAID_VOLUME_RL_RAID5;
409 		*qual = G_RAID_VOLUME_RLQ_R5RS;
410 	} else if (strcasecmp(str, "RAID5") == 0 ||
411 		   strcasecmp(str, "RAID5-LA") == 0) {
412 		*level = G_RAID_VOLUME_RL_RAID5;
413 		*qual = G_RAID_VOLUME_RLQ_R5LA;
414 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
415 		*level = G_RAID_VOLUME_RL_RAID5;
416 		*qual = G_RAID_VOLUME_RLQ_R5LS;
417 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
418 		*level = G_RAID_VOLUME_RL_RAID6;
419 		*qual = G_RAID_VOLUME_RLQ_R6RA;
420 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
421 		*level = G_RAID_VOLUME_RL_RAID6;
422 		*qual = G_RAID_VOLUME_RLQ_R6RS;
423 	} else if (strcasecmp(str, "RAID6") == 0 ||
424 		   strcasecmp(str, "RAID6-LA") == 0) {
425 		*level = G_RAID_VOLUME_RL_RAID6;
426 		*qual = G_RAID_VOLUME_RLQ_R6LA;
427 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
428 		*level = G_RAID_VOLUME_RL_RAID6;
429 		*qual = G_RAID_VOLUME_RLQ_R6LS;
430 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
431 		*level = G_RAID_VOLUME_RL_RAIDMDF;
432 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
433 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
434 		*level = G_RAID_VOLUME_RL_RAIDMDF;
435 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
436 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
437 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
438 		*level = G_RAID_VOLUME_RL_RAIDMDF;
439 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
440 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
441 		*level = G_RAID_VOLUME_RL_RAIDMDF;
442 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
443 	} else if (strcasecmp(str, "RAID10") == 0 ||
444 		   strcasecmp(str, "RAID1E") == 0 ||
445 		   strcasecmp(str, "RAID1E-A") == 0) {
446 		*level = G_RAID_VOLUME_RL_RAID1E;
447 		*qual = G_RAID_VOLUME_RLQ_R1EA;
448 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
449 		*level = G_RAID_VOLUME_RL_RAID1E;
450 		*qual = G_RAID_VOLUME_RLQ_R1EO;
451 	} else if (strcasecmp(str, "SINGLE") == 0)
452 		*level = G_RAID_VOLUME_RL_SINGLE;
453 	else if (strcasecmp(str, "CONCAT") == 0)
454 		*level = G_RAID_VOLUME_RL_CONCAT;
455 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
456 		*level = G_RAID_VOLUME_RL_RAID5E;
457 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
458 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
459 		*level = G_RAID_VOLUME_RL_RAID5E;
460 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
461 	} else if (strcasecmp(str, "RAID5E") == 0 ||
462 		   strcasecmp(str, "RAID5E-LA") == 0) {
463 		*level = G_RAID_VOLUME_RL_RAID5E;
464 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
465 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
466 		*level = G_RAID_VOLUME_RL_RAID5E;
467 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
468 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
469 		*level = G_RAID_VOLUME_RL_RAID5EE;
470 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
471 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
472 		*level = G_RAID_VOLUME_RL_RAID5EE;
473 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
474 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
475 		   strcasecmp(str, "RAID5EE-LA") == 0) {
476 		*level = G_RAID_VOLUME_RL_RAID5EE;
477 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
478 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
479 		*level = G_RAID_VOLUME_RL_RAID5EE;
480 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
481 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
482 		*level = G_RAID_VOLUME_RL_RAID5R;
483 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
484 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
485 		*level = G_RAID_VOLUME_RL_RAID5R;
486 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
487 	} else if (strcasecmp(str, "RAID5R") == 0 ||
488 		   strcasecmp(str, "RAID5R-LA") == 0) {
489 		*level = G_RAID_VOLUME_RL_RAID5R;
490 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
491 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
492 		*level = G_RAID_VOLUME_RL_RAID5R;
493 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
494 	} else
495 		return (-1);
496 	return (0);
497 }
498 
499 const char *
500 g_raid_get_diskname(struct g_raid_disk *disk)
501 {
502 
503 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
504 		return ("[unknown]");
505 	return (disk->d_consumer->provider->name);
506 }
507 
508 void
509 g_raid_get_disk_info(struct g_raid_disk *disk)
510 {
511 	struct g_consumer *cp = disk->d_consumer;
512 	int error, len;
513 
514 	/* Read kernel dumping information. */
515 	disk->d_kd.offset = 0;
516 	disk->d_kd.length = OFF_MAX;
517 	len = sizeof(disk->d_kd);
518 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
519 	if (error)
520 		disk->d_kd.di.dumper = NULL;
521 	if (disk->d_kd.di.dumper == NULL)
522 		G_RAID_DEBUG1(2, disk->d_softc,
523 		    "Dumping not supported by %s: %d.",
524 		    cp->provider->name, error);
525 
526 	/* Read BIO_DELETE support. */
527 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
528 	if (error)
529 		disk->d_candelete = 0;
530 	if (!disk->d_candelete)
531 		G_RAID_DEBUG1(2, disk->d_softc,
532 		    "BIO_DELETE not supported by %s: %d.",
533 		    cp->provider->name, error);
534 }
535 
536 void
537 g_raid_report_disk_state(struct g_raid_disk *disk)
538 {
539 	struct g_raid_subdisk *sd;
540 	int len, state;
541 	uint32_t s;
542 
543 	if (disk->d_consumer == NULL)
544 		return;
545 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
546 		s = G_STATE_ACTIVE; /* XXX */
547 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
548 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
549 		s = G_STATE_FAILED;
550 	} else {
551 		state = G_RAID_SUBDISK_S_ACTIVE;
552 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
553 			if (sd->sd_state < state)
554 				state = sd->sd_state;
555 		}
556 		if (state == G_RAID_SUBDISK_S_FAILED)
557 			s = G_STATE_FAILED;
558 		else if (state == G_RAID_SUBDISK_S_NEW ||
559 		    state == G_RAID_SUBDISK_S_REBUILD)
560 			s = G_STATE_REBUILD;
561 		else if (state == G_RAID_SUBDISK_S_STALE ||
562 		    state == G_RAID_SUBDISK_S_RESYNC)
563 			s = G_STATE_RESYNC;
564 		else
565 			s = G_STATE_ACTIVE;
566 	}
567 	len = sizeof(s);
568 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
569 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
570 	    g_raid_get_diskname(disk), s);
571 }
572 
573 void
574 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
575 {
576 
577 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
578 	    g_raid_get_diskname(disk),
579 	    g_raid_disk_state2str(disk->d_state),
580 	    g_raid_disk_state2str(state));
581 	disk->d_state = state;
582 	g_raid_report_disk_state(disk);
583 }
584 
585 void
586 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
587 {
588 
589 	G_RAID_DEBUG1(0, sd->sd_softc,
590 	    "Subdisk %s:%d-%s state changed from %s to %s.",
591 	    sd->sd_volume->v_name, sd->sd_pos,
592 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
593 	    g_raid_subdisk_state2str(sd->sd_state),
594 	    g_raid_subdisk_state2str(state));
595 	sd->sd_state = state;
596 	if (sd->sd_disk)
597 		g_raid_report_disk_state(sd->sd_disk);
598 }
599 
600 void
601 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
602 {
603 
604 	G_RAID_DEBUG1(0, vol->v_softc,
605 	    "Volume %s state changed from %s to %s.",
606 	    vol->v_name,
607 	    g_raid_volume_state2str(vol->v_state),
608 	    g_raid_volume_state2str(state));
609 	vol->v_state = state;
610 }
611 
612 /*
613  * --- Events handling functions ---
614  * Events in geom_raid are used to maintain subdisks and volumes status
615  * from one thread to simplify locking.
616  */
617 static void
618 g_raid_event_free(struct g_raid_event *ep)
619 {
620 
621 	free(ep, M_RAID);
622 }
623 
624 int
625 g_raid_event_send(void *arg, int event, int flags)
626 {
627 	struct g_raid_softc *sc;
628 	struct g_raid_event *ep;
629 	int error;
630 
631 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
632 		sc = ((struct g_raid_volume *)arg)->v_softc;
633 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
634 		sc = ((struct g_raid_disk *)arg)->d_softc;
635 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
636 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
637 	} else {
638 		sc = arg;
639 	}
640 	ep = malloc(sizeof(*ep), M_RAID,
641 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
642 	if (ep == NULL)
643 		return (ENOMEM);
644 	ep->e_tgt = arg;
645 	ep->e_event = event;
646 	ep->e_flags = flags;
647 	ep->e_error = 0;
648 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
649 	mtx_lock(&sc->sc_queue_mtx);
650 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
651 	mtx_unlock(&sc->sc_queue_mtx);
652 	wakeup(sc);
653 
654 	if ((flags & G_RAID_EVENT_WAIT) == 0)
655 		return (0);
656 
657 	sx_assert(&sc->sc_lock, SX_XLOCKED);
658 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
659 	sx_xunlock(&sc->sc_lock);
660 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
661 		mtx_lock(&sc->sc_queue_mtx);
662 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
663 		    hz * 5);
664 	}
665 	error = ep->e_error;
666 	g_raid_event_free(ep);
667 	sx_xlock(&sc->sc_lock);
668 	return (error);
669 }
670 
671 static void
672 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
673 {
674 	struct g_raid_event *ep, *tmpep;
675 
676 	sx_assert(&sc->sc_lock, SX_XLOCKED);
677 
678 	mtx_lock(&sc->sc_queue_mtx);
679 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
680 		if (ep->e_tgt != tgt)
681 			continue;
682 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
683 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
684 			g_raid_event_free(ep);
685 		else {
686 			ep->e_error = ECANCELED;
687 			wakeup(ep);
688 		}
689 	}
690 	mtx_unlock(&sc->sc_queue_mtx);
691 }
692 
693 static int
694 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
695 {
696 	struct g_raid_event *ep;
697 	int	res = 0;
698 
699 	sx_assert(&sc->sc_lock, SX_XLOCKED);
700 
701 	mtx_lock(&sc->sc_queue_mtx);
702 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
703 		if (ep->e_tgt != tgt)
704 			continue;
705 		res = 1;
706 		break;
707 	}
708 	mtx_unlock(&sc->sc_queue_mtx);
709 	return (res);
710 }
711 
712 /*
713  * Return the number of disks in given state.
714  * If state is equal to -1, count all connected disks.
715  */
716 u_int
717 g_raid_ndisks(struct g_raid_softc *sc, int state)
718 {
719 	struct g_raid_disk *disk;
720 	u_int n;
721 
722 	sx_assert(&sc->sc_lock, SX_LOCKED);
723 
724 	n = 0;
725 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
726 		if (disk->d_state == state || state == -1)
727 			n++;
728 	}
729 	return (n);
730 }
731 
732 /*
733  * Return the number of subdisks in given state.
734  * If state is equal to -1, count all connected disks.
735  */
736 u_int
737 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
738 {
739 	struct g_raid_subdisk *subdisk;
740 	struct g_raid_softc *sc;
741 	u_int i, n ;
742 
743 	sc = vol->v_softc;
744 	sx_assert(&sc->sc_lock, SX_LOCKED);
745 
746 	n = 0;
747 	for (i = 0; i < vol->v_disks_count; i++) {
748 		subdisk = &vol->v_subdisks[i];
749 		if ((state == -1 &&
750 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
751 		    subdisk->sd_state == state)
752 			n++;
753 	}
754 	return (n);
755 }
756 
757 /*
758  * Return the first subdisk in given state.
759  * If state is equal to -1, then the first connected disks.
760  */
761 struct g_raid_subdisk *
762 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
763 {
764 	struct g_raid_subdisk *sd;
765 	struct g_raid_softc *sc;
766 	u_int i;
767 
768 	sc = vol->v_softc;
769 	sx_assert(&sc->sc_lock, SX_LOCKED);
770 
771 	for (i = 0; i < vol->v_disks_count; i++) {
772 		sd = &vol->v_subdisks[i];
773 		if ((state == -1 &&
774 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
775 		    sd->sd_state == state)
776 			return (sd);
777 	}
778 	return (NULL);
779 }
780 
781 struct g_consumer *
782 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
783 {
784 	struct g_consumer *cp;
785 	struct g_provider *pp;
786 
787 	g_topology_assert();
788 
789 	if (strncmp(name, "/dev/", 5) == 0)
790 		name += 5;
791 	pp = g_provider_by_name(name);
792 	if (pp == NULL)
793 		return (NULL);
794 	cp = g_new_consumer(sc->sc_geom);
795 	if (g_attach(cp, pp) != 0) {
796 		g_destroy_consumer(cp);
797 		return (NULL);
798 	}
799 	if (g_access(cp, 1, 1, 1) != 0) {
800 		g_detach(cp);
801 		g_destroy_consumer(cp);
802 		return (NULL);
803 	}
804 	return (cp);
805 }
806 
807 static u_int
808 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
809 {
810 	struct bio *bp;
811 	u_int nreqs = 0;
812 
813 	mtx_lock(&sc->sc_queue_mtx);
814 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
815 		if (bp->bio_from == cp)
816 			nreqs++;
817 	}
818 	mtx_unlock(&sc->sc_queue_mtx);
819 	return (nreqs);
820 }
821 
822 u_int
823 g_raid_nopens(struct g_raid_softc *sc)
824 {
825 	struct g_raid_volume *vol;
826 	u_int opens;
827 
828 	opens = 0;
829 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
830 		if (vol->v_provider_open != 0)
831 			opens++;
832 	}
833 	return (opens);
834 }
835 
836 static int
837 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
838 {
839 
840 	if (cp->index > 0) {
841 		G_RAID_DEBUG1(2, sc,
842 		    "I/O requests for %s exist, can't destroy it now.",
843 		    cp->provider->name);
844 		return (1);
845 	}
846 	if (g_raid_nrequests(sc, cp) > 0) {
847 		G_RAID_DEBUG1(2, sc,
848 		    "I/O requests for %s in queue, can't destroy it now.",
849 		    cp->provider->name);
850 		return (1);
851 	}
852 	return (0);
853 }
854 
855 static void
856 g_raid_destroy_consumer(void *arg, int flags __unused)
857 {
858 	struct g_consumer *cp;
859 
860 	g_topology_assert();
861 
862 	cp = arg;
863 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
864 	g_detach(cp);
865 	g_destroy_consumer(cp);
866 }
867 
868 void
869 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
870 {
871 	struct g_provider *pp;
872 	int retaste_wait;
873 
874 	g_topology_assert_not();
875 
876 	g_topology_lock();
877 	cp->private = NULL;
878 	if (g_raid_consumer_is_busy(sc, cp))
879 		goto out;
880 	pp = cp->provider;
881 	retaste_wait = 0;
882 	if (cp->acw == 1) {
883 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
884 			retaste_wait = 1;
885 	}
886 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
887 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
888 	if (retaste_wait) {
889 		/*
890 		 * After retaste event was send (inside g_access()), we can send
891 		 * event to detach and destroy consumer.
892 		 * A class, which has consumer to the given provider connected
893 		 * will not receive retaste event for the provider.
894 		 * This is the way how I ignore retaste events when I close
895 		 * consumers opened for write: I detach and destroy consumer
896 		 * after retaste event is sent.
897 		 */
898 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
899 		goto out;
900 	}
901 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
902 	g_detach(cp);
903 	g_destroy_consumer(cp);
904 out:
905 	g_topology_unlock();
906 }
907 
908 static void
909 g_raid_orphan(struct g_consumer *cp)
910 {
911 	struct g_raid_disk *disk;
912 
913 	g_topology_assert();
914 
915 	disk = cp->private;
916 	if (disk == NULL)
917 		return;
918 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
919 	    G_RAID_EVENT_DISK);
920 }
921 
922 static void
923 g_raid_clean(struct g_raid_volume *vol, int acw)
924 {
925 	struct g_raid_softc *sc;
926 	int timeout;
927 
928 	sc = vol->v_softc;
929 	g_topology_assert_not();
930 	sx_assert(&sc->sc_lock, SX_XLOCKED);
931 
932 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
933 //		return;
934 	if (!vol->v_dirty)
935 		return;
936 	if (vol->v_writes > 0)
937 		return;
938 	if (acw > 0 || (acw == -1 &&
939 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
940 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
941 		if (!g_raid_shutdown && timeout > 0)
942 			return;
943 	}
944 	vol->v_dirty = 0;
945 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
946 	    vol->v_name);
947 	g_raid_write_metadata(sc, vol, NULL, NULL);
948 }
949 
950 static void
951 g_raid_dirty(struct g_raid_volume *vol)
952 {
953 	struct g_raid_softc *sc;
954 
955 	sc = vol->v_softc;
956 	g_topology_assert_not();
957 	sx_assert(&sc->sc_lock, SX_XLOCKED);
958 
959 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
960 //		return;
961 	vol->v_dirty = 1;
962 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
963 	    vol->v_name);
964 	g_raid_write_metadata(sc, vol, NULL, NULL);
965 }
966 
967 void
968 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
969 {
970 	struct g_raid_softc *sc;
971 	struct g_raid_volume *vol;
972 	struct g_raid_subdisk *sd;
973 	struct bio_queue_head queue;
974 	struct bio *cbp;
975 	int i;
976 
977 	vol = tr->tro_volume;
978 	sc = vol->v_softc;
979 
980 	/*
981 	 * Allocate all bios before sending any request, so we can return
982 	 * ENOMEM in nice and clean way.
983 	 */
984 	bioq_init(&queue);
985 	for (i = 0; i < vol->v_disks_count; i++) {
986 		sd = &vol->v_subdisks[i];
987 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
988 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
989 			continue;
990 		cbp = g_clone_bio(bp);
991 		if (cbp == NULL)
992 			goto failure;
993 		cbp->bio_caller1 = sd;
994 		bioq_insert_tail(&queue, cbp);
995 	}
996 	for (cbp = bioq_first(&queue); cbp != NULL;
997 	    cbp = bioq_first(&queue)) {
998 		bioq_remove(&queue, cbp);
999 		sd = cbp->bio_caller1;
1000 		cbp->bio_caller1 = NULL;
1001 		g_raid_subdisk_iostart(sd, cbp);
1002 	}
1003 	return;
1004 failure:
1005 	for (cbp = bioq_first(&queue); cbp != NULL;
1006 	    cbp = bioq_first(&queue)) {
1007 		bioq_remove(&queue, cbp);
1008 		g_destroy_bio(cbp);
1009 	}
1010 	if (bp->bio_error == 0)
1011 		bp->bio_error = ENOMEM;
1012 	g_raid_iodone(bp, bp->bio_error);
1013 }
1014 
1015 static void
1016 g_raid_tr_kerneldump_common_done(struct bio *bp)
1017 {
1018 
1019 	bp->bio_flags |= BIO_DONE;
1020 }
1021 
1022 int
1023 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1024     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1025 {
1026 	struct g_raid_softc *sc;
1027 	struct g_raid_volume *vol;
1028 	struct bio bp;
1029 
1030 	vol = tr->tro_volume;
1031 	sc = vol->v_softc;
1032 
1033 	bzero(&bp, sizeof(bp));
1034 	bp.bio_cmd = BIO_WRITE;
1035 	bp.bio_done = g_raid_tr_kerneldump_common_done;
1036 	bp.bio_attribute = NULL;
1037 	bp.bio_offset = offset;
1038 	bp.bio_length = length;
1039 	bp.bio_data = virtual;
1040 	bp.bio_to = vol->v_provider;
1041 
1042 	g_raid_start(&bp);
1043 	while (!(bp.bio_flags & BIO_DONE)) {
1044 		G_RAID_DEBUG1(4, sc, "Poll...");
1045 		g_raid_poll(sc);
1046 		DELAY(10);
1047 	}
1048 
1049 	return (bp.bio_error != 0 ? EIO : 0);
1050 }
1051 
1052 static int
1053 g_raid_dump(void *arg,
1054     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1055 {
1056 	struct g_raid_volume *vol;
1057 	int error;
1058 
1059 	vol = (struct g_raid_volume *)arg;
1060 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1061 	    (long long unsigned)offset, (long long unsigned)length);
1062 
1063 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1064 	    virtual, physical, offset, length);
1065 	return (error);
1066 }
1067 
1068 static void
1069 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1070 {
1071 	struct g_kerneldump *gkd;
1072 	struct g_provider *pp;
1073 	struct g_raid_volume *vol;
1074 
1075 	gkd = (struct g_kerneldump*)bp->bio_data;
1076 	pp = bp->bio_to;
1077 	vol = pp->private;
1078 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1079 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1080 	gkd->di.dumper = g_raid_dump;
1081 	gkd->di.priv = vol;
1082 	gkd->di.blocksize = vol->v_sectorsize;
1083 	gkd->di.maxiosize = DFLTPHYS;
1084 	gkd->di.mediaoffset = gkd->offset;
1085 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1086 		gkd->length = vol->v_mediasize - gkd->offset;
1087 	gkd->di.mediasize = gkd->length;
1088 	g_io_deliver(bp, 0);
1089 }
1090 
1091 static void
1092 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1093 {
1094 	struct g_provider *pp;
1095 	struct g_raid_volume *vol;
1096 	struct g_raid_subdisk *sd;
1097 	int *val;
1098 	int i;
1099 
1100 	val = (int *)bp->bio_data;
1101 	pp = bp->bio_to;
1102 	vol = pp->private;
1103 	*val = 0;
1104 	for (i = 0; i < vol->v_disks_count; i++) {
1105 		sd = &vol->v_subdisks[i];
1106 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1107 			continue;
1108 		if (sd->sd_disk->d_candelete) {
1109 			*val = 1;
1110 			break;
1111 		}
1112 	}
1113 	g_io_deliver(bp, 0);
1114 }
1115 
1116 static void
1117 g_raid_start(struct bio *bp)
1118 {
1119 	struct g_raid_softc *sc;
1120 
1121 	sc = bp->bio_to->geom->softc;
1122 	/*
1123 	 * If sc == NULL or there are no valid disks, provider's error
1124 	 * should be set and g_raid_start() should not be called at all.
1125 	 */
1126 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1127 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1128 //	    bp->bio_to->error, bp->bio_to->name));
1129 	G_RAID_LOGREQ(3, bp, "Request received.");
1130 
1131 	switch (bp->bio_cmd) {
1132 	case BIO_READ:
1133 	case BIO_WRITE:
1134 	case BIO_DELETE:
1135 	case BIO_FLUSH:
1136 		break;
1137 	case BIO_GETATTR:
1138 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1139 			g_raid_candelete(sc, bp);
1140 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1141 			g_raid_kerneldump(sc, bp);
1142 		else
1143 			g_io_deliver(bp, EOPNOTSUPP);
1144 		return;
1145 	default:
1146 		g_io_deliver(bp, EOPNOTSUPP);
1147 		return;
1148 	}
1149 	mtx_lock(&sc->sc_queue_mtx);
1150 	bioq_disksort(&sc->sc_queue, bp);
1151 	mtx_unlock(&sc->sc_queue_mtx);
1152 	if (!dumping) {
1153 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1154 		wakeup(sc);
1155 	}
1156 }
1157 
1158 static int
1159 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1160 {
1161 	/*
1162 	 * 5 cases:
1163 	 * (1) bp entirely below NO
1164 	 * (2) bp entirely above NO
1165 	 * (3) bp start below, but end in range YES
1166 	 * (4) bp entirely within YES
1167 	 * (5) bp starts within, ends above YES
1168 	 *
1169 	 * lock range 10-19 (offset 10 length 10)
1170 	 * (1) 1-5: first if kicks it out
1171 	 * (2) 30-35: second if kicks it out
1172 	 * (3) 5-15: passes both ifs
1173 	 * (4) 12-14: passes both ifs
1174 	 * (5) 19-20: passes both
1175 	 */
1176 	off_t lend = lstart + len - 1;
1177 	off_t bstart = bp->bio_offset;
1178 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1179 
1180 	if (bend < lstart)
1181 		return (0);
1182 	if (lend < bstart)
1183 		return (0);
1184 	return (1);
1185 }
1186 
1187 static int
1188 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1189 {
1190 	struct g_raid_lock *lp;
1191 
1192 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1193 
1194 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1195 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1196 			return (1);
1197 	}
1198 	return (0);
1199 }
1200 
1201 static void
1202 g_raid_start_request(struct bio *bp)
1203 {
1204 	struct g_raid_softc *sc;
1205 	struct g_raid_volume *vol;
1206 
1207 	sc = bp->bio_to->geom->softc;
1208 	sx_assert(&sc->sc_lock, SX_LOCKED);
1209 	vol = bp->bio_to->private;
1210 
1211 	/*
1212 	 * Check to see if this item is in a locked range.  If so,
1213 	 * queue it to our locked queue and return.  We'll requeue
1214 	 * it when the range is unlocked.  Internal I/O for the
1215 	 * rebuild/rescan/recovery process is excluded from this
1216 	 * check so we can actually do the recovery.
1217 	 */
1218 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1219 	    g_raid_is_in_locked_range(vol, bp)) {
1220 		G_RAID_LOGREQ(3, bp, "Defer request.");
1221 		bioq_insert_tail(&vol->v_locked, bp);
1222 		return;
1223 	}
1224 
1225 	/*
1226 	 * If we're actually going to do the write/delete, then
1227 	 * update the idle stats for the volume.
1228 	 */
1229 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1230 		if (!vol->v_dirty)
1231 			g_raid_dirty(vol);
1232 		vol->v_writes++;
1233 	}
1234 
1235 	/*
1236 	 * Put request onto inflight queue, so we can check if new
1237 	 * synchronization requests don't collide with it.  Then tell
1238 	 * the transformation layer to start the I/O.
1239 	 */
1240 	bioq_insert_tail(&vol->v_inflight, bp);
1241 	G_RAID_LOGREQ(4, bp, "Request started");
1242 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1243 }
1244 
1245 static void
1246 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1247 {
1248 	off_t off, len;
1249 	struct bio *nbp;
1250 	struct g_raid_lock *lp;
1251 
1252 	vol->v_pending_lock = 0;
1253 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1254 		if (lp->l_pending) {
1255 			off = lp->l_offset;
1256 			len = lp->l_length;
1257 			lp->l_pending = 0;
1258 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1259 				if (g_raid_bio_overlaps(nbp, off, len))
1260 					lp->l_pending++;
1261 			}
1262 			if (lp->l_pending) {
1263 				vol->v_pending_lock = 1;
1264 				G_RAID_DEBUG1(4, vol->v_softc,
1265 				    "Deferred lock(%jd, %jd) has %d pending",
1266 				    (intmax_t)off, (intmax_t)(off + len),
1267 				    lp->l_pending);
1268 				continue;
1269 			}
1270 			G_RAID_DEBUG1(4, vol->v_softc,
1271 			    "Deferred lock of %jd to %jd completed",
1272 			    (intmax_t)off, (intmax_t)(off + len));
1273 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1274 		}
1275 	}
1276 }
1277 
1278 void
1279 g_raid_iodone(struct bio *bp, int error)
1280 {
1281 	struct g_raid_softc *sc;
1282 	struct g_raid_volume *vol;
1283 
1284 	sc = bp->bio_to->geom->softc;
1285 	sx_assert(&sc->sc_lock, SX_LOCKED);
1286 	vol = bp->bio_to->private;
1287 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1288 
1289 	/* Update stats if we done write/delete. */
1290 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1291 		vol->v_writes--;
1292 		vol->v_last_write = time_uptime;
1293 	}
1294 
1295 	bioq_remove(&vol->v_inflight, bp);
1296 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1297 		g_raid_finish_with_locked_ranges(vol, bp);
1298 	getmicrouptime(&vol->v_last_done);
1299 	g_io_deliver(bp, error);
1300 }
1301 
1302 int
1303 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1304     struct bio *ignore, void *argp)
1305 {
1306 	struct g_raid_softc *sc;
1307 	struct g_raid_lock *lp;
1308 	struct bio *bp;
1309 
1310 	sc = vol->v_softc;
1311 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1312 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1313 	lp->l_offset = off;
1314 	lp->l_length = len;
1315 	lp->l_callback_arg = argp;
1316 
1317 	lp->l_pending = 0;
1318 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1319 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1320 			lp->l_pending++;
1321 	}
1322 
1323 	/*
1324 	 * If there are any writes that are pending, we return EBUSY.  All
1325 	 * callers will have to wait until all pending writes clear.
1326 	 */
1327 	if (lp->l_pending > 0) {
1328 		vol->v_pending_lock = 1;
1329 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1330 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1331 		return (EBUSY);
1332 	}
1333 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1334 	    (intmax_t)off, (intmax_t)(off+len));
1335 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1336 	return (0);
1337 }
1338 
1339 int
1340 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1341 {
1342 	struct g_raid_lock *lp;
1343 	struct g_raid_softc *sc;
1344 	struct bio *bp;
1345 
1346 	sc = vol->v_softc;
1347 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1348 		if (lp->l_offset == off && lp->l_length == len) {
1349 			LIST_REMOVE(lp, l_next);
1350 			/* XXX
1351 			 * Right now we just put them all back on the queue
1352 			 * and hope for the best.  We hope this because any
1353 			 * locked ranges will go right back on this list
1354 			 * when the worker thread runs.
1355 			 * XXX
1356 			 */
1357 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1358 			    (intmax_t)lp->l_offset,
1359 			    (intmax_t)(lp->l_offset+lp->l_length));
1360 			mtx_lock(&sc->sc_queue_mtx);
1361 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1362 				bioq_disksort(&sc->sc_queue, bp);
1363 			mtx_unlock(&sc->sc_queue_mtx);
1364 			free(lp, M_RAID);
1365 			return (0);
1366 		}
1367 	}
1368 	return (EINVAL);
1369 }
1370 
1371 void
1372 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1373 {
1374 	struct g_consumer *cp;
1375 	struct g_raid_disk *disk, *tdisk;
1376 
1377 	bp->bio_caller1 = sd;
1378 
1379 	/*
1380 	 * Make sure that the disk is present. Generally it is a task of
1381 	 * transformation layers to not send requests to absent disks, but
1382 	 * it is better to be safe and report situation then sorry.
1383 	 */
1384 	if (sd->sd_disk == NULL) {
1385 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1386 nodisk:
1387 		bp->bio_from = NULL;
1388 		bp->bio_to = NULL;
1389 		bp->bio_error = ENXIO;
1390 		g_raid_disk_done(bp);
1391 		return;
1392 	}
1393 	disk = sd->sd_disk;
1394 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1395 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1396 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1397 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1398 		goto nodisk;
1399 	}
1400 
1401 	cp = disk->d_consumer;
1402 	bp->bio_from = cp;
1403 	bp->bio_to = cp->provider;
1404 	cp->index++;
1405 
1406 	/* Update average disks load. */
1407 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1408 		if (tdisk->d_consumer == NULL)
1409 			tdisk->d_load = 0;
1410 		else
1411 			tdisk->d_load = (tdisk->d_consumer->index *
1412 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1413 	}
1414 
1415 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1416 	if (dumping) {
1417 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1418 		if (bp->bio_cmd == BIO_WRITE) {
1419 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1420 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1421 		} else
1422 			bp->bio_error = EOPNOTSUPP;
1423 		g_raid_disk_done(bp);
1424 	} else {
1425 		bp->bio_done = g_raid_disk_done;
1426 		bp->bio_offset += sd->sd_offset;
1427 		G_RAID_LOGREQ(3, bp, "Sending request.");
1428 		g_io_request(bp, cp);
1429 	}
1430 }
1431 
1432 int
1433 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1434     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1435 {
1436 
1437 	if (sd->sd_disk == NULL)
1438 		return (ENXIO);
1439 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1440 		return (EOPNOTSUPP);
1441 	return (dump_write(&sd->sd_disk->d_kd.di,
1442 	    virtual, physical,
1443 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1444 	    length));
1445 }
1446 
1447 static void
1448 g_raid_disk_done(struct bio *bp)
1449 {
1450 	struct g_raid_softc *sc;
1451 	struct g_raid_subdisk *sd;
1452 
1453 	sd = bp->bio_caller1;
1454 	sc = sd->sd_softc;
1455 	mtx_lock(&sc->sc_queue_mtx);
1456 	bioq_disksort(&sc->sc_queue, bp);
1457 	mtx_unlock(&sc->sc_queue_mtx);
1458 	if (!dumping)
1459 		wakeup(sc);
1460 }
1461 
1462 static void
1463 g_raid_disk_done_request(struct bio *bp)
1464 {
1465 	struct g_raid_softc *sc;
1466 	struct g_raid_disk *disk;
1467 	struct g_raid_subdisk *sd;
1468 	struct g_raid_volume *vol;
1469 
1470 	g_topology_assert_not();
1471 
1472 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1473 	sd = bp->bio_caller1;
1474 	sc = sd->sd_softc;
1475 	vol = sd->sd_volume;
1476 	if (bp->bio_from != NULL) {
1477 		bp->bio_from->index--;
1478 		disk = bp->bio_from->private;
1479 		if (disk == NULL)
1480 			g_raid_kill_consumer(sc, bp->bio_from);
1481 	}
1482 	bp->bio_offset -= sd->sd_offset;
1483 
1484 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1485 }
1486 
1487 static void
1488 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1489 {
1490 
1491 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1492 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1493 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1494 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1495 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1496 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1497 	else
1498 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1499 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1500 		KASSERT(ep->e_error == 0,
1501 		    ("Error cannot be handled."));
1502 		g_raid_event_free(ep);
1503 	} else {
1504 		ep->e_flags |= G_RAID_EVENT_DONE;
1505 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1506 		mtx_lock(&sc->sc_queue_mtx);
1507 		wakeup(ep);
1508 		mtx_unlock(&sc->sc_queue_mtx);
1509 	}
1510 }
1511 
1512 /*
1513  * Worker thread.
1514  */
1515 static void
1516 g_raid_worker(void *arg)
1517 {
1518 	struct g_raid_softc *sc;
1519 	struct g_raid_event *ep;
1520 	struct g_raid_volume *vol;
1521 	struct bio *bp;
1522 	struct timeval now, t;
1523 	int timeout, rv;
1524 
1525 	sc = arg;
1526 	thread_lock(curthread);
1527 	sched_prio(curthread, PRIBIO);
1528 	thread_unlock(curthread);
1529 
1530 	sx_xlock(&sc->sc_lock);
1531 	for (;;) {
1532 		mtx_lock(&sc->sc_queue_mtx);
1533 		/*
1534 		 * First take a look at events.
1535 		 * This is important to handle events before any I/O requests.
1536 		 */
1537 		bp = NULL;
1538 		vol = NULL;
1539 		rv = 0;
1540 		ep = TAILQ_FIRST(&sc->sc_events);
1541 		if (ep != NULL)
1542 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1543 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1544 			;
1545 		else {
1546 			getmicrouptime(&now);
1547 			t = now;
1548 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1549 				if (bioq_first(&vol->v_inflight) == NULL &&
1550 				    vol->v_tr &&
1551 				    timevalcmp(&vol->v_last_done, &t, < ))
1552 					t = vol->v_last_done;
1553 			}
1554 			timevalsub(&t, &now);
1555 			timeout = g_raid_idle_threshold +
1556 			    t.tv_sec * 1000000 + t.tv_usec;
1557 			if (timeout > 0) {
1558 				/*
1559 				 * Two steps to avoid overflows at HZ=1000
1560 				 * and idle timeouts > 2.1s.  Some rounding
1561 				 * errors can occur, but they are < 1tick,
1562 				 * which is deemed to be close enough for
1563 				 * this purpose.
1564 				 */
1565 				int micpertic = 1000000 / hz;
1566 				timeout = (timeout + micpertic - 1) / micpertic;
1567 				sx_xunlock(&sc->sc_lock);
1568 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1569 				    PRIBIO | PDROP, "-", timeout);
1570 				sx_xlock(&sc->sc_lock);
1571 				goto process;
1572 			} else
1573 				rv = EWOULDBLOCK;
1574 		}
1575 		mtx_unlock(&sc->sc_queue_mtx);
1576 process:
1577 		if (ep != NULL) {
1578 			g_raid_handle_event(sc, ep);
1579 		} else if (bp != NULL) {
1580 			if (bp->bio_to != NULL &&
1581 			    bp->bio_to->geom == sc->sc_geom)
1582 				g_raid_start_request(bp);
1583 			else
1584 				g_raid_disk_done_request(bp);
1585 		} else if (rv == EWOULDBLOCK) {
1586 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1587 				g_raid_clean(vol, -1);
1588 				if (bioq_first(&vol->v_inflight) == NULL &&
1589 				    vol->v_tr) {
1590 					t.tv_sec = g_raid_idle_threshold / 1000000;
1591 					t.tv_usec = g_raid_idle_threshold % 1000000;
1592 					timevaladd(&t, &vol->v_last_done);
1593 					getmicrouptime(&now);
1594 					if (timevalcmp(&t, &now, <= )) {
1595 						G_RAID_TR_IDLE(vol->v_tr);
1596 						vol->v_last_done = now;
1597 					}
1598 				}
1599 			}
1600 		}
1601 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1602 			g_raid_destroy_node(sc, 1);	/* May not return. */
1603 	}
1604 }
1605 
1606 static void
1607 g_raid_poll(struct g_raid_softc *sc)
1608 {
1609 	struct g_raid_event *ep;
1610 	struct bio *bp;
1611 
1612 	sx_xlock(&sc->sc_lock);
1613 	mtx_lock(&sc->sc_queue_mtx);
1614 	/*
1615 	 * First take a look at events.
1616 	 * This is important to handle events before any I/O requests.
1617 	 */
1618 	ep = TAILQ_FIRST(&sc->sc_events);
1619 	if (ep != NULL) {
1620 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1621 		mtx_unlock(&sc->sc_queue_mtx);
1622 		g_raid_handle_event(sc, ep);
1623 		goto out;
1624 	}
1625 	bp = bioq_takefirst(&sc->sc_queue);
1626 	if (bp != NULL) {
1627 		mtx_unlock(&sc->sc_queue_mtx);
1628 		if (bp->bio_from == NULL ||
1629 		    bp->bio_from->geom != sc->sc_geom)
1630 			g_raid_start_request(bp);
1631 		else
1632 			g_raid_disk_done_request(bp);
1633 	}
1634 out:
1635 	sx_xunlock(&sc->sc_lock);
1636 }
1637 
1638 static void
1639 g_raid_launch_provider(struct g_raid_volume *vol)
1640 {
1641 	struct g_raid_disk *disk;
1642 	struct g_raid_softc *sc;
1643 	struct g_provider *pp;
1644 	char name[G_RAID_MAX_VOLUMENAME];
1645 	char   announce_buf[80], buf1[32];
1646 	off_t off;
1647 
1648 	sc = vol->v_softc;
1649 	sx_assert(&sc->sc_lock, SX_LOCKED);
1650 
1651 	g_topology_lock();
1652 	/* Try to name provider with volume name. */
1653 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1654 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1655 	    g_provider_by_name(name) != NULL) {
1656 		/* Otherwise use sequential volume number. */
1657 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1658 	}
1659 
1660 	/*
1661 	 * Create a /dev/ar%d that the old ataraid(4) stack once
1662 	 * created as an alias for /dev/raid/r%d if requested.
1663 	 * This helps going from stable/7 ataraid devices to newer
1664 	 * FreeBSD releases. sbruno 07 MAY 2013
1665 	 */
1666 
1667         if (ar_legacy_aliases) {
1668 		snprintf(announce_buf, sizeof(announce_buf),
1669                         "kern.devalias.%s", name);
1670                 snprintf(buf1, sizeof(buf1),
1671                         "ar%d", vol->v_global_id);
1672                 setenv(announce_buf, buf1);
1673         }
1674 
1675 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1676 	pp->private = vol;
1677 	pp->mediasize = vol->v_mediasize;
1678 	pp->sectorsize = vol->v_sectorsize;
1679 	pp->stripesize = 0;
1680 	pp->stripeoffset = 0;
1681 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1682 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1683 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1684 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1685 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1686 		    disk->d_consumer != NULL &&
1687 		    disk->d_consumer->provider != NULL) {
1688 			pp->stripesize = disk->d_consumer->provider->stripesize;
1689 			off = disk->d_consumer->provider->stripeoffset;
1690 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1691 			if (off > 0)
1692 				pp->stripeoffset %= off;
1693 		}
1694 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1695 			pp->stripesize *= (vol->v_disks_count - 1);
1696 			pp->stripeoffset *= (vol->v_disks_count - 1);
1697 		}
1698 	} else
1699 		pp->stripesize = vol->v_strip_size;
1700 	vol->v_provider = pp;
1701 	g_error_provider(pp, 0);
1702 	g_topology_unlock();
1703 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1704 	    pp->name, vol->v_name);
1705 }
1706 
1707 static void
1708 g_raid_destroy_provider(struct g_raid_volume *vol)
1709 {
1710 	struct g_raid_softc *sc;
1711 	struct g_provider *pp;
1712 	struct bio *bp, *tmp;
1713 
1714 	g_topology_assert_not();
1715 	sc = vol->v_softc;
1716 	pp = vol->v_provider;
1717 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1718 
1719 	g_topology_lock();
1720 	g_error_provider(pp, ENXIO);
1721 	mtx_lock(&sc->sc_queue_mtx);
1722 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1723 		if (bp->bio_to != pp)
1724 			continue;
1725 		bioq_remove(&sc->sc_queue, bp);
1726 		g_io_deliver(bp, ENXIO);
1727 	}
1728 	mtx_unlock(&sc->sc_queue_mtx);
1729 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1730 	    pp->name, vol->v_name);
1731 	g_wither_provider(pp, ENXIO);
1732 	g_topology_unlock();
1733 	vol->v_provider = NULL;
1734 }
1735 
1736 /*
1737  * Update device state.
1738  */
1739 static int
1740 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1741 {
1742 	struct g_raid_softc *sc;
1743 
1744 	sc = vol->v_softc;
1745 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1746 
1747 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1748 	    g_raid_volume_event2str(event),
1749 	    vol->v_name);
1750 	switch (event) {
1751 	case G_RAID_VOLUME_E_DOWN:
1752 		if (vol->v_provider != NULL)
1753 			g_raid_destroy_provider(vol);
1754 		break;
1755 	case G_RAID_VOLUME_E_UP:
1756 		if (vol->v_provider == NULL)
1757 			g_raid_launch_provider(vol);
1758 		break;
1759 	case G_RAID_VOLUME_E_START:
1760 		if (vol->v_tr)
1761 			G_RAID_TR_START(vol->v_tr);
1762 		return (0);
1763 	default:
1764 		if (sc->sc_md)
1765 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1766 		return (0);
1767 	}
1768 
1769 	/* Manage root mount release. */
1770 	if (vol->v_starting) {
1771 		vol->v_starting = 0;
1772 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1773 		root_mount_rel(vol->v_rootmount);
1774 		vol->v_rootmount = NULL;
1775 	}
1776 	if (vol->v_stopping && vol->v_provider_open == 0)
1777 		g_raid_destroy_volume(vol);
1778 	return (0);
1779 }
1780 
1781 /*
1782  * Update subdisk state.
1783  */
1784 static int
1785 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1786 {
1787 	struct g_raid_softc *sc;
1788 	struct g_raid_volume *vol;
1789 
1790 	sc = sd->sd_softc;
1791 	vol = sd->sd_volume;
1792 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1793 
1794 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1795 	    g_raid_subdisk_event2str(event),
1796 	    vol->v_name, sd->sd_pos,
1797 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1798 	if (vol->v_tr)
1799 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1800 
1801 	return (0);
1802 }
1803 
1804 /*
1805  * Update disk state.
1806  */
1807 static int
1808 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1809 {
1810 	struct g_raid_softc *sc;
1811 
1812 	sc = disk->d_softc;
1813 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1814 
1815 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1816 	    g_raid_disk_event2str(event),
1817 	    g_raid_get_diskname(disk));
1818 
1819 	if (sc->sc_md)
1820 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1821 	return (0);
1822 }
1823 
1824 /*
1825  * Node event.
1826  */
1827 static int
1828 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1829 {
1830 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1831 
1832 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1833 	    g_raid_node_event2str(event));
1834 
1835 	if (event == G_RAID_NODE_E_WAKE)
1836 		return (0);
1837 	if (sc->sc_md)
1838 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1839 	return (0);
1840 }
1841 
1842 static int
1843 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1844 {
1845 	struct g_raid_volume *vol;
1846 	struct g_raid_softc *sc;
1847 	int dcw, opens, error = 0;
1848 
1849 	g_topology_assert();
1850 	sc = pp->geom->softc;
1851 	vol = pp->private;
1852 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1853 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1854 
1855 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1856 	    acr, acw, ace);
1857 	dcw = pp->acw + acw;
1858 
1859 	g_topology_unlock();
1860 	sx_xlock(&sc->sc_lock);
1861 	/* Deny new opens while dying. */
1862 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1863 		error = ENXIO;
1864 		goto out;
1865 	}
1866 	/* Deny write opens for read-only volumes. */
1867 	if (vol->v_read_only && acw > 0) {
1868 		error = EROFS;
1869 		goto out;
1870 	}
1871 	if (dcw == 0)
1872 		g_raid_clean(vol, dcw);
1873 	vol->v_provider_open += acr + acw + ace;
1874 	/* Handle delayed node destruction. */
1875 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1876 	    vol->v_provider_open == 0) {
1877 		/* Count open volumes. */
1878 		opens = g_raid_nopens(sc);
1879 		if (opens == 0) {
1880 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1881 			/* Wake up worker to make it selfdestruct. */
1882 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1883 		}
1884 	}
1885 	/* Handle open volume destruction. */
1886 	if (vol->v_stopping && vol->v_provider_open == 0)
1887 		g_raid_destroy_volume(vol);
1888 out:
1889 	sx_xunlock(&sc->sc_lock);
1890 	g_topology_lock();
1891 	return (error);
1892 }
1893 
1894 struct g_raid_softc *
1895 g_raid_create_node(struct g_class *mp,
1896     const char *name, struct g_raid_md_object *md)
1897 {
1898 	struct g_raid_softc *sc;
1899 	struct g_geom *gp;
1900 	int error;
1901 
1902 	g_topology_assert();
1903 	G_RAID_DEBUG(1, "Creating array %s.", name);
1904 
1905 	gp = g_new_geomf(mp, "%s", name);
1906 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1907 	gp->start = g_raid_start;
1908 	gp->orphan = g_raid_orphan;
1909 	gp->access = g_raid_access;
1910 	gp->dumpconf = g_raid_dumpconf;
1911 
1912 	sc->sc_md = md;
1913 	sc->sc_geom = gp;
1914 	sc->sc_flags = 0;
1915 	TAILQ_INIT(&sc->sc_volumes);
1916 	TAILQ_INIT(&sc->sc_disks);
1917 	sx_init(&sc->sc_lock, "graid:lock");
1918 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1919 	TAILQ_INIT(&sc->sc_events);
1920 	bioq_init(&sc->sc_queue);
1921 	gp->softc = sc;
1922 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1923 	    "g_raid %s", name);
1924 	if (error != 0) {
1925 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1926 		mtx_destroy(&sc->sc_queue_mtx);
1927 		sx_destroy(&sc->sc_lock);
1928 		g_destroy_geom(sc->sc_geom);
1929 		free(sc, M_RAID);
1930 		return (NULL);
1931 	}
1932 
1933 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1934 	return (sc);
1935 }
1936 
1937 struct g_raid_volume *
1938 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1939 {
1940 	struct g_raid_volume	*vol, *vol1;
1941 	int i;
1942 
1943 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1944 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1945 	vol->v_softc = sc;
1946 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1947 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1948 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1949 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1950 	vol->v_rotate_parity = 1;
1951 	bioq_init(&vol->v_inflight);
1952 	bioq_init(&vol->v_locked);
1953 	LIST_INIT(&vol->v_locks);
1954 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1955 		vol->v_subdisks[i].sd_softc = sc;
1956 		vol->v_subdisks[i].sd_volume = vol;
1957 		vol->v_subdisks[i].sd_pos = i;
1958 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1959 	}
1960 
1961 	/* Find free ID for this volume. */
1962 	g_topology_lock();
1963 	vol1 = vol;
1964 	if (id >= 0) {
1965 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1966 			if (vol1->v_global_id == id)
1967 				break;
1968 		}
1969 	}
1970 	if (vol1 != NULL) {
1971 		for (id = 0; ; id++) {
1972 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1973 				if (vol1->v_global_id == id)
1974 					break;
1975 			}
1976 			if (vol1 == NULL)
1977 				break;
1978 		}
1979 	}
1980 	vol->v_global_id = id;
1981 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1982 	g_topology_unlock();
1983 
1984 	/* Delay root mounting. */
1985 	vol->v_rootmount = root_mount_hold("GRAID");
1986 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1987 	vol->v_starting = 1;
1988 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1989 	return (vol);
1990 }
1991 
1992 struct g_raid_disk *
1993 g_raid_create_disk(struct g_raid_softc *sc)
1994 {
1995 	struct g_raid_disk	*disk;
1996 
1997 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1998 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1999 	disk->d_softc = sc;
2000 	disk->d_state = G_RAID_DISK_S_NONE;
2001 	TAILQ_INIT(&disk->d_subdisks);
2002 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
2003 	return (disk);
2004 }
2005 
2006 int g_raid_start_volume(struct g_raid_volume *vol)
2007 {
2008 	struct g_raid_tr_class *class;
2009 	struct g_raid_tr_object *obj;
2010 	int status;
2011 
2012 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
2013 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
2014 		if (!class->trc_enable)
2015 			continue;
2016 		G_RAID_DEBUG1(2, vol->v_softc,
2017 		    "Tasting volume %s for %s transformation.",
2018 		    vol->v_name, class->name);
2019 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2020 		    M_WAITOK);
2021 		obj->tro_class = class;
2022 		obj->tro_volume = vol;
2023 		status = G_RAID_TR_TASTE(obj, vol);
2024 		if (status != G_RAID_TR_TASTE_FAIL)
2025 			break;
2026 		kobj_delete((kobj_t)obj, M_RAID);
2027 	}
2028 	if (class == NULL) {
2029 		G_RAID_DEBUG1(0, vol->v_softc,
2030 		    "No transformation module found for %s.",
2031 		    vol->v_name);
2032 		vol->v_tr = NULL;
2033 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2034 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2035 		    G_RAID_EVENT_VOLUME);
2036 		return (-1);
2037 	}
2038 	G_RAID_DEBUG1(2, vol->v_softc,
2039 	    "Transformation module %s chosen for %s.",
2040 	    class->name, vol->v_name);
2041 	vol->v_tr = obj;
2042 	return (0);
2043 }
2044 
2045 int
2046 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2047 {
2048 	struct g_raid_volume *vol, *tmpv;
2049 	struct g_raid_disk *disk, *tmpd;
2050 	int error = 0;
2051 
2052 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2053 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2054 		if (g_raid_destroy_volume(vol))
2055 			error = EBUSY;
2056 	}
2057 	if (error)
2058 		return (error);
2059 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2060 		if (g_raid_destroy_disk(disk))
2061 			error = EBUSY;
2062 	}
2063 	if (error)
2064 		return (error);
2065 	if (sc->sc_md) {
2066 		G_RAID_MD_FREE(sc->sc_md);
2067 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2068 		sc->sc_md = NULL;
2069 	}
2070 	if (sc->sc_geom != NULL) {
2071 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2072 		g_topology_lock();
2073 		sc->sc_geom->softc = NULL;
2074 		g_wither_geom(sc->sc_geom, ENXIO);
2075 		g_topology_unlock();
2076 		sc->sc_geom = NULL;
2077 	} else
2078 		G_RAID_DEBUG(1, "Array destroyed.");
2079 	if (worker) {
2080 		g_raid_event_cancel(sc, sc);
2081 		mtx_destroy(&sc->sc_queue_mtx);
2082 		sx_xunlock(&sc->sc_lock);
2083 		sx_destroy(&sc->sc_lock);
2084 		wakeup(&sc->sc_stopping);
2085 		free(sc, M_RAID);
2086 		curthread->td_pflags &= ~TDP_GEOM;
2087 		G_RAID_DEBUG(1, "Thread exiting.");
2088 		kproc_exit(0);
2089 	} else {
2090 		/* Wake up worker to make it selfdestruct. */
2091 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2092 	}
2093 	return (0);
2094 }
2095 
2096 int
2097 g_raid_destroy_volume(struct g_raid_volume *vol)
2098 {
2099 	struct g_raid_softc *sc;
2100 	struct g_raid_disk *disk;
2101 	int i;
2102 
2103 	sc = vol->v_softc;
2104 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2105 	vol->v_stopping = 1;
2106 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2107 		if (vol->v_tr) {
2108 			G_RAID_TR_STOP(vol->v_tr);
2109 			return (EBUSY);
2110 		} else
2111 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2112 	}
2113 	if (g_raid_event_check(sc, vol) != 0)
2114 		return (EBUSY);
2115 	if (vol->v_provider != NULL)
2116 		return (EBUSY);
2117 	if (vol->v_provider_open != 0)
2118 		return (EBUSY);
2119 	if (vol->v_tr) {
2120 		G_RAID_TR_FREE(vol->v_tr);
2121 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2122 		vol->v_tr = NULL;
2123 	}
2124 	if (vol->v_rootmount)
2125 		root_mount_rel(vol->v_rootmount);
2126 	g_topology_lock();
2127 	LIST_REMOVE(vol, v_global_next);
2128 	g_topology_unlock();
2129 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2130 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2131 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2132 		disk = vol->v_subdisks[i].sd_disk;
2133 		if (disk == NULL)
2134 			continue;
2135 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2136 	}
2137 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2138 	if (sc->sc_md)
2139 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2140 	g_raid_event_cancel(sc, vol);
2141 	free(vol, M_RAID);
2142 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2143 		/* Wake up worker to let it selfdestruct. */
2144 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2145 	}
2146 	return (0);
2147 }
2148 
2149 int
2150 g_raid_destroy_disk(struct g_raid_disk *disk)
2151 {
2152 	struct g_raid_softc *sc;
2153 	struct g_raid_subdisk *sd, *tmp;
2154 
2155 	sc = disk->d_softc;
2156 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2157 	if (disk->d_consumer) {
2158 		g_raid_kill_consumer(sc, disk->d_consumer);
2159 		disk->d_consumer = NULL;
2160 	}
2161 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2162 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2163 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2164 		    G_RAID_EVENT_SUBDISK);
2165 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2166 		sd->sd_disk = NULL;
2167 	}
2168 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2169 	if (sc->sc_md)
2170 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2171 	g_raid_event_cancel(sc, disk);
2172 	free(disk, M_RAID);
2173 	return (0);
2174 }
2175 
2176 int
2177 g_raid_destroy(struct g_raid_softc *sc, int how)
2178 {
2179 	int error, opens;
2180 
2181 	g_topology_assert_not();
2182 	if (sc == NULL)
2183 		return (ENXIO);
2184 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2185 
2186 	/* Count open volumes. */
2187 	opens = g_raid_nopens(sc);
2188 
2189 	/* React on some opened volumes. */
2190 	if (opens > 0) {
2191 		switch (how) {
2192 		case G_RAID_DESTROY_SOFT:
2193 			G_RAID_DEBUG1(1, sc,
2194 			    "%d volumes are still open.",
2195 			    opens);
2196 			sx_xunlock(&sc->sc_lock);
2197 			return (EBUSY);
2198 		case G_RAID_DESTROY_DELAYED:
2199 			G_RAID_DEBUG1(1, sc,
2200 			    "Array will be destroyed on last close.");
2201 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2202 			sx_xunlock(&sc->sc_lock);
2203 			return (EBUSY);
2204 		case G_RAID_DESTROY_HARD:
2205 			G_RAID_DEBUG1(1, sc,
2206 			    "%d volumes are still open.",
2207 			    opens);
2208 		}
2209 	}
2210 
2211 	/* Mark node for destruction. */
2212 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2213 	/* Wake up worker to let it selfdestruct. */
2214 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2215 	/* Sleep until node destroyed. */
2216 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2217 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2218 	return (error == EWOULDBLOCK ? EBUSY : 0);
2219 }
2220 
2221 static void
2222 g_raid_taste_orphan(struct g_consumer *cp)
2223 {
2224 
2225 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2226 	    cp->provider->name));
2227 }
2228 
2229 static struct g_geom *
2230 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2231 {
2232 	struct g_consumer *cp;
2233 	struct g_geom *gp, *geom;
2234 	struct g_raid_md_class *class;
2235 	struct g_raid_md_object *obj;
2236 	int status;
2237 
2238 	g_topology_assert();
2239 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2240 	if (!g_raid_enable)
2241 		return (NULL);
2242 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2243 
2244 	gp = g_new_geomf(mp, "raid:taste");
2245 	/*
2246 	 * This orphan function should be never called.
2247 	 */
2248 	gp->orphan = g_raid_taste_orphan;
2249 	cp = g_new_consumer(gp);
2250 	g_attach(cp, pp);
2251 
2252 	geom = NULL;
2253 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2254 		if (!class->mdc_enable)
2255 			continue;
2256 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2257 		    pp->name, class->name);
2258 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2259 		    M_WAITOK);
2260 		obj->mdo_class = class;
2261 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2262 		if (status != G_RAID_MD_TASTE_NEW)
2263 			kobj_delete((kobj_t)obj, M_RAID);
2264 		if (status != G_RAID_MD_TASTE_FAIL)
2265 			break;
2266 	}
2267 
2268 	g_detach(cp);
2269 	g_destroy_consumer(cp);
2270 	g_destroy_geom(gp);
2271 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2272 	return (geom);
2273 }
2274 
2275 int
2276 g_raid_create_node_format(const char *format, struct gctl_req *req,
2277     struct g_geom **gp)
2278 {
2279 	struct g_raid_md_class *class;
2280 	struct g_raid_md_object *obj;
2281 	int status;
2282 
2283 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2284 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2285 		if (strcasecmp(class->name, format) == 0)
2286 			break;
2287 	}
2288 	if (class == NULL) {
2289 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2290 		return (G_RAID_MD_TASTE_FAIL);
2291 	}
2292 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2293 	    M_WAITOK);
2294 	obj->mdo_class = class;
2295 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2296 	if (status != G_RAID_MD_TASTE_NEW)
2297 		kobj_delete((kobj_t)obj, M_RAID);
2298 	return (status);
2299 }
2300 
2301 static int
2302 g_raid_destroy_geom(struct gctl_req *req __unused,
2303     struct g_class *mp __unused, struct g_geom *gp)
2304 {
2305 	struct g_raid_softc *sc;
2306 	int error;
2307 
2308 	g_topology_unlock();
2309 	sc = gp->softc;
2310 	sx_xlock(&sc->sc_lock);
2311 	g_cancel_event(sc);
2312 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2313 	g_topology_lock();
2314 	return (error);
2315 }
2316 
2317 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2318     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2319 {
2320 
2321 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2322 		return;
2323 	if (sc->sc_md)
2324 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2325 }
2326 
2327 void g_raid_fail_disk(struct g_raid_softc *sc,
2328     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2329 {
2330 
2331 	if (disk == NULL)
2332 		disk = sd->sd_disk;
2333 	if (disk == NULL) {
2334 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2335 		return;
2336 	}
2337 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2338 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2339 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2340 		return;
2341 	}
2342 	if (sc->sc_md)
2343 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2344 }
2345 
2346 static void
2347 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2348     struct g_consumer *cp, struct g_provider *pp)
2349 {
2350 	struct g_raid_softc *sc;
2351 	struct g_raid_volume *vol;
2352 	struct g_raid_subdisk *sd;
2353 	struct g_raid_disk *disk;
2354 	int i, s;
2355 
2356 	g_topology_assert();
2357 
2358 	sc = gp->softc;
2359 	if (sc == NULL)
2360 		return;
2361 	if (pp != NULL) {
2362 		vol = pp->private;
2363 		g_topology_unlock();
2364 		sx_xlock(&sc->sc_lock);
2365 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2366 		    sc->sc_md->mdo_class->name,
2367 		    g_raid_volume_level2str(vol->v_raid_level,
2368 		    vol->v_raid_level_qualifier));
2369 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2370 		    vol->v_name);
2371 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2372 		    g_raid_volume_level2str(vol->v_raid_level,
2373 		    vol->v_raid_level_qualifier));
2374 		sbuf_printf(sb,
2375 		    "%s<Transformation>%s</Transformation>\n", indent,
2376 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2377 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2378 		    vol->v_disks_count);
2379 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2380 		    vol->v_strip_size);
2381 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2382 		    g_raid_volume_state2str(vol->v_state));
2383 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2384 		    vol->v_dirty ? "Yes" : "No");
2385 		sbuf_printf(sb, "%s<Subdisks>", indent);
2386 		for (i = 0; i < vol->v_disks_count; i++) {
2387 			sd = &vol->v_subdisks[i];
2388 			if (sd->sd_disk != NULL &&
2389 			    sd->sd_disk->d_consumer != NULL) {
2390 				sbuf_printf(sb, "%s ",
2391 				    g_raid_get_diskname(sd->sd_disk));
2392 			} else {
2393 				sbuf_printf(sb, "NONE ");
2394 			}
2395 			sbuf_printf(sb, "(%s",
2396 			    g_raid_subdisk_state2str(sd->sd_state));
2397 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2398 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2399 				sbuf_printf(sb, " %d%%",
2400 				    (int)(sd->sd_rebuild_pos * 100 /
2401 				     sd->sd_size));
2402 			}
2403 			sbuf_printf(sb, ")");
2404 			if (i + 1 < vol->v_disks_count)
2405 				sbuf_printf(sb, ", ");
2406 		}
2407 		sbuf_printf(sb, "</Subdisks>\n");
2408 		sx_xunlock(&sc->sc_lock);
2409 		g_topology_lock();
2410 	} else if (cp != NULL) {
2411 		disk = cp->private;
2412 		if (disk == NULL)
2413 			return;
2414 		g_topology_unlock();
2415 		sx_xlock(&sc->sc_lock);
2416 		sbuf_printf(sb, "%s<State>%s", indent,
2417 		    g_raid_disk_state2str(disk->d_state));
2418 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2419 			sbuf_printf(sb, " (");
2420 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2421 				sbuf_printf(sb, "%s",
2422 				    g_raid_subdisk_state2str(sd->sd_state));
2423 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2424 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2425 					sbuf_printf(sb, " %d%%",
2426 					    (int)(sd->sd_rebuild_pos * 100 /
2427 					     sd->sd_size));
2428 				}
2429 				if (TAILQ_NEXT(sd, sd_next))
2430 					sbuf_printf(sb, ", ");
2431 			}
2432 			sbuf_printf(sb, ")");
2433 		}
2434 		sbuf_printf(sb, "</State>\n");
2435 		sbuf_printf(sb, "%s<Subdisks>", indent);
2436 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2437 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2438 			    sd->sd_volume->v_global_id,
2439 			    sd->sd_volume->v_name,
2440 			    sd->sd_pos, sd->sd_offset);
2441 			if (TAILQ_NEXT(sd, sd_next))
2442 				sbuf_printf(sb, ", ");
2443 		}
2444 		sbuf_printf(sb, "</Subdisks>\n");
2445 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2446 		    disk->d_read_errs);
2447 		sx_xunlock(&sc->sc_lock);
2448 		g_topology_lock();
2449 	} else {
2450 		g_topology_unlock();
2451 		sx_xlock(&sc->sc_lock);
2452 		if (sc->sc_md) {
2453 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2454 			    sc->sc_md->mdo_class->name);
2455 		}
2456 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2457 			s = 0xff;
2458 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2459 				if (vol->v_state < s)
2460 					s = vol->v_state;
2461 			}
2462 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2463 			    g_raid_volume_state2str(s));
2464 		}
2465 		sx_xunlock(&sc->sc_lock);
2466 		g_topology_lock();
2467 	}
2468 }
2469 
2470 static void
2471 g_raid_shutdown_post_sync(void *arg, int howto)
2472 {
2473 	struct g_class *mp;
2474 	struct g_geom *gp, *gp2;
2475 	struct g_raid_softc *sc;
2476 	struct g_raid_volume *vol;
2477 
2478 	mp = arg;
2479 	DROP_GIANT();
2480 	g_topology_lock();
2481 	g_raid_shutdown = 1;
2482 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2483 		if ((sc = gp->softc) == NULL)
2484 			continue;
2485 		g_topology_unlock();
2486 		sx_xlock(&sc->sc_lock);
2487 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2488 			g_raid_clean(vol, -1);
2489 		g_cancel_event(sc);
2490 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2491 		g_topology_lock();
2492 	}
2493 	g_topology_unlock();
2494 	PICKUP_GIANT();
2495 }
2496 
2497 static void
2498 g_raid_init(struct g_class *mp)
2499 {
2500 
2501 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2502 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2503 	if (g_raid_post_sync == NULL)
2504 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2505 	g_raid_started = 1;
2506 }
2507 
2508 static void
2509 g_raid_fini(struct g_class *mp)
2510 {
2511 
2512 	if (g_raid_post_sync != NULL)
2513 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2514 	g_raid_started = 0;
2515 }
2516 
2517 int
2518 g_raid_md_modevent(module_t mod, int type, void *arg)
2519 {
2520 	struct g_raid_md_class *class, *c, *nc;
2521 	int error;
2522 
2523 	error = 0;
2524 	class = arg;
2525 	switch (type) {
2526 	case MOD_LOAD:
2527 		c = LIST_FIRST(&g_raid_md_classes);
2528 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2529 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2530 		else {
2531 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2532 			    nc->mdc_priority < class->mdc_priority)
2533 				c = nc;
2534 			LIST_INSERT_AFTER(c, class, mdc_list);
2535 		}
2536 		if (g_raid_started)
2537 			g_retaste(&g_raid_class);
2538 		break;
2539 	case MOD_UNLOAD:
2540 		LIST_REMOVE(class, mdc_list);
2541 		break;
2542 	default:
2543 		error = EOPNOTSUPP;
2544 		break;
2545 	}
2546 
2547 	return (error);
2548 }
2549 
2550 int
2551 g_raid_tr_modevent(module_t mod, int type, void *arg)
2552 {
2553 	struct g_raid_tr_class *class, *c, *nc;
2554 	int error;
2555 
2556 	error = 0;
2557 	class = arg;
2558 	switch (type) {
2559 	case MOD_LOAD:
2560 		c = LIST_FIRST(&g_raid_tr_classes);
2561 		if (c == NULL || c->trc_priority > class->trc_priority)
2562 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2563 		else {
2564 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2565 			    nc->trc_priority < class->trc_priority)
2566 				c = nc;
2567 			LIST_INSERT_AFTER(c, class, trc_list);
2568 		}
2569 		break;
2570 	case MOD_UNLOAD:
2571 		LIST_REMOVE(class, trc_list);
2572 		break;
2573 	default:
2574 		error = EOPNOTSUPP;
2575 		break;
2576 	}
2577 
2578 	return (error);
2579 }
2580 
2581 /*
2582  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2583  * to reduce module priority, allowing submodules to register them first.
2584  */
2585 static moduledata_t g_raid_mod = {
2586 	"g_raid",
2587 	g_modevent,
2588 	&g_raid_class
2589 };
2590 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2591 MODULE_VERSION(geom_raid, 0);
2592