xref: /freebsd/sys/geom/raid/g_raid.c (revision 2e3f49888ec8851bafb22011533217487764fdb0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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/param.h>
30 #include <sys/systm.h>
31 #include <sys/bio.h>
32 #include <sys/eventhandler.h>
33 #include <sys/kernel.h>
34 #include <sys/kthread.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/reboot.h>
42 #include <sys/sbuf.h>
43 #include <sys/sched.h>
44 #include <sys/sysctl.h>
45 
46 #include <vm/uma.h>
47 
48 #include <geom/geom.h>
49 #include <geom/geom_dbg.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, void *virtual, off_t offset, size_t length)
1037 {
1038 	struct g_raid_volume *vol;
1039 	int error;
1040 
1041 	vol = (struct g_raid_volume *)arg;
1042 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1043 	    (long long unsigned)offset, (long long unsigned)length);
1044 
1045 	error = G_RAID_TR_KERNELDUMP(vol->v_tr, virtual, offset, length);
1046 	return (error);
1047 }
1048 
1049 static void
1050 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1051 {
1052 	struct g_kerneldump *gkd;
1053 	struct g_provider *pp;
1054 	struct g_raid_volume *vol;
1055 
1056 	gkd = (struct g_kerneldump*)bp->bio_data;
1057 	pp = bp->bio_to;
1058 	vol = pp->private;
1059 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1060 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1061 	gkd->di.dumper = g_raid_dump;
1062 	gkd->di.priv = vol;
1063 	gkd->di.blocksize = vol->v_sectorsize;
1064 	gkd->di.maxiosize = DFLTPHYS;
1065 	gkd->di.mediaoffset = gkd->offset;
1066 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1067 		gkd->length = vol->v_mediasize - gkd->offset;
1068 	gkd->di.mediasize = gkd->length;
1069 	g_io_deliver(bp, 0);
1070 }
1071 
1072 static void
1073 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1074 {
1075 	struct g_provider *pp;
1076 	struct g_raid_volume *vol;
1077 	struct g_raid_subdisk *sd;
1078 	int i, val;
1079 
1080 	pp = bp->bio_to;
1081 	vol = pp->private;
1082 	for (i = 0; i < vol->v_disks_count; i++) {
1083 		sd = &vol->v_subdisks[i];
1084 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1085 			continue;
1086 		if (sd->sd_disk->d_candelete)
1087 			break;
1088 	}
1089 	val = i < vol->v_disks_count;
1090 	g_handleattr(bp, "GEOM::candelete", &val, sizeof(val));
1091 }
1092 
1093 static void
1094 g_raid_start(struct bio *bp)
1095 {
1096 	struct g_raid_softc *sc;
1097 
1098 	sc = bp->bio_to->geom->softc;
1099 	/*
1100 	 * If sc == NULL or there are no valid disks, provider's error
1101 	 * should be set and g_raid_start() should not be called at all.
1102 	 */
1103 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1104 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1105 //	    bp->bio_to->error, bp->bio_to->name));
1106 	G_RAID_LOGREQ(3, bp, "Request received.");
1107 
1108 	switch (bp->bio_cmd) {
1109 	case BIO_READ:
1110 	case BIO_WRITE:
1111 	case BIO_DELETE:
1112 	case BIO_FLUSH:
1113 	case BIO_SPEEDUP:
1114 		break;
1115 	case BIO_GETATTR:
1116 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1117 			g_raid_candelete(sc, bp);
1118 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1119 			g_raid_kerneldump(sc, bp);
1120 		else
1121 			g_io_deliver(bp, EOPNOTSUPP);
1122 		return;
1123 	default:
1124 		g_io_deliver(bp, EOPNOTSUPP);
1125 		return;
1126 	}
1127 	mtx_lock(&sc->sc_queue_mtx);
1128 	bioq_insert_tail(&sc->sc_queue, bp);
1129 	mtx_unlock(&sc->sc_queue_mtx);
1130 	if (!dumping) {
1131 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1132 		wakeup(sc);
1133 	}
1134 }
1135 
1136 static int
1137 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1138 {
1139 	/*
1140 	 * 5 cases:
1141 	 * (1) bp entirely below NO
1142 	 * (2) bp entirely above NO
1143 	 * (3) bp start below, but end in range YES
1144 	 * (4) bp entirely within YES
1145 	 * (5) bp starts within, ends above YES
1146 	 *
1147 	 * lock range 10-19 (offset 10 length 10)
1148 	 * (1) 1-5: first if kicks it out
1149 	 * (2) 30-35: second if kicks it out
1150 	 * (3) 5-15: passes both ifs
1151 	 * (4) 12-14: passes both ifs
1152 	 * (5) 19-20: passes both
1153 	 */
1154 	off_t lend = lstart + len - 1;
1155 	off_t bstart = bp->bio_offset;
1156 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1157 
1158 	if (bend < lstart)
1159 		return (0);
1160 	if (lend < bstart)
1161 		return (0);
1162 	return (1);
1163 }
1164 
1165 static int
1166 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1167 {
1168 	struct g_raid_lock *lp;
1169 
1170 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1171 
1172 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1173 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1174 			return (1);
1175 	}
1176 	return (0);
1177 }
1178 
1179 static void
1180 g_raid_start_request(struct bio *bp)
1181 {
1182 	struct g_raid_softc *sc __diagused;
1183 	struct g_raid_volume *vol;
1184 
1185 	sc = bp->bio_to->geom->softc;
1186 	sx_assert(&sc->sc_lock, SX_LOCKED);
1187 	vol = bp->bio_to->private;
1188 
1189 	/*
1190 	 * Check to see if this item is in a locked range.  If so,
1191 	 * queue it to our locked queue and return.  We'll requeue
1192 	 * it when the range is unlocked.  Internal I/O for the
1193 	 * rebuild/rescan/recovery process is excluded from this
1194 	 * check so we can actually do the recovery.
1195 	 */
1196 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1197 	    g_raid_is_in_locked_range(vol, bp)) {
1198 		G_RAID_LOGREQ(3, bp, "Defer request.");
1199 		bioq_insert_tail(&vol->v_locked, bp);
1200 		return;
1201 	}
1202 
1203 	/*
1204 	 * If we're actually going to do the write/delete, then
1205 	 * update the idle stats for the volume.
1206 	 */
1207 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1208 		if (!vol->v_dirty)
1209 			g_raid_dirty(vol);
1210 		vol->v_writes++;
1211 	}
1212 
1213 	/*
1214 	 * Put request onto inflight queue, so we can check if new
1215 	 * synchronization requests don't collide with it.  Then tell
1216 	 * the transformation layer to start the I/O.
1217 	 */
1218 	bioq_insert_tail(&vol->v_inflight, bp);
1219 	G_RAID_LOGREQ(4, bp, "Request started");
1220 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1221 }
1222 
1223 static void
1224 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1225 {
1226 	off_t off, len;
1227 	struct bio *nbp;
1228 	struct g_raid_lock *lp;
1229 
1230 	vol->v_pending_lock = 0;
1231 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1232 		if (lp->l_pending) {
1233 			off = lp->l_offset;
1234 			len = lp->l_length;
1235 			lp->l_pending = 0;
1236 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1237 				if (g_raid_bio_overlaps(nbp, off, len))
1238 					lp->l_pending++;
1239 			}
1240 			if (lp->l_pending) {
1241 				vol->v_pending_lock = 1;
1242 				G_RAID_DEBUG1(4, vol->v_softc,
1243 				    "Deferred lock(%jd, %jd) has %d pending",
1244 				    (intmax_t)off, (intmax_t)(off + len),
1245 				    lp->l_pending);
1246 				continue;
1247 			}
1248 			G_RAID_DEBUG1(4, vol->v_softc,
1249 			    "Deferred lock of %jd to %jd completed",
1250 			    (intmax_t)off, (intmax_t)(off + len));
1251 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1252 		}
1253 	}
1254 }
1255 
1256 void
1257 g_raid_iodone(struct bio *bp, int error)
1258 {
1259 	struct g_raid_softc *sc __diagused;
1260 	struct g_raid_volume *vol;
1261 
1262 	sc = bp->bio_to->geom->softc;
1263 	sx_assert(&sc->sc_lock, SX_LOCKED);
1264 	vol = bp->bio_to->private;
1265 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1266 
1267 	/* Update stats if we done write/delete. */
1268 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1269 		vol->v_writes--;
1270 		vol->v_last_write = time_uptime;
1271 	}
1272 
1273 	bioq_remove(&vol->v_inflight, bp);
1274 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1275 		g_raid_finish_with_locked_ranges(vol, bp);
1276 	getmicrouptime(&vol->v_last_done);
1277 	g_io_deliver(bp, error);
1278 }
1279 
1280 int
1281 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1282     struct bio *ignore, void *argp)
1283 {
1284 	struct g_raid_softc *sc;
1285 	struct g_raid_lock *lp;
1286 	struct bio *bp;
1287 
1288 	sc = vol->v_softc;
1289 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1290 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1291 	lp->l_offset = off;
1292 	lp->l_length = len;
1293 	lp->l_callback_arg = argp;
1294 
1295 	lp->l_pending = 0;
1296 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1297 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1298 			lp->l_pending++;
1299 	}
1300 
1301 	/*
1302 	 * If there are any writes that are pending, we return EBUSY.  All
1303 	 * callers will have to wait until all pending writes clear.
1304 	 */
1305 	if (lp->l_pending > 0) {
1306 		vol->v_pending_lock = 1;
1307 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1308 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1309 		return (EBUSY);
1310 	}
1311 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1312 	    (intmax_t)off, (intmax_t)(off+len));
1313 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1314 	return (0);
1315 }
1316 
1317 int
1318 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1319 {
1320 	struct g_raid_lock *lp;
1321 	struct g_raid_softc *sc;
1322 	struct bio *bp;
1323 
1324 	sc = vol->v_softc;
1325 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1326 		if (lp->l_offset == off && lp->l_length == len) {
1327 			LIST_REMOVE(lp, l_next);
1328 			/* XXX
1329 			 * Right now we just put them all back on the queue
1330 			 * and hope for the best.  We hope this because any
1331 			 * locked ranges will go right back on this list
1332 			 * when the worker thread runs.
1333 			 * XXX
1334 			 */
1335 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1336 			    (intmax_t)lp->l_offset,
1337 			    (intmax_t)(lp->l_offset+lp->l_length));
1338 			mtx_lock(&sc->sc_queue_mtx);
1339 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1340 				bioq_insert_tail(&sc->sc_queue, bp);
1341 			mtx_unlock(&sc->sc_queue_mtx);
1342 			free(lp, M_RAID);
1343 			return (0);
1344 		}
1345 	}
1346 	return (EINVAL);
1347 }
1348 
1349 void
1350 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1351 {
1352 	struct g_consumer *cp;
1353 	struct g_raid_disk *disk, *tdisk;
1354 
1355 	bp->bio_caller1 = sd;
1356 
1357 	/*
1358 	 * Make sure that the disk is present. Generally it is a task of
1359 	 * transformation layers to not send requests to absent disks, but
1360 	 * it is better to be safe and report situation then sorry.
1361 	 */
1362 	if (sd->sd_disk == NULL) {
1363 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1364 nodisk:
1365 		bp->bio_from = NULL;
1366 		bp->bio_to = NULL;
1367 		bp->bio_error = ENXIO;
1368 		g_raid_disk_done(bp);
1369 		return;
1370 	}
1371 	disk = sd->sd_disk;
1372 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1373 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1374 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1375 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1376 		goto nodisk;
1377 	}
1378 
1379 	cp = disk->d_consumer;
1380 	bp->bio_from = cp;
1381 	bp->bio_to = cp->provider;
1382 	cp->index++;
1383 
1384 	/* Update average disks load. */
1385 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1386 		if (tdisk->d_consumer == NULL)
1387 			tdisk->d_load = 0;
1388 		else
1389 			tdisk->d_load = (tdisk->d_consumer->index *
1390 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1391 	}
1392 
1393 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1394 	if (dumping) {
1395 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1396 		if (bp->bio_cmd == BIO_WRITE) {
1397 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1398 			    bp->bio_data, bp->bio_offset, bp->bio_length);
1399 		} else
1400 			bp->bio_error = EOPNOTSUPP;
1401 		g_raid_disk_done(bp);
1402 	} else {
1403 		bp->bio_done = g_raid_disk_done;
1404 		bp->bio_offset += sd->sd_offset;
1405 		G_RAID_LOGREQ(3, bp, "Sending request.");
1406 		g_io_request(bp, cp);
1407 	}
1408 }
1409 
1410 int
1411 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, void *virtual,
1412     off_t offset, size_t length)
1413 {
1414 
1415 	if (sd->sd_disk == NULL)
1416 		return (ENXIO);
1417 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1418 		return (EOPNOTSUPP);
1419 	return (dump_write(&sd->sd_disk->d_kd.di, virtual,
1420 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, length));
1421 }
1422 
1423 static void
1424 g_raid_disk_done(struct bio *bp)
1425 {
1426 	struct g_raid_softc *sc;
1427 	struct g_raid_subdisk *sd;
1428 
1429 	sd = bp->bio_caller1;
1430 	sc = sd->sd_softc;
1431 	mtx_lock(&sc->sc_queue_mtx);
1432 	bioq_insert_tail(&sc->sc_queue, bp);
1433 	mtx_unlock(&sc->sc_queue_mtx);
1434 	if (!dumping)
1435 		wakeup(sc);
1436 }
1437 
1438 static void
1439 g_raid_disk_done_request(struct bio *bp)
1440 {
1441 	struct g_raid_softc *sc;
1442 	struct g_raid_disk *disk;
1443 	struct g_raid_subdisk *sd;
1444 	struct g_raid_volume *vol;
1445 
1446 	g_topology_assert_not();
1447 
1448 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1449 	sd = bp->bio_caller1;
1450 	sc = sd->sd_softc;
1451 	vol = sd->sd_volume;
1452 	if (bp->bio_from != NULL) {
1453 		bp->bio_from->index--;
1454 		disk = bp->bio_from->private;
1455 		if (disk == NULL)
1456 			g_raid_kill_consumer(sc, bp->bio_from);
1457 	}
1458 	bp->bio_offset -= sd->sd_offset;
1459 
1460 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1461 }
1462 
1463 static void
1464 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1465 {
1466 
1467 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1468 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1469 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1470 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1471 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1472 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1473 	else
1474 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1475 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1476 		KASSERT(ep->e_error == 0,
1477 		    ("Error cannot be handled."));
1478 		g_raid_event_free(ep);
1479 	} else {
1480 		ep->e_flags |= G_RAID_EVENT_DONE;
1481 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1482 		mtx_lock(&sc->sc_queue_mtx);
1483 		wakeup(ep);
1484 		mtx_unlock(&sc->sc_queue_mtx);
1485 	}
1486 }
1487 
1488 /*
1489  * Worker thread.
1490  */
1491 static void
1492 g_raid_worker(void *arg)
1493 {
1494 	struct g_raid_softc *sc;
1495 	struct g_raid_event *ep;
1496 	struct g_raid_volume *vol;
1497 	struct bio *bp;
1498 	struct timeval now, t;
1499 	int timeout, rv;
1500 
1501 	sc = arg;
1502 	thread_lock(curthread);
1503 	sched_prio(curthread, PRIBIO);
1504 	thread_unlock(curthread);
1505 
1506 	sx_xlock(&sc->sc_lock);
1507 	for (;;) {
1508 		mtx_lock(&sc->sc_queue_mtx);
1509 		/*
1510 		 * First take a look at events.
1511 		 * This is important to handle events before any I/O requests.
1512 		 */
1513 		bp = NULL;
1514 		vol = NULL;
1515 		rv = 0;
1516 		ep = TAILQ_FIRST(&sc->sc_events);
1517 		if (ep != NULL)
1518 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1519 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1520 			;
1521 		else {
1522 			getmicrouptime(&now);
1523 			t = now;
1524 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1525 				if (bioq_first(&vol->v_inflight) == NULL &&
1526 				    vol->v_tr &&
1527 				    timevalcmp(&vol->v_last_done, &t, < ))
1528 					t = vol->v_last_done;
1529 			}
1530 			timevalsub(&t, &now);
1531 			timeout = g_raid_idle_threshold +
1532 			    t.tv_sec * 1000000 + t.tv_usec;
1533 			if (timeout > 0) {
1534 				/*
1535 				 * Two steps to avoid overflows at HZ=1000
1536 				 * and idle timeouts > 2.1s.  Some rounding
1537 				 * errors can occur, but they are < 1tick,
1538 				 * which is deemed to be close enough for
1539 				 * this purpose.
1540 				 */
1541 				int micpertic = 1000000 / hz;
1542 				timeout = (timeout + micpertic - 1) / micpertic;
1543 				sx_xunlock(&sc->sc_lock);
1544 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1545 				    PRIBIO | PDROP, "-", timeout);
1546 				sx_xlock(&sc->sc_lock);
1547 				goto process;
1548 			} else
1549 				rv = EWOULDBLOCK;
1550 		}
1551 		mtx_unlock(&sc->sc_queue_mtx);
1552 process:
1553 		if (ep != NULL) {
1554 			g_raid_handle_event(sc, ep);
1555 		} else if (bp != NULL) {
1556 			if (bp->bio_to != NULL &&
1557 			    bp->bio_to->geom == sc->sc_geom)
1558 				g_raid_start_request(bp);
1559 			else
1560 				g_raid_disk_done_request(bp);
1561 		} else if (rv == EWOULDBLOCK) {
1562 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1563 				g_raid_clean(vol, -1);
1564 				if (bioq_first(&vol->v_inflight) == NULL &&
1565 				    vol->v_tr) {
1566 					t.tv_sec = g_raid_idle_threshold / 1000000;
1567 					t.tv_usec = g_raid_idle_threshold % 1000000;
1568 					timevaladd(&t, &vol->v_last_done);
1569 					getmicrouptime(&now);
1570 					if (timevalcmp(&t, &now, <= )) {
1571 						G_RAID_TR_IDLE(vol->v_tr);
1572 						vol->v_last_done = now;
1573 					}
1574 				}
1575 			}
1576 		}
1577 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1578 			g_raid_destroy_node(sc, 1);	/* May not return. */
1579 	}
1580 }
1581 
1582 static void
1583 g_raid_poll(struct g_raid_softc *sc)
1584 {
1585 	struct g_raid_event *ep;
1586 	struct bio *bp;
1587 
1588 	sx_xlock(&sc->sc_lock);
1589 	mtx_lock(&sc->sc_queue_mtx);
1590 	/*
1591 	 * First take a look at events.
1592 	 * This is important to handle events before any I/O requests.
1593 	 */
1594 	ep = TAILQ_FIRST(&sc->sc_events);
1595 	if (ep != NULL) {
1596 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1597 		mtx_unlock(&sc->sc_queue_mtx);
1598 		g_raid_handle_event(sc, ep);
1599 		goto out;
1600 	}
1601 	bp = bioq_takefirst(&sc->sc_queue);
1602 	if (bp != NULL) {
1603 		mtx_unlock(&sc->sc_queue_mtx);
1604 		if (bp->bio_from == NULL ||
1605 		    bp->bio_from->geom != sc->sc_geom)
1606 			g_raid_start_request(bp);
1607 		else
1608 			g_raid_disk_done_request(bp);
1609 	}
1610 out:
1611 	sx_xunlock(&sc->sc_lock);
1612 }
1613 
1614 static void
1615 g_raid_launch_provider(struct g_raid_volume *vol)
1616 {
1617 	struct g_raid_disk *disk;
1618 	struct g_raid_subdisk *sd;
1619 	struct g_raid_softc *sc;
1620 	struct g_provider *pp;
1621 	char name[G_RAID_MAX_VOLUMENAME];
1622 	off_t off;
1623 	int i;
1624 
1625 	sc = vol->v_softc;
1626 	sx_assert(&sc->sc_lock, SX_LOCKED);
1627 
1628 	g_topology_lock();
1629 	/* Try to name provider with volume name. */
1630 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1631 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1632 	    g_provider_by_name(name) != NULL) {
1633 		/* Otherwise use sequential volume number. */
1634 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1635 	}
1636 
1637 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1638 	pp->flags |= G_PF_DIRECT_RECEIVE;
1639 	if (vol->v_tr->tro_class->trc_accept_unmapped) {
1640 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1641 		for (i = 0; i < vol->v_disks_count; i++) {
1642 			sd = &vol->v_subdisks[i];
1643 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1644 				continue;
1645 			if ((sd->sd_disk->d_consumer->provider->flags &
1646 			    G_PF_ACCEPT_UNMAPPED) == 0)
1647 				pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
1648 		}
1649 	}
1650 	pp->private = vol;
1651 	pp->mediasize = vol->v_mediasize;
1652 	pp->sectorsize = vol->v_sectorsize;
1653 	pp->stripesize = 0;
1654 	pp->stripeoffset = 0;
1655 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1656 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1657 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1658 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1659 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1660 		    disk->d_consumer != NULL &&
1661 		    disk->d_consumer->provider != NULL) {
1662 			pp->stripesize = disk->d_consumer->provider->stripesize;
1663 			off = disk->d_consumer->provider->stripeoffset;
1664 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1665 			if (off > 0)
1666 				pp->stripeoffset %= off;
1667 		}
1668 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1669 			pp->stripesize *= (vol->v_disks_count - 1);
1670 			pp->stripeoffset *= (vol->v_disks_count - 1);
1671 		}
1672 	} else
1673 		pp->stripesize = vol->v_strip_size;
1674 	vol->v_provider = pp;
1675 	g_error_provider(pp, 0);
1676 	g_topology_unlock();
1677 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1678 	    pp->name, vol->v_name);
1679 }
1680 
1681 static void
1682 g_raid_destroy_provider(struct g_raid_volume *vol)
1683 {
1684 	struct g_raid_softc *sc;
1685 	struct g_provider *pp;
1686 	struct bio *bp, *tmp;
1687 
1688 	g_topology_assert_not();
1689 	sc = vol->v_softc;
1690 	pp = vol->v_provider;
1691 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1692 
1693 	g_topology_lock();
1694 	g_error_provider(pp, ENXIO);
1695 	mtx_lock(&sc->sc_queue_mtx);
1696 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1697 		if (bp->bio_to != pp)
1698 			continue;
1699 		bioq_remove(&sc->sc_queue, bp);
1700 		g_io_deliver(bp, ENXIO);
1701 	}
1702 	mtx_unlock(&sc->sc_queue_mtx);
1703 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1704 	    pp->name, vol->v_name);
1705 	g_wither_provider(pp, ENXIO);
1706 	g_topology_unlock();
1707 	vol->v_provider = NULL;
1708 }
1709 
1710 /*
1711  * Update device state.
1712  */
1713 static int
1714 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1715 {
1716 	struct g_raid_softc *sc;
1717 
1718 	sc = vol->v_softc;
1719 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1720 
1721 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1722 	    g_raid_volume_event2str(event),
1723 	    vol->v_name);
1724 	switch (event) {
1725 	case G_RAID_VOLUME_E_DOWN:
1726 		if (vol->v_provider != NULL)
1727 			g_raid_destroy_provider(vol);
1728 		break;
1729 	case G_RAID_VOLUME_E_UP:
1730 		if (vol->v_provider == NULL)
1731 			g_raid_launch_provider(vol);
1732 		break;
1733 	case G_RAID_VOLUME_E_START:
1734 		if (vol->v_tr)
1735 			G_RAID_TR_START(vol->v_tr);
1736 		return (0);
1737 	default:
1738 		if (sc->sc_md)
1739 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1740 		return (0);
1741 	}
1742 
1743 	/* Manage root mount release. */
1744 	if (vol->v_starting) {
1745 		vol->v_starting = 0;
1746 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1747 		root_mount_rel(vol->v_rootmount);
1748 		vol->v_rootmount = NULL;
1749 	}
1750 	if (vol->v_stopping && vol->v_provider_open == 0)
1751 		g_raid_destroy_volume(vol);
1752 	return (0);
1753 }
1754 
1755 /*
1756  * Update subdisk state.
1757  */
1758 static int
1759 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1760 {
1761 	struct g_raid_softc *sc;
1762 	struct g_raid_volume *vol;
1763 
1764 	sc = sd->sd_softc;
1765 	vol = sd->sd_volume;
1766 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1767 
1768 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1769 	    g_raid_subdisk_event2str(event),
1770 	    vol->v_name, sd->sd_pos,
1771 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1772 	if (vol->v_tr)
1773 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1774 
1775 	return (0);
1776 }
1777 
1778 /*
1779  * Update disk state.
1780  */
1781 static int
1782 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1783 {
1784 	struct g_raid_softc *sc;
1785 
1786 	sc = disk->d_softc;
1787 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1788 
1789 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1790 	    g_raid_disk_event2str(event),
1791 	    g_raid_get_diskname(disk));
1792 
1793 	if (sc->sc_md)
1794 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1795 	return (0);
1796 }
1797 
1798 /*
1799  * Node event.
1800  */
1801 static int
1802 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1803 {
1804 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1805 
1806 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1807 	    g_raid_node_event2str(event));
1808 
1809 	if (event == G_RAID_NODE_E_WAKE)
1810 		return (0);
1811 	if (sc->sc_md)
1812 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1813 	return (0);
1814 }
1815 
1816 static int
1817 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1818 {
1819 	struct g_raid_volume *vol;
1820 	struct g_raid_softc *sc;
1821 	int dcw, opens, error = 0;
1822 
1823 	g_topology_assert();
1824 	sc = pp->geom->softc;
1825 	vol = pp->private;
1826 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1827 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1828 
1829 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1830 	    acr, acw, ace);
1831 	dcw = pp->acw + acw;
1832 
1833 	g_topology_unlock();
1834 	sx_xlock(&sc->sc_lock);
1835 	/* Deny new opens while dying. */
1836 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1837 		error = ENXIO;
1838 		goto out;
1839 	}
1840 	/* Deny write opens for read-only volumes. */
1841 	if (vol->v_read_only && acw > 0) {
1842 		error = EROFS;
1843 		goto out;
1844 	}
1845 	if (dcw == 0)
1846 		g_raid_clean(vol, dcw);
1847 	vol->v_provider_open += acr + acw + ace;
1848 	/* Handle delayed node destruction. */
1849 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1850 	    vol->v_provider_open == 0) {
1851 		/* Count open volumes. */
1852 		opens = g_raid_nopens(sc);
1853 		if (opens == 0) {
1854 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1855 			/* Wake up worker to make it selfdestruct. */
1856 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1857 		}
1858 	}
1859 	/* Handle open volume destruction. */
1860 	if (vol->v_stopping && vol->v_provider_open == 0)
1861 		g_raid_destroy_volume(vol);
1862 out:
1863 	sx_xunlock(&sc->sc_lock);
1864 	g_topology_lock();
1865 	return (error);
1866 }
1867 
1868 struct g_raid_softc *
1869 g_raid_create_node(struct g_class *mp,
1870     const char *name, struct g_raid_md_object *md)
1871 {
1872 	struct g_raid_softc *sc;
1873 	struct g_geom *gp;
1874 	int error;
1875 
1876 	g_topology_assert();
1877 	G_RAID_DEBUG(1, "Creating array %s.", name);
1878 
1879 	gp = g_new_geomf(mp, "%s", name);
1880 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1881 	gp->start = g_raid_start;
1882 	gp->orphan = g_raid_orphan;
1883 	gp->access = g_raid_access;
1884 	gp->dumpconf = g_raid_dumpconf;
1885 
1886 	sc->sc_md = md;
1887 	sc->sc_geom = gp;
1888 	sc->sc_flags = 0;
1889 	TAILQ_INIT(&sc->sc_volumes);
1890 	TAILQ_INIT(&sc->sc_disks);
1891 	sx_init(&sc->sc_lock, "graid:lock");
1892 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1893 	TAILQ_INIT(&sc->sc_events);
1894 	bioq_init(&sc->sc_queue);
1895 	gp->softc = sc;
1896 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1897 	    "g_raid %s", name);
1898 	if (error != 0) {
1899 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1900 		mtx_destroy(&sc->sc_queue_mtx);
1901 		sx_destroy(&sc->sc_lock);
1902 		g_destroy_geom(sc->sc_geom);
1903 		free(sc, M_RAID);
1904 		return (NULL);
1905 	}
1906 
1907 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1908 	return (sc);
1909 }
1910 
1911 struct g_raid_volume *
1912 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1913 {
1914 	struct g_raid_volume	*vol, *vol1;
1915 	int i;
1916 
1917 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1918 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1919 	vol->v_softc = sc;
1920 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1921 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1922 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1923 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1924 	vol->v_rotate_parity = 1;
1925 	bioq_init(&vol->v_inflight);
1926 	bioq_init(&vol->v_locked);
1927 	LIST_INIT(&vol->v_locks);
1928 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1929 		vol->v_subdisks[i].sd_softc = sc;
1930 		vol->v_subdisks[i].sd_volume = vol;
1931 		vol->v_subdisks[i].sd_pos = i;
1932 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1933 	}
1934 
1935 	/* Find free ID for this volume. */
1936 	g_topology_lock();
1937 	vol1 = vol;
1938 	if (id >= 0) {
1939 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1940 			if (vol1->v_global_id == id)
1941 				break;
1942 		}
1943 	}
1944 	if (vol1 != NULL) {
1945 		for (id = 0; ; id++) {
1946 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1947 				if (vol1->v_global_id == id)
1948 					break;
1949 			}
1950 			if (vol1 == NULL)
1951 				break;
1952 		}
1953 	}
1954 	vol->v_global_id = id;
1955 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1956 	g_topology_unlock();
1957 
1958 	/* Delay root mounting. */
1959 	vol->v_rootmount = root_mount_hold("GRAID");
1960 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1961 	vol->v_starting = 1;
1962 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1963 	return (vol);
1964 }
1965 
1966 struct g_raid_disk *
1967 g_raid_create_disk(struct g_raid_softc *sc)
1968 {
1969 	struct g_raid_disk	*disk;
1970 
1971 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1972 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1973 	disk->d_softc = sc;
1974 	disk->d_state = G_RAID_DISK_S_NONE;
1975 	TAILQ_INIT(&disk->d_subdisks);
1976 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1977 	return (disk);
1978 }
1979 
1980 int g_raid_start_volume(struct g_raid_volume *vol)
1981 {
1982 	struct g_raid_tr_class *class;
1983 	struct g_raid_tr_object *obj;
1984 	int status;
1985 
1986 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1987 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1988 		if (!class->trc_enable)
1989 			continue;
1990 		G_RAID_DEBUG1(2, vol->v_softc,
1991 		    "Tasting volume %s for %s transformation.",
1992 		    vol->v_name, class->name);
1993 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1994 		    M_WAITOK);
1995 		obj->tro_class = class;
1996 		obj->tro_volume = vol;
1997 		status = G_RAID_TR_TASTE(obj, vol);
1998 		if (status != G_RAID_TR_TASTE_FAIL)
1999 			break;
2000 		kobj_delete((kobj_t)obj, M_RAID);
2001 	}
2002 	if (class == NULL) {
2003 		G_RAID_DEBUG1(0, vol->v_softc,
2004 		    "No transformation module found for %s.",
2005 		    vol->v_name);
2006 		vol->v_tr = NULL;
2007 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2008 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2009 		    G_RAID_EVENT_VOLUME);
2010 		return (-1);
2011 	}
2012 	G_RAID_DEBUG1(2, vol->v_softc,
2013 	    "Transformation module %s chosen for %s.",
2014 	    class->name, vol->v_name);
2015 	vol->v_tr = obj;
2016 	return (0);
2017 }
2018 
2019 int
2020 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2021 {
2022 	struct g_raid_volume *vol, *tmpv;
2023 	struct g_raid_disk *disk, *tmpd;
2024 	int error = 0;
2025 
2026 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2027 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2028 		if (g_raid_destroy_volume(vol))
2029 			error = EBUSY;
2030 	}
2031 	if (error)
2032 		return (error);
2033 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2034 		if (g_raid_destroy_disk(disk))
2035 			error = EBUSY;
2036 	}
2037 	if (error)
2038 		return (error);
2039 	if (sc->sc_md) {
2040 		G_RAID_MD_FREE(sc->sc_md);
2041 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2042 		sc->sc_md = NULL;
2043 	}
2044 	if (sc->sc_geom != NULL) {
2045 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2046 		g_topology_lock();
2047 		sc->sc_geom->softc = NULL;
2048 		g_wither_geom(sc->sc_geom, ENXIO);
2049 		g_topology_unlock();
2050 		sc->sc_geom = NULL;
2051 	} else
2052 		G_RAID_DEBUG(1, "Array destroyed.");
2053 	if (worker) {
2054 		g_raid_event_cancel(sc, sc);
2055 		mtx_destroy(&sc->sc_queue_mtx);
2056 		sx_xunlock(&sc->sc_lock);
2057 		sx_destroy(&sc->sc_lock);
2058 		wakeup(&sc->sc_stopping);
2059 		free(sc, M_RAID);
2060 		curthread->td_pflags &= ~TDP_GEOM;
2061 		G_RAID_DEBUG(1, "Thread exiting.");
2062 		kproc_exit(0);
2063 	} else {
2064 		/* Wake up worker to make it selfdestruct. */
2065 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2066 	}
2067 	return (0);
2068 }
2069 
2070 int
2071 g_raid_destroy_volume(struct g_raid_volume *vol)
2072 {
2073 	struct g_raid_softc *sc;
2074 	struct g_raid_disk *disk;
2075 	int i;
2076 
2077 	sc = vol->v_softc;
2078 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2079 	vol->v_stopping = 1;
2080 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2081 		if (vol->v_tr) {
2082 			G_RAID_TR_STOP(vol->v_tr);
2083 			return (EBUSY);
2084 		} else
2085 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2086 	}
2087 	if (g_raid_event_check(sc, vol) != 0)
2088 		return (EBUSY);
2089 	if (vol->v_provider != NULL)
2090 		return (EBUSY);
2091 	if (vol->v_provider_open != 0)
2092 		return (EBUSY);
2093 	if (vol->v_tr) {
2094 		G_RAID_TR_FREE(vol->v_tr);
2095 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2096 		vol->v_tr = NULL;
2097 	}
2098 	if (vol->v_rootmount)
2099 		root_mount_rel(vol->v_rootmount);
2100 	g_topology_lock();
2101 	LIST_REMOVE(vol, v_global_next);
2102 	g_topology_unlock();
2103 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2104 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2105 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2106 		disk = vol->v_subdisks[i].sd_disk;
2107 		if (disk == NULL)
2108 			continue;
2109 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2110 	}
2111 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2112 	if (sc->sc_md)
2113 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2114 	g_raid_event_cancel(sc, vol);
2115 	free(vol, M_RAID);
2116 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2117 		/* Wake up worker to let it selfdestruct. */
2118 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2119 	}
2120 	return (0);
2121 }
2122 
2123 int
2124 g_raid_destroy_disk(struct g_raid_disk *disk)
2125 {
2126 	struct g_raid_softc *sc;
2127 	struct g_raid_subdisk *sd, *tmp;
2128 
2129 	sc = disk->d_softc;
2130 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2131 	if (disk->d_consumer) {
2132 		g_raid_kill_consumer(sc, disk->d_consumer);
2133 		disk->d_consumer = NULL;
2134 	}
2135 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2136 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2137 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2138 		    G_RAID_EVENT_SUBDISK);
2139 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2140 		sd->sd_disk = NULL;
2141 	}
2142 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2143 	if (sc->sc_md)
2144 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2145 	g_raid_event_cancel(sc, disk);
2146 	free(disk, M_RAID);
2147 	return (0);
2148 }
2149 
2150 int
2151 g_raid_destroy(struct g_raid_softc *sc, int how)
2152 {
2153 	int error, opens;
2154 
2155 	g_topology_assert_not();
2156 	if (sc == NULL)
2157 		return (ENXIO);
2158 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2159 
2160 	/* Count open volumes. */
2161 	opens = g_raid_nopens(sc);
2162 
2163 	/* React on some opened volumes. */
2164 	if (opens > 0) {
2165 		switch (how) {
2166 		case G_RAID_DESTROY_SOFT:
2167 			G_RAID_DEBUG1(1, sc,
2168 			    "%d volumes are still open.",
2169 			    opens);
2170 			sx_xunlock(&sc->sc_lock);
2171 			return (EBUSY);
2172 		case G_RAID_DESTROY_DELAYED:
2173 			G_RAID_DEBUG1(1, sc,
2174 			    "Array will be destroyed on last close.");
2175 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2176 			sx_xunlock(&sc->sc_lock);
2177 			return (EBUSY);
2178 		case G_RAID_DESTROY_HARD:
2179 			G_RAID_DEBUG1(1, sc,
2180 			    "%d volumes are still open.",
2181 			    opens);
2182 		}
2183 	}
2184 
2185 	/* Mark node for destruction. */
2186 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2187 	/* Wake up worker to let it selfdestruct. */
2188 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2189 	/* Sleep until node destroyed. */
2190 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2191 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2192 	return (error == EWOULDBLOCK ? EBUSY : 0);
2193 }
2194 
2195 static void
2196 g_raid_taste_orphan(struct g_consumer *cp)
2197 {
2198 
2199 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2200 	    cp->provider->name));
2201 }
2202 
2203 static struct g_geom *
2204 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2205 {
2206 	struct g_consumer *cp;
2207 	struct g_geom *gp, *geom;
2208 	struct g_raid_md_class *class;
2209 	struct g_raid_md_object *obj;
2210 	int status;
2211 
2212 	g_topology_assert();
2213 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2214 	if (!g_raid_enable)
2215 		return (NULL);
2216 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2217 
2218 	geom = NULL;
2219 	status = G_RAID_MD_TASTE_FAIL;
2220 	gp = g_new_geomf(mp, "raid:taste");
2221 	/*
2222 	 * This orphan function should be never called.
2223 	 */
2224 	gp->orphan = g_raid_taste_orphan;
2225 	cp = g_new_consumer(gp);
2226 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
2227 	if (g_attach(cp, pp) != 0)
2228 		goto ofail2;
2229 	if (g_access(cp, 1, 0, 0) != 0)
2230 		goto ofail;
2231 
2232 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2233 		if (!class->mdc_enable)
2234 			continue;
2235 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2236 		    pp->name, class->name);
2237 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2238 		    M_WAITOK);
2239 		obj->mdo_class = class;
2240 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2241 		if (status != G_RAID_MD_TASTE_NEW)
2242 			kobj_delete((kobj_t)obj, M_RAID);
2243 		if (status != G_RAID_MD_TASTE_FAIL)
2244 			break;
2245 	}
2246 
2247 	if (status == G_RAID_MD_TASTE_FAIL)
2248 		(void)g_access(cp, -1, 0, 0);
2249 ofail:
2250 	g_detach(cp);
2251 ofail2:
2252 	g_destroy_consumer(cp);
2253 	g_destroy_geom(gp);
2254 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2255 	return (geom);
2256 }
2257 
2258 int
2259 g_raid_create_node_format(const char *format, struct gctl_req *req,
2260     struct g_geom **gp)
2261 {
2262 	struct g_raid_md_class *class;
2263 	struct g_raid_md_object *obj;
2264 	int status;
2265 
2266 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2267 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2268 		if (strcasecmp(class->name, format) == 0)
2269 			break;
2270 	}
2271 	if (class == NULL) {
2272 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2273 		return (G_RAID_MD_TASTE_FAIL);
2274 	}
2275 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2276 	    M_WAITOK);
2277 	obj->mdo_class = class;
2278 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2279 	if (status != G_RAID_MD_TASTE_NEW)
2280 		kobj_delete((kobj_t)obj, M_RAID);
2281 	return (status);
2282 }
2283 
2284 static int
2285 g_raid_destroy_geom(struct gctl_req *req __unused,
2286     struct g_class *mp __unused, struct g_geom *gp)
2287 {
2288 	struct g_raid_softc *sc;
2289 	int error;
2290 
2291 	g_topology_unlock();
2292 	sc = gp->softc;
2293 	sx_xlock(&sc->sc_lock);
2294 	g_cancel_event(sc);
2295 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2296 	g_topology_lock();
2297 	return (error);
2298 }
2299 
2300 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2301     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2302 {
2303 
2304 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2305 		return;
2306 	if (sc->sc_md)
2307 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2308 }
2309 
2310 void g_raid_fail_disk(struct g_raid_softc *sc,
2311     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2312 {
2313 
2314 	if (disk == NULL)
2315 		disk = sd->sd_disk;
2316 	if (disk == NULL) {
2317 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2318 		return;
2319 	}
2320 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2321 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2322 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2323 		return;
2324 	}
2325 	if (sc->sc_md)
2326 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2327 }
2328 
2329 static void
2330 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2331     struct g_consumer *cp, struct g_provider *pp)
2332 {
2333 	struct g_raid_softc *sc;
2334 	struct g_raid_volume *vol;
2335 	struct g_raid_subdisk *sd;
2336 	struct g_raid_disk *disk;
2337 	int i, s;
2338 
2339 	g_topology_assert();
2340 
2341 	sc = gp->softc;
2342 	if (sc == NULL)
2343 		return;
2344 	if (pp != NULL) {
2345 		vol = pp->private;
2346 		g_topology_unlock();
2347 		sx_xlock(&sc->sc_lock);
2348 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2349 		    sc->sc_md->mdo_class->name,
2350 		    g_raid_volume_level2str(vol->v_raid_level,
2351 		    vol->v_raid_level_qualifier));
2352 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2353 		    vol->v_name);
2354 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2355 		    g_raid_volume_level2str(vol->v_raid_level,
2356 		    vol->v_raid_level_qualifier));
2357 		sbuf_printf(sb,
2358 		    "%s<Transformation>%s</Transformation>\n", indent,
2359 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2360 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2361 		    vol->v_disks_count);
2362 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2363 		    vol->v_strip_size);
2364 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2365 		    g_raid_volume_state2str(vol->v_state));
2366 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2367 		    vol->v_dirty ? "Yes" : "No");
2368 		sbuf_printf(sb, "%s<Subdisks>", indent);
2369 		for (i = 0; i < vol->v_disks_count; i++) {
2370 			sd = &vol->v_subdisks[i];
2371 			if (sd->sd_disk != NULL &&
2372 			    sd->sd_disk->d_consumer != NULL) {
2373 				sbuf_printf(sb, "%s ",
2374 				    g_raid_get_diskname(sd->sd_disk));
2375 			} else {
2376 				sbuf_cat(sb, "NONE ");
2377 			}
2378 			sbuf_printf(sb, "(%s",
2379 			    g_raid_subdisk_state2str(sd->sd_state));
2380 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2381 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2382 				sbuf_printf(sb, " %d%%",
2383 				    (int)(sd->sd_rebuild_pos * 100 /
2384 				     sd->sd_size));
2385 			}
2386 			sbuf_cat(sb, ")");
2387 			if (i + 1 < vol->v_disks_count)
2388 				sbuf_cat(sb, ", ");
2389 		}
2390 		sbuf_cat(sb, "</Subdisks>\n");
2391 		sx_xunlock(&sc->sc_lock);
2392 		g_topology_lock();
2393 	} else if (cp != NULL) {
2394 		disk = cp->private;
2395 		if (disk == NULL)
2396 			return;
2397 		g_topology_unlock();
2398 		sx_xlock(&sc->sc_lock);
2399 		sbuf_printf(sb, "%s<State>%s", indent,
2400 		    g_raid_disk_state2str(disk->d_state));
2401 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2402 			sbuf_cat(sb, " (");
2403 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2404 				sbuf_printf(sb, "%s",
2405 				    g_raid_subdisk_state2str(sd->sd_state));
2406 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2407 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2408 					sbuf_printf(sb, " %d%%",
2409 					    (int)(sd->sd_rebuild_pos * 100 /
2410 					     sd->sd_size));
2411 				}
2412 				if (TAILQ_NEXT(sd, sd_next))
2413 					sbuf_cat(sb, ", ");
2414 			}
2415 			sbuf_cat(sb, ")");
2416 		}
2417 		sbuf_cat(sb, "</State>\n");
2418 		sbuf_printf(sb, "%s<Subdisks>", indent);
2419 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2420 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2421 			    sd->sd_volume->v_global_id,
2422 			    sd->sd_volume->v_name,
2423 			    sd->sd_pos, (uintmax_t)sd->sd_offset);
2424 			if (TAILQ_NEXT(sd, sd_next))
2425 				sbuf_cat(sb, ", ");
2426 		}
2427 		sbuf_cat(sb, "</Subdisks>\n");
2428 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2429 		    disk->d_read_errs);
2430 		sx_xunlock(&sc->sc_lock);
2431 		g_topology_lock();
2432 	} else {
2433 		g_topology_unlock();
2434 		sx_xlock(&sc->sc_lock);
2435 		if (sc->sc_md) {
2436 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2437 			    sc->sc_md->mdo_class->name);
2438 		}
2439 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2440 			s = 0xff;
2441 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2442 				if (vol->v_state < s)
2443 					s = vol->v_state;
2444 			}
2445 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2446 			    g_raid_volume_state2str(s));
2447 		}
2448 		sx_xunlock(&sc->sc_lock);
2449 		g_topology_lock();
2450 	}
2451 }
2452 
2453 static void
2454 g_raid_shutdown_post_sync(void *arg, int howto)
2455 {
2456 	struct g_class *mp;
2457 	struct g_geom *gp, *gp2;
2458 	struct g_raid_softc *sc;
2459 	struct g_raid_volume *vol;
2460 
2461 	if ((howto & RB_NOSYNC) != 0)
2462 		return;
2463 
2464 	mp = arg;
2465 	g_topology_lock();
2466 	g_raid_shutdown = 1;
2467 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2468 		if ((sc = gp->softc) == NULL)
2469 			continue;
2470 		g_topology_unlock();
2471 		sx_xlock(&sc->sc_lock);
2472 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2473 			g_raid_clean(vol, -1);
2474 		g_cancel_event(sc);
2475 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2476 		g_topology_lock();
2477 	}
2478 	g_topology_unlock();
2479 }
2480 
2481 static void
2482 g_raid_init(struct g_class *mp)
2483 {
2484 
2485 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2486 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2487 	if (g_raid_post_sync == NULL)
2488 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2489 	g_raid_started = 1;
2490 }
2491 
2492 static void
2493 g_raid_fini(struct g_class *mp)
2494 {
2495 
2496 	if (g_raid_post_sync != NULL)
2497 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2498 	g_raid_started = 0;
2499 }
2500 
2501 int
2502 g_raid_md_modevent(module_t mod, int type, void *arg)
2503 {
2504 	struct g_raid_md_class *class, *c, *nc;
2505 	int error;
2506 
2507 	error = 0;
2508 	class = arg;
2509 	switch (type) {
2510 	case MOD_LOAD:
2511 		c = LIST_FIRST(&g_raid_md_classes);
2512 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2513 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2514 		else {
2515 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2516 			    nc->mdc_priority < class->mdc_priority)
2517 				c = nc;
2518 			LIST_INSERT_AFTER(c, class, mdc_list);
2519 		}
2520 		if (g_raid_started)
2521 			g_retaste(&g_raid_class);
2522 		break;
2523 	case MOD_UNLOAD:
2524 		LIST_REMOVE(class, mdc_list);
2525 		break;
2526 	default:
2527 		error = EOPNOTSUPP;
2528 		break;
2529 	}
2530 
2531 	return (error);
2532 }
2533 
2534 int
2535 g_raid_tr_modevent(module_t mod, int type, void *arg)
2536 {
2537 	struct g_raid_tr_class *class, *c, *nc;
2538 	int error;
2539 
2540 	error = 0;
2541 	class = arg;
2542 	switch (type) {
2543 	case MOD_LOAD:
2544 		c = LIST_FIRST(&g_raid_tr_classes);
2545 		if (c == NULL || c->trc_priority > class->trc_priority)
2546 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2547 		else {
2548 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2549 			    nc->trc_priority < class->trc_priority)
2550 				c = nc;
2551 			LIST_INSERT_AFTER(c, class, trc_list);
2552 		}
2553 		break;
2554 	case MOD_UNLOAD:
2555 		LIST_REMOVE(class, trc_list);
2556 		break;
2557 	default:
2558 		error = EOPNOTSUPP;
2559 		break;
2560 	}
2561 
2562 	return (error);
2563 }
2564 
2565 /*
2566  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2567  * to reduce module priority, allowing submodules to register them first.
2568  */
2569 static moduledata_t g_raid_mod = {
2570 	"g_raid",
2571 	g_modevent,
2572 	&g_raid_class
2573 };
2574 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2575 MODULE_VERSION(geom_raid, 0);
2576