xref: /freebsd/sys/geom/raid/g_raid.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/bio.h>
40 #include <sys/sbuf.h>
41 #include <sys/sysctl.h>
42 #include <sys/malloc.h>
43 #include <sys/eventhandler.h>
44 #include <vm/uma.h>
45 #include <geom/geom.h>
46 #include <sys/proc.h>
47 #include <sys/kthread.h>
48 #include <sys/sched.h>
49 #include <geom/raid/g_raid.h>
50 #include "g_raid_md_if.h"
51 #include "g_raid_tr_if.h"
52 
53 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
54 
55 SYSCTL_DECL(_kern_geom);
56 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
57 int g_raid_enable = 1;
58 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN,
59     &g_raid_enable, 0, "Enable on-disk metadata taste");
60 u_int g_raid_aggressive_spare = 0;
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN,
62     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0,
65     "Debug level");
66 int g_raid_read_err_thresh = 10;
67 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN,
68     &g_raid_read_err_thresh, 0,
69     "Number of read errors equated to disk failure");
70 u_int g_raid_start_timeout = 30;
71 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN,
72     &g_raid_start_timeout, 0,
73     "Time to wait for all array components");
74 static u_int g_raid_clean_time = 5;
75 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN,
76     &g_raid_clean_time, 0, "Mark volume as clean when idling");
77 static u_int g_raid_disconnect_on_failure = 1;
78 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN,
79     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
80 static u_int g_raid_name_format = 0;
81 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN,
82     &g_raid_name_format, 0, "Providers name format.");
83 static u_int g_raid_idle_threshold = 1000000;
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN,
85     &g_raid_idle_threshold, 1000000,
86     "Time in microseconds to consider a volume idle.");
87 
88 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
89 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
90 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
91 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
92 } while (0)
93 
94 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
95     LIST_HEAD_INITIALIZER(g_raid_md_classes);
96 
97 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
98     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
99 
100 LIST_HEAD(, g_raid_volume) g_raid_volumes =
101     LIST_HEAD_INITIALIZER(g_raid_volumes);
102 
103 static eventhandler_tag g_raid_post_sync = NULL;
104 static int g_raid_started = 0;
105 static int g_raid_shutdown = 0;
106 
107 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
108     struct g_geom *gp);
109 static g_taste_t g_raid_taste;
110 static void g_raid_init(struct g_class *mp);
111 static void g_raid_fini(struct g_class *mp);
112 
113 struct g_class g_raid_class = {
114 	.name = G_RAID_CLASS_NAME,
115 	.version = G_VERSION,
116 	.ctlreq = g_raid_ctl,
117 	.taste = g_raid_taste,
118 	.destroy_geom = g_raid_destroy_geom,
119 	.init = g_raid_init,
120 	.fini = g_raid_fini
121 };
122 
123 static void g_raid_destroy_provider(struct g_raid_volume *vol);
124 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
125 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
126 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
127 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
128 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
129     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
130 static void g_raid_start(struct bio *bp);
131 static void g_raid_start_request(struct bio *bp);
132 static void g_raid_disk_done(struct bio *bp);
133 static void g_raid_poll(struct g_raid_softc *sc);
134 
135 static const char *
136 g_raid_node_event2str(int event)
137 {
138 
139 	switch (event) {
140 	case G_RAID_NODE_E_WAKE:
141 		return ("WAKE");
142 	case G_RAID_NODE_E_START:
143 		return ("START");
144 	default:
145 		return ("INVALID");
146 	}
147 }
148 
149 const char *
150 g_raid_disk_state2str(int state)
151 {
152 
153 	switch (state) {
154 	case G_RAID_DISK_S_NONE:
155 		return ("NONE");
156 	case G_RAID_DISK_S_OFFLINE:
157 		return ("OFFLINE");
158 	case G_RAID_DISK_S_DISABLED:
159 		return ("DISABLED");
160 	case G_RAID_DISK_S_FAILED:
161 		return ("FAILED");
162 	case G_RAID_DISK_S_STALE_FAILED:
163 		return ("STALE_FAILED");
164 	case G_RAID_DISK_S_SPARE:
165 		return ("SPARE");
166 	case G_RAID_DISK_S_STALE:
167 		return ("STALE");
168 	case G_RAID_DISK_S_ACTIVE:
169 		return ("ACTIVE");
170 	default:
171 		return ("INVALID");
172 	}
173 }
174 
175 static const char *
176 g_raid_disk_event2str(int event)
177 {
178 
179 	switch (event) {
180 	case G_RAID_DISK_E_DISCONNECTED:
181 		return ("DISCONNECTED");
182 	default:
183 		return ("INVALID");
184 	}
185 }
186 
187 const char *
188 g_raid_subdisk_state2str(int state)
189 {
190 
191 	switch (state) {
192 	case G_RAID_SUBDISK_S_NONE:
193 		return ("NONE");
194 	case G_RAID_SUBDISK_S_FAILED:
195 		return ("FAILED");
196 	case G_RAID_SUBDISK_S_NEW:
197 		return ("NEW");
198 	case G_RAID_SUBDISK_S_REBUILD:
199 		return ("REBUILD");
200 	case G_RAID_SUBDISK_S_UNINITIALIZED:
201 		return ("UNINITIALIZED");
202 	case G_RAID_SUBDISK_S_STALE:
203 		return ("STALE");
204 	case G_RAID_SUBDISK_S_RESYNC:
205 		return ("RESYNC");
206 	case G_RAID_SUBDISK_S_ACTIVE:
207 		return ("ACTIVE");
208 	default:
209 		return ("INVALID");
210 	}
211 }
212 
213 static const char *
214 g_raid_subdisk_event2str(int event)
215 {
216 
217 	switch (event) {
218 	case G_RAID_SUBDISK_E_NEW:
219 		return ("NEW");
220 	case G_RAID_SUBDISK_E_FAILED:
221 		return ("FAILED");
222 	case G_RAID_SUBDISK_E_DISCONNECTED:
223 		return ("DISCONNECTED");
224 	default:
225 		return ("INVALID");
226 	}
227 }
228 
229 const char *
230 g_raid_volume_state2str(int state)
231 {
232 
233 	switch (state) {
234 	case G_RAID_VOLUME_S_STARTING:
235 		return ("STARTING");
236 	case G_RAID_VOLUME_S_BROKEN:
237 		return ("BROKEN");
238 	case G_RAID_VOLUME_S_DEGRADED:
239 		return ("DEGRADED");
240 	case G_RAID_VOLUME_S_SUBOPTIMAL:
241 		return ("SUBOPTIMAL");
242 	case G_RAID_VOLUME_S_OPTIMAL:
243 		return ("OPTIMAL");
244 	case G_RAID_VOLUME_S_UNSUPPORTED:
245 		return ("UNSUPPORTED");
246 	case G_RAID_VOLUME_S_STOPPED:
247 		return ("STOPPED");
248 	default:
249 		return ("INVALID");
250 	}
251 }
252 
253 static const char *
254 g_raid_volume_event2str(int event)
255 {
256 
257 	switch (event) {
258 	case G_RAID_VOLUME_E_UP:
259 		return ("UP");
260 	case G_RAID_VOLUME_E_DOWN:
261 		return ("DOWN");
262 	case G_RAID_VOLUME_E_START:
263 		return ("START");
264 	case G_RAID_VOLUME_E_STARTMD:
265 		return ("STARTMD");
266 	default:
267 		return ("INVALID");
268 	}
269 }
270 
271 const char *
272 g_raid_volume_level2str(int level, int qual)
273 {
274 
275 	switch (level) {
276 	case G_RAID_VOLUME_RL_RAID0:
277 		return ("RAID0");
278 	case G_RAID_VOLUME_RL_RAID1:
279 		return ("RAID1");
280 	case G_RAID_VOLUME_RL_RAID3:
281 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
282 			return ("RAID3-P0");
283 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
284 			return ("RAID3-PN");
285 		return ("RAID3");
286 	case G_RAID_VOLUME_RL_RAID4:
287 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
288 			return ("RAID4-P0");
289 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
290 			return ("RAID4-PN");
291 		return ("RAID4");
292 	case G_RAID_VOLUME_RL_RAID5:
293 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
294 			return ("RAID5-RA");
295 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
296 			return ("RAID5-RS");
297 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
298 			return ("RAID5-LA");
299 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
300 			return ("RAID5-LS");
301 		return ("RAID5");
302 	case G_RAID_VOLUME_RL_RAID6:
303 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
304 			return ("RAID6-RA");
305 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
306 			return ("RAID6-RS");
307 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
308 			return ("RAID6-LA");
309 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
310 			return ("RAID6-LS");
311 		return ("RAID6");
312 	case G_RAID_VOLUME_RL_RAIDMDF:
313 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
314 			return ("RAIDMDF-RA");
315 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
316 			return ("RAIDMDF-RS");
317 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
318 			return ("RAIDMDF-LA");
319 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
320 			return ("RAIDMDF-LS");
321 		return ("RAIDMDF");
322 	case G_RAID_VOLUME_RL_RAID1E:
323 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
324 			return ("RAID1E-A");
325 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
326 			return ("RAID1E-O");
327 		return ("RAID1E");
328 	case G_RAID_VOLUME_RL_SINGLE:
329 		return ("SINGLE");
330 	case G_RAID_VOLUME_RL_CONCAT:
331 		return ("CONCAT");
332 	case G_RAID_VOLUME_RL_RAID5E:
333 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
334 			return ("RAID5E-RA");
335 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
336 			return ("RAID5E-RS");
337 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
338 			return ("RAID5E-LA");
339 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
340 			return ("RAID5E-LS");
341 		return ("RAID5E");
342 	case G_RAID_VOLUME_RL_RAID5EE:
343 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
344 			return ("RAID5EE-RA");
345 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
346 			return ("RAID5EE-RS");
347 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
348 			return ("RAID5EE-LA");
349 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
350 			return ("RAID5EE-LS");
351 		return ("RAID5EE");
352 	case G_RAID_VOLUME_RL_RAID5R:
353 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
354 			return ("RAID5R-RA");
355 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
356 			return ("RAID5R-RS");
357 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
358 			return ("RAID5R-LA");
359 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
360 			return ("RAID5R-LS");
361 		return ("RAID5E");
362 	default:
363 		return ("UNKNOWN");
364 	}
365 }
366 
367 int
368 g_raid_volume_str2level(const char *str, int *level, int *qual)
369 {
370 
371 	*level = G_RAID_VOLUME_RL_UNKNOWN;
372 	*qual = G_RAID_VOLUME_RLQ_NONE;
373 	if (strcasecmp(str, "RAID0") == 0)
374 		*level = G_RAID_VOLUME_RL_RAID0;
375 	else if (strcasecmp(str, "RAID1") == 0)
376 		*level = G_RAID_VOLUME_RL_RAID1;
377 	else if (strcasecmp(str, "RAID3-P0") == 0) {
378 		*level = G_RAID_VOLUME_RL_RAID3;
379 		*qual = G_RAID_VOLUME_RLQ_R3P0;
380 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
381 		   strcasecmp(str, "RAID3") == 0) {
382 		*level = G_RAID_VOLUME_RL_RAID3;
383 		*qual = G_RAID_VOLUME_RLQ_R3PN;
384 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
385 		*level = G_RAID_VOLUME_RL_RAID4;
386 		*qual = G_RAID_VOLUME_RLQ_R4P0;
387 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
388 		   strcasecmp(str, "RAID4") == 0) {
389 		*level = G_RAID_VOLUME_RL_RAID4;
390 		*qual = G_RAID_VOLUME_RLQ_R4PN;
391 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
392 		*level = G_RAID_VOLUME_RL_RAID5;
393 		*qual = G_RAID_VOLUME_RLQ_R5RA;
394 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
395 		*level = G_RAID_VOLUME_RL_RAID5;
396 		*qual = G_RAID_VOLUME_RLQ_R5RS;
397 	} else if (strcasecmp(str, "RAID5") == 0 ||
398 		   strcasecmp(str, "RAID5-LA") == 0) {
399 		*level = G_RAID_VOLUME_RL_RAID5;
400 		*qual = G_RAID_VOLUME_RLQ_R5LA;
401 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
402 		*level = G_RAID_VOLUME_RL_RAID5;
403 		*qual = G_RAID_VOLUME_RLQ_R5LS;
404 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
405 		*level = G_RAID_VOLUME_RL_RAID6;
406 		*qual = G_RAID_VOLUME_RLQ_R6RA;
407 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
408 		*level = G_RAID_VOLUME_RL_RAID6;
409 		*qual = G_RAID_VOLUME_RLQ_R6RS;
410 	} else if (strcasecmp(str, "RAID6") == 0 ||
411 		   strcasecmp(str, "RAID6-LA") == 0) {
412 		*level = G_RAID_VOLUME_RL_RAID6;
413 		*qual = G_RAID_VOLUME_RLQ_R6LA;
414 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
415 		*level = G_RAID_VOLUME_RL_RAID6;
416 		*qual = G_RAID_VOLUME_RLQ_R6LS;
417 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
418 		*level = G_RAID_VOLUME_RL_RAIDMDF;
419 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
420 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
421 		*level = G_RAID_VOLUME_RL_RAIDMDF;
422 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
423 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
424 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
425 		*level = G_RAID_VOLUME_RL_RAIDMDF;
426 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
427 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
428 		*level = G_RAID_VOLUME_RL_RAIDMDF;
429 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
430 	} else if (strcasecmp(str, "RAID10") == 0 ||
431 		   strcasecmp(str, "RAID1E") == 0 ||
432 		   strcasecmp(str, "RAID1E-A") == 0) {
433 		*level = G_RAID_VOLUME_RL_RAID1E;
434 		*qual = G_RAID_VOLUME_RLQ_R1EA;
435 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
436 		*level = G_RAID_VOLUME_RL_RAID1E;
437 		*qual = G_RAID_VOLUME_RLQ_R1EO;
438 	} else if (strcasecmp(str, "SINGLE") == 0)
439 		*level = G_RAID_VOLUME_RL_SINGLE;
440 	else if (strcasecmp(str, "CONCAT") == 0)
441 		*level = G_RAID_VOLUME_RL_CONCAT;
442 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
443 		*level = G_RAID_VOLUME_RL_RAID5E;
444 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
445 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
446 		*level = G_RAID_VOLUME_RL_RAID5E;
447 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
448 	} else if (strcasecmp(str, "RAID5E") == 0 ||
449 		   strcasecmp(str, "RAID5E-LA") == 0) {
450 		*level = G_RAID_VOLUME_RL_RAID5E;
451 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
452 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
453 		*level = G_RAID_VOLUME_RL_RAID5E;
454 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
455 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
456 		*level = G_RAID_VOLUME_RL_RAID5EE;
457 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
458 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
459 		*level = G_RAID_VOLUME_RL_RAID5EE;
460 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
461 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
462 		   strcasecmp(str, "RAID5EE-LA") == 0) {
463 		*level = G_RAID_VOLUME_RL_RAID5EE;
464 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
465 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
466 		*level = G_RAID_VOLUME_RL_RAID5EE;
467 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
468 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
469 		*level = G_RAID_VOLUME_RL_RAID5R;
470 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
471 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
472 		*level = G_RAID_VOLUME_RL_RAID5R;
473 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
474 	} else if (strcasecmp(str, "RAID5R") == 0 ||
475 		   strcasecmp(str, "RAID5R-LA") == 0) {
476 		*level = G_RAID_VOLUME_RL_RAID5R;
477 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
478 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
479 		*level = G_RAID_VOLUME_RL_RAID5R;
480 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
481 	} else
482 		return (-1);
483 	return (0);
484 }
485 
486 const char *
487 g_raid_get_diskname(struct g_raid_disk *disk)
488 {
489 
490 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
491 		return ("[unknown]");
492 	return (disk->d_consumer->provider->name);
493 }
494 
495 void
496 g_raid_get_disk_info(struct g_raid_disk *disk)
497 {
498 	struct g_consumer *cp = disk->d_consumer;
499 	int error, len;
500 
501 	/* Read kernel dumping information. */
502 	disk->d_kd.offset = 0;
503 	disk->d_kd.length = OFF_MAX;
504 	len = sizeof(disk->d_kd);
505 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
506 	if (error)
507 		disk->d_kd.di.dumper = NULL;
508 	if (disk->d_kd.di.dumper == NULL)
509 		G_RAID_DEBUG1(2, disk->d_softc,
510 		    "Dumping not supported by %s: %d.",
511 		    cp->provider->name, error);
512 
513 	/* Read BIO_DELETE support. */
514 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
515 	if (error)
516 		disk->d_candelete = 0;
517 	if (!disk->d_candelete)
518 		G_RAID_DEBUG1(2, disk->d_softc,
519 		    "BIO_DELETE not supported by %s: %d.",
520 		    cp->provider->name, error);
521 }
522 
523 void
524 g_raid_report_disk_state(struct g_raid_disk *disk)
525 {
526 	struct g_raid_subdisk *sd;
527 	int len, state;
528 	uint32_t s;
529 
530 	if (disk->d_consumer == NULL)
531 		return;
532 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
533 		s = G_STATE_ACTIVE; /* XXX */
534 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
535 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
536 		s = G_STATE_FAILED;
537 	} else {
538 		state = G_RAID_SUBDISK_S_ACTIVE;
539 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
540 			if (sd->sd_state < state)
541 				state = sd->sd_state;
542 		}
543 		if (state == G_RAID_SUBDISK_S_FAILED)
544 			s = G_STATE_FAILED;
545 		else if (state == G_RAID_SUBDISK_S_NEW ||
546 		    state == G_RAID_SUBDISK_S_REBUILD)
547 			s = G_STATE_REBUILD;
548 		else if (state == G_RAID_SUBDISK_S_STALE ||
549 		    state == G_RAID_SUBDISK_S_RESYNC)
550 			s = G_STATE_RESYNC;
551 		else
552 			s = G_STATE_ACTIVE;
553 	}
554 	len = sizeof(s);
555 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
556 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
557 	    g_raid_get_diskname(disk), s);
558 }
559 
560 void
561 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
562 {
563 
564 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
565 	    g_raid_get_diskname(disk),
566 	    g_raid_disk_state2str(disk->d_state),
567 	    g_raid_disk_state2str(state));
568 	disk->d_state = state;
569 	g_raid_report_disk_state(disk);
570 }
571 
572 void
573 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
574 {
575 
576 	G_RAID_DEBUG1(0, sd->sd_softc,
577 	    "Subdisk %s:%d-%s state changed from %s to %s.",
578 	    sd->sd_volume->v_name, sd->sd_pos,
579 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
580 	    g_raid_subdisk_state2str(sd->sd_state),
581 	    g_raid_subdisk_state2str(state));
582 	sd->sd_state = state;
583 	if (sd->sd_disk)
584 		g_raid_report_disk_state(sd->sd_disk);
585 }
586 
587 void
588 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
589 {
590 
591 	G_RAID_DEBUG1(0, vol->v_softc,
592 	    "Volume %s state changed from %s to %s.",
593 	    vol->v_name,
594 	    g_raid_volume_state2str(vol->v_state),
595 	    g_raid_volume_state2str(state));
596 	vol->v_state = state;
597 }
598 
599 /*
600  * --- Events handling functions ---
601  * Events in geom_raid are used to maintain subdisks and volumes status
602  * from one thread to simplify locking.
603  */
604 static void
605 g_raid_event_free(struct g_raid_event *ep)
606 {
607 
608 	free(ep, M_RAID);
609 }
610 
611 int
612 g_raid_event_send(void *arg, int event, int flags)
613 {
614 	struct g_raid_softc *sc;
615 	struct g_raid_event *ep;
616 	int error;
617 
618 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
619 		sc = ((struct g_raid_volume *)arg)->v_softc;
620 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
621 		sc = ((struct g_raid_disk *)arg)->d_softc;
622 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
623 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
624 	} else {
625 		sc = arg;
626 	}
627 	ep = malloc(sizeof(*ep), M_RAID,
628 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
629 	if (ep == NULL)
630 		return (ENOMEM);
631 	ep->e_tgt = arg;
632 	ep->e_event = event;
633 	ep->e_flags = flags;
634 	ep->e_error = 0;
635 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
636 	mtx_lock(&sc->sc_queue_mtx);
637 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
638 	mtx_unlock(&sc->sc_queue_mtx);
639 	wakeup(sc);
640 
641 	if ((flags & G_RAID_EVENT_WAIT) == 0)
642 		return (0);
643 
644 	sx_assert(&sc->sc_lock, SX_XLOCKED);
645 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
646 	sx_xunlock(&sc->sc_lock);
647 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
648 		mtx_lock(&sc->sc_queue_mtx);
649 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
650 		    hz * 5);
651 	}
652 	error = ep->e_error;
653 	g_raid_event_free(ep);
654 	sx_xlock(&sc->sc_lock);
655 	return (error);
656 }
657 
658 static void
659 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
660 {
661 	struct g_raid_event *ep, *tmpep;
662 
663 	sx_assert(&sc->sc_lock, SX_XLOCKED);
664 
665 	mtx_lock(&sc->sc_queue_mtx);
666 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
667 		if (ep->e_tgt != tgt)
668 			continue;
669 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
670 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
671 			g_raid_event_free(ep);
672 		else {
673 			ep->e_error = ECANCELED;
674 			wakeup(ep);
675 		}
676 	}
677 	mtx_unlock(&sc->sc_queue_mtx);
678 }
679 
680 static int
681 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
682 {
683 	struct g_raid_event *ep;
684 	int	res = 0;
685 
686 	sx_assert(&sc->sc_lock, SX_XLOCKED);
687 
688 	mtx_lock(&sc->sc_queue_mtx);
689 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
690 		if (ep->e_tgt != tgt)
691 			continue;
692 		res = 1;
693 		break;
694 	}
695 	mtx_unlock(&sc->sc_queue_mtx);
696 	return (res);
697 }
698 
699 /*
700  * Return the number of disks in given state.
701  * If state is equal to -1, count all connected disks.
702  */
703 u_int
704 g_raid_ndisks(struct g_raid_softc *sc, int state)
705 {
706 	struct g_raid_disk *disk;
707 	u_int n;
708 
709 	sx_assert(&sc->sc_lock, SX_LOCKED);
710 
711 	n = 0;
712 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
713 		if (disk->d_state == state || state == -1)
714 			n++;
715 	}
716 	return (n);
717 }
718 
719 /*
720  * Return the number of subdisks in given state.
721  * If state is equal to -1, count all connected disks.
722  */
723 u_int
724 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
725 {
726 	struct g_raid_subdisk *subdisk;
727 	struct g_raid_softc *sc;
728 	u_int i, n ;
729 
730 	sc = vol->v_softc;
731 	sx_assert(&sc->sc_lock, SX_LOCKED);
732 
733 	n = 0;
734 	for (i = 0; i < vol->v_disks_count; i++) {
735 		subdisk = &vol->v_subdisks[i];
736 		if ((state == -1 &&
737 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
738 		    subdisk->sd_state == state)
739 			n++;
740 	}
741 	return (n);
742 }
743 
744 /*
745  * Return the first subdisk in given state.
746  * If state is equal to -1, then the first connected disks.
747  */
748 struct g_raid_subdisk *
749 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
750 {
751 	struct g_raid_subdisk *sd;
752 	struct g_raid_softc *sc;
753 	u_int i;
754 
755 	sc = vol->v_softc;
756 	sx_assert(&sc->sc_lock, SX_LOCKED);
757 
758 	for (i = 0; i < vol->v_disks_count; i++) {
759 		sd = &vol->v_subdisks[i];
760 		if ((state == -1 &&
761 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
762 		    sd->sd_state == state)
763 			return (sd);
764 	}
765 	return (NULL);
766 }
767 
768 struct g_consumer *
769 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
770 {
771 	struct g_consumer *cp;
772 	struct g_provider *pp;
773 
774 	g_topology_assert();
775 
776 	if (strncmp(name, "/dev/", 5) == 0)
777 		name += 5;
778 	pp = g_provider_by_name(name);
779 	if (pp == NULL)
780 		return (NULL);
781 	cp = g_new_consumer(sc->sc_geom);
782 	cp->flags |= G_CF_DIRECT_RECEIVE;
783 	if (g_attach(cp, pp) != 0) {
784 		g_destroy_consumer(cp);
785 		return (NULL);
786 	}
787 	if (g_access(cp, 1, 1, 1) != 0) {
788 		g_detach(cp);
789 		g_destroy_consumer(cp);
790 		return (NULL);
791 	}
792 	return (cp);
793 }
794 
795 static u_int
796 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
797 {
798 	struct bio *bp;
799 	u_int nreqs = 0;
800 
801 	mtx_lock(&sc->sc_queue_mtx);
802 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
803 		if (bp->bio_from == cp)
804 			nreqs++;
805 	}
806 	mtx_unlock(&sc->sc_queue_mtx);
807 	return (nreqs);
808 }
809 
810 u_int
811 g_raid_nopens(struct g_raid_softc *sc)
812 {
813 	struct g_raid_volume *vol;
814 	u_int opens;
815 
816 	opens = 0;
817 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
818 		if (vol->v_provider_open != 0)
819 			opens++;
820 	}
821 	return (opens);
822 }
823 
824 static int
825 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
826 {
827 
828 	if (cp->index > 0) {
829 		G_RAID_DEBUG1(2, sc,
830 		    "I/O requests for %s exist, can't destroy it now.",
831 		    cp->provider->name);
832 		return (1);
833 	}
834 	if (g_raid_nrequests(sc, cp) > 0) {
835 		G_RAID_DEBUG1(2, sc,
836 		    "I/O requests for %s in queue, can't destroy it now.",
837 		    cp->provider->name);
838 		return (1);
839 	}
840 	return (0);
841 }
842 
843 static void
844 g_raid_destroy_consumer(void *arg, int flags __unused)
845 {
846 	struct g_consumer *cp;
847 
848 	g_topology_assert();
849 
850 	cp = arg;
851 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
852 	g_detach(cp);
853 	g_destroy_consumer(cp);
854 }
855 
856 void
857 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
858 {
859 	struct g_provider *pp;
860 	int retaste_wait;
861 
862 	g_topology_assert_not();
863 
864 	g_topology_lock();
865 	cp->private = NULL;
866 	if (g_raid_consumer_is_busy(sc, cp))
867 		goto out;
868 	pp = cp->provider;
869 	retaste_wait = 0;
870 	if (cp->acw == 1) {
871 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
872 			retaste_wait = 1;
873 	}
874 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
875 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
876 	if (retaste_wait) {
877 		/*
878 		 * After retaste event was send (inside g_access()), we can send
879 		 * event to detach and destroy consumer.
880 		 * A class, which has consumer to the given provider connected
881 		 * will not receive retaste event for the provider.
882 		 * This is the way how I ignore retaste events when I close
883 		 * consumers opened for write: I detach and destroy consumer
884 		 * after retaste event is sent.
885 		 */
886 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
887 		goto out;
888 	}
889 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
890 	g_detach(cp);
891 	g_destroy_consumer(cp);
892 out:
893 	g_topology_unlock();
894 }
895 
896 static void
897 g_raid_orphan(struct g_consumer *cp)
898 {
899 	struct g_raid_disk *disk;
900 
901 	g_topology_assert();
902 
903 	disk = cp->private;
904 	if (disk == NULL)
905 		return;
906 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
907 	    G_RAID_EVENT_DISK);
908 }
909 
910 static void
911 g_raid_clean(struct g_raid_volume *vol, int acw)
912 {
913 	struct g_raid_softc *sc;
914 	int timeout;
915 
916 	sc = vol->v_softc;
917 	g_topology_assert_not();
918 	sx_assert(&sc->sc_lock, SX_XLOCKED);
919 
920 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
921 //		return;
922 	if (!vol->v_dirty)
923 		return;
924 	if (vol->v_writes > 0)
925 		return;
926 	if (acw > 0 || (acw == -1 &&
927 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
928 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
929 		if (!g_raid_shutdown && timeout > 0)
930 			return;
931 	}
932 	vol->v_dirty = 0;
933 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
934 	    vol->v_name);
935 	g_raid_write_metadata(sc, vol, NULL, NULL);
936 }
937 
938 static void
939 g_raid_dirty(struct g_raid_volume *vol)
940 {
941 	struct g_raid_softc *sc;
942 
943 	sc = vol->v_softc;
944 	g_topology_assert_not();
945 	sx_assert(&sc->sc_lock, SX_XLOCKED);
946 
947 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
948 //		return;
949 	vol->v_dirty = 1;
950 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
951 	    vol->v_name);
952 	g_raid_write_metadata(sc, vol, NULL, NULL);
953 }
954 
955 void
956 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
957 {
958 	struct g_raid_volume *vol;
959 	struct g_raid_subdisk *sd;
960 	struct bio_queue_head queue;
961 	struct bio *cbp;
962 	int i;
963 
964 	vol = tr->tro_volume;
965 
966 	/*
967 	 * Allocate all bios before sending any request, so we can return
968 	 * ENOMEM in nice and clean way.
969 	 */
970 	bioq_init(&queue);
971 	for (i = 0; i < vol->v_disks_count; i++) {
972 		sd = &vol->v_subdisks[i];
973 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
974 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
975 			continue;
976 		cbp = g_clone_bio(bp);
977 		if (cbp == NULL)
978 			goto failure;
979 		cbp->bio_caller1 = sd;
980 		bioq_insert_tail(&queue, cbp);
981 	}
982 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
983 		sd = cbp->bio_caller1;
984 		cbp->bio_caller1 = NULL;
985 		g_raid_subdisk_iostart(sd, cbp);
986 	}
987 	return;
988 failure:
989 	while ((cbp = bioq_takefirst(&queue)) != NULL)
990 		g_destroy_bio(cbp);
991 	if (bp->bio_error == 0)
992 		bp->bio_error = ENOMEM;
993 	g_raid_iodone(bp, bp->bio_error);
994 }
995 
996 static void
997 g_raid_tr_kerneldump_common_done(struct bio *bp)
998 {
999 
1000 	bp->bio_flags |= BIO_DONE;
1001 }
1002 
1003 int
1004 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1005     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1006 {
1007 	struct g_raid_softc *sc;
1008 	struct g_raid_volume *vol;
1009 	struct bio bp;
1010 
1011 	vol = tr->tro_volume;
1012 	sc = vol->v_softc;
1013 
1014 	g_reset_bio(&bp);
1015 	bp.bio_cmd = BIO_WRITE;
1016 	bp.bio_done = g_raid_tr_kerneldump_common_done;
1017 	bp.bio_attribute = NULL;
1018 	bp.bio_offset = offset;
1019 	bp.bio_length = length;
1020 	bp.bio_data = virtual;
1021 	bp.bio_to = vol->v_provider;
1022 
1023 	g_raid_start(&bp);
1024 	while (!(bp.bio_flags & BIO_DONE)) {
1025 		G_RAID_DEBUG1(4, sc, "Poll...");
1026 		g_raid_poll(sc);
1027 		DELAY(10);
1028 	}
1029 
1030 	return (bp.bio_error != 0 ? EIO : 0);
1031 }
1032 
1033 static int
1034 g_raid_dump(void *arg,
1035     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1036 {
1037 	struct g_raid_volume *vol;
1038 	int error;
1039 
1040 	vol = (struct g_raid_volume *)arg;
1041 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1042 	    (long long unsigned)offset, (long long unsigned)length);
1043 
1044 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1045 	    virtual, physical, 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 		break;
1114 	case BIO_GETATTR:
1115 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1116 			g_raid_candelete(sc, bp);
1117 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1118 			g_raid_kerneldump(sc, bp);
1119 		else
1120 			g_io_deliver(bp, EOPNOTSUPP);
1121 		return;
1122 	default:
1123 		g_io_deliver(bp, EOPNOTSUPP);
1124 		return;
1125 	}
1126 	mtx_lock(&sc->sc_queue_mtx);
1127 	bioq_insert_tail(&sc->sc_queue, bp);
1128 	mtx_unlock(&sc->sc_queue_mtx);
1129 	if (!dumping) {
1130 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1131 		wakeup(sc);
1132 	}
1133 }
1134 
1135 static int
1136 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1137 {
1138 	/*
1139 	 * 5 cases:
1140 	 * (1) bp entirely below NO
1141 	 * (2) bp entirely above NO
1142 	 * (3) bp start below, but end in range YES
1143 	 * (4) bp entirely within YES
1144 	 * (5) bp starts within, ends above YES
1145 	 *
1146 	 * lock range 10-19 (offset 10 length 10)
1147 	 * (1) 1-5: first if kicks it out
1148 	 * (2) 30-35: second if kicks it out
1149 	 * (3) 5-15: passes both ifs
1150 	 * (4) 12-14: passes both ifs
1151 	 * (5) 19-20: passes both
1152 	 */
1153 	off_t lend = lstart + len - 1;
1154 	off_t bstart = bp->bio_offset;
1155 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1156 
1157 	if (bend < lstart)
1158 		return (0);
1159 	if (lend < bstart)
1160 		return (0);
1161 	return (1);
1162 }
1163 
1164 static int
1165 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1166 {
1167 	struct g_raid_lock *lp;
1168 
1169 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1170 
1171 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1172 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1173 			return (1);
1174 	}
1175 	return (0);
1176 }
1177 
1178 static void
1179 g_raid_start_request(struct bio *bp)
1180 {
1181 	struct g_raid_softc *sc;
1182 	struct g_raid_volume *vol;
1183 
1184 	sc = bp->bio_to->geom->softc;
1185 	sx_assert(&sc->sc_lock, SX_LOCKED);
1186 	vol = bp->bio_to->private;
1187 
1188 	/*
1189 	 * Check to see if this item is in a locked range.  If so,
1190 	 * queue it to our locked queue and return.  We'll requeue
1191 	 * it when the range is unlocked.  Internal I/O for the
1192 	 * rebuild/rescan/recovery process is excluded from this
1193 	 * check so we can actually do the recovery.
1194 	 */
1195 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1196 	    g_raid_is_in_locked_range(vol, bp)) {
1197 		G_RAID_LOGREQ(3, bp, "Defer request.");
1198 		bioq_insert_tail(&vol->v_locked, bp);
1199 		return;
1200 	}
1201 
1202 	/*
1203 	 * If we're actually going to do the write/delete, then
1204 	 * update the idle stats for the volume.
1205 	 */
1206 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1207 		if (!vol->v_dirty)
1208 			g_raid_dirty(vol);
1209 		vol->v_writes++;
1210 	}
1211 
1212 	/*
1213 	 * Put request onto inflight queue, so we can check if new
1214 	 * synchronization requests don't collide with it.  Then tell
1215 	 * the transformation layer to start the I/O.
1216 	 */
1217 	bioq_insert_tail(&vol->v_inflight, bp);
1218 	G_RAID_LOGREQ(4, bp, "Request started");
1219 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1220 }
1221 
1222 static void
1223 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1224 {
1225 	off_t off, len;
1226 	struct bio *nbp;
1227 	struct g_raid_lock *lp;
1228 
1229 	vol->v_pending_lock = 0;
1230 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1231 		if (lp->l_pending) {
1232 			off = lp->l_offset;
1233 			len = lp->l_length;
1234 			lp->l_pending = 0;
1235 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1236 				if (g_raid_bio_overlaps(nbp, off, len))
1237 					lp->l_pending++;
1238 			}
1239 			if (lp->l_pending) {
1240 				vol->v_pending_lock = 1;
1241 				G_RAID_DEBUG1(4, vol->v_softc,
1242 				    "Deferred lock(%jd, %jd) has %d pending",
1243 				    (intmax_t)off, (intmax_t)(off + len),
1244 				    lp->l_pending);
1245 				continue;
1246 			}
1247 			G_RAID_DEBUG1(4, vol->v_softc,
1248 			    "Deferred lock of %jd to %jd completed",
1249 			    (intmax_t)off, (intmax_t)(off + len));
1250 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1251 		}
1252 	}
1253 }
1254 
1255 void
1256 g_raid_iodone(struct bio *bp, int error)
1257 {
1258 	struct g_raid_softc *sc;
1259 	struct g_raid_volume *vol;
1260 
1261 	sc = bp->bio_to->geom->softc;
1262 	sx_assert(&sc->sc_lock, SX_LOCKED);
1263 	vol = bp->bio_to->private;
1264 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1265 
1266 	/* Update stats if we done write/delete. */
1267 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1268 		vol->v_writes--;
1269 		vol->v_last_write = time_uptime;
1270 	}
1271 
1272 	bioq_remove(&vol->v_inflight, bp);
1273 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1274 		g_raid_finish_with_locked_ranges(vol, bp);
1275 	getmicrouptime(&vol->v_last_done);
1276 	g_io_deliver(bp, error);
1277 }
1278 
1279 int
1280 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1281     struct bio *ignore, void *argp)
1282 {
1283 	struct g_raid_softc *sc;
1284 	struct g_raid_lock *lp;
1285 	struct bio *bp;
1286 
1287 	sc = vol->v_softc;
1288 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1289 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1290 	lp->l_offset = off;
1291 	lp->l_length = len;
1292 	lp->l_callback_arg = argp;
1293 
1294 	lp->l_pending = 0;
1295 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1296 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1297 			lp->l_pending++;
1298 	}
1299 
1300 	/*
1301 	 * If there are any writes that are pending, we return EBUSY.  All
1302 	 * callers will have to wait until all pending writes clear.
1303 	 */
1304 	if (lp->l_pending > 0) {
1305 		vol->v_pending_lock = 1;
1306 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1307 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1308 		return (EBUSY);
1309 	}
1310 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1311 	    (intmax_t)off, (intmax_t)(off+len));
1312 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1313 	return (0);
1314 }
1315 
1316 int
1317 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1318 {
1319 	struct g_raid_lock *lp;
1320 	struct g_raid_softc *sc;
1321 	struct bio *bp;
1322 
1323 	sc = vol->v_softc;
1324 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1325 		if (lp->l_offset == off && lp->l_length == len) {
1326 			LIST_REMOVE(lp, l_next);
1327 			/* XXX
1328 			 * Right now we just put them all back on the queue
1329 			 * and hope for the best.  We hope this because any
1330 			 * locked ranges will go right back on this list
1331 			 * when the worker thread runs.
1332 			 * XXX
1333 			 */
1334 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1335 			    (intmax_t)lp->l_offset,
1336 			    (intmax_t)(lp->l_offset+lp->l_length));
1337 			mtx_lock(&sc->sc_queue_mtx);
1338 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1339 				bioq_insert_tail(&sc->sc_queue, bp);
1340 			mtx_unlock(&sc->sc_queue_mtx);
1341 			free(lp, M_RAID);
1342 			return (0);
1343 		}
1344 	}
1345 	return (EINVAL);
1346 }
1347 
1348 void
1349 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1350 {
1351 	struct g_consumer *cp;
1352 	struct g_raid_disk *disk, *tdisk;
1353 
1354 	bp->bio_caller1 = sd;
1355 
1356 	/*
1357 	 * Make sure that the disk is present. Generally it is a task of
1358 	 * transformation layers to not send requests to absent disks, but
1359 	 * it is better to be safe and report situation then sorry.
1360 	 */
1361 	if (sd->sd_disk == NULL) {
1362 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1363 nodisk:
1364 		bp->bio_from = NULL;
1365 		bp->bio_to = NULL;
1366 		bp->bio_error = ENXIO;
1367 		g_raid_disk_done(bp);
1368 		return;
1369 	}
1370 	disk = sd->sd_disk;
1371 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1372 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1373 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1374 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1375 		goto nodisk;
1376 	}
1377 
1378 	cp = disk->d_consumer;
1379 	bp->bio_from = cp;
1380 	bp->bio_to = cp->provider;
1381 	cp->index++;
1382 
1383 	/* Update average disks load. */
1384 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1385 		if (tdisk->d_consumer == NULL)
1386 			tdisk->d_load = 0;
1387 		else
1388 			tdisk->d_load = (tdisk->d_consumer->index *
1389 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1390 	}
1391 
1392 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1393 	if (dumping) {
1394 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1395 		if (bp->bio_cmd == BIO_WRITE) {
1396 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1397 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1398 		} else
1399 			bp->bio_error = EOPNOTSUPP;
1400 		g_raid_disk_done(bp);
1401 	} else {
1402 		bp->bio_done = g_raid_disk_done;
1403 		bp->bio_offset += sd->sd_offset;
1404 		G_RAID_LOGREQ(3, bp, "Sending request.");
1405 		g_io_request(bp, cp);
1406 	}
1407 }
1408 
1409 int
1410 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1411     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1412 {
1413 
1414 	if (sd->sd_disk == NULL)
1415 		return (ENXIO);
1416 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1417 		return (EOPNOTSUPP);
1418 	return (dump_write(&sd->sd_disk->d_kd.di,
1419 	    virtual, physical,
1420 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1421 	    length));
1422 }
1423 
1424 static void
1425 g_raid_disk_done(struct bio *bp)
1426 {
1427 	struct g_raid_softc *sc;
1428 	struct g_raid_subdisk *sd;
1429 
1430 	sd = bp->bio_caller1;
1431 	sc = sd->sd_softc;
1432 	mtx_lock(&sc->sc_queue_mtx);
1433 	bioq_insert_tail(&sc->sc_queue, bp);
1434 	mtx_unlock(&sc->sc_queue_mtx);
1435 	if (!dumping)
1436 		wakeup(sc);
1437 }
1438 
1439 static void
1440 g_raid_disk_done_request(struct bio *bp)
1441 {
1442 	struct g_raid_softc *sc;
1443 	struct g_raid_disk *disk;
1444 	struct g_raid_subdisk *sd;
1445 	struct g_raid_volume *vol;
1446 
1447 	g_topology_assert_not();
1448 
1449 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1450 	sd = bp->bio_caller1;
1451 	sc = sd->sd_softc;
1452 	vol = sd->sd_volume;
1453 	if (bp->bio_from != NULL) {
1454 		bp->bio_from->index--;
1455 		disk = bp->bio_from->private;
1456 		if (disk == NULL)
1457 			g_raid_kill_consumer(sc, bp->bio_from);
1458 	}
1459 	bp->bio_offset -= sd->sd_offset;
1460 
1461 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1462 }
1463 
1464 static void
1465 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1466 {
1467 
1468 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1469 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1470 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1471 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1472 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1473 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1474 	else
1475 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1476 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1477 		KASSERT(ep->e_error == 0,
1478 		    ("Error cannot be handled."));
1479 		g_raid_event_free(ep);
1480 	} else {
1481 		ep->e_flags |= G_RAID_EVENT_DONE;
1482 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1483 		mtx_lock(&sc->sc_queue_mtx);
1484 		wakeup(ep);
1485 		mtx_unlock(&sc->sc_queue_mtx);
1486 	}
1487 }
1488 
1489 /*
1490  * Worker thread.
1491  */
1492 static void
1493 g_raid_worker(void *arg)
1494 {
1495 	struct g_raid_softc *sc;
1496 	struct g_raid_event *ep;
1497 	struct g_raid_volume *vol;
1498 	struct bio *bp;
1499 	struct timeval now, t;
1500 	int timeout, rv;
1501 
1502 	sc = arg;
1503 	thread_lock(curthread);
1504 	sched_prio(curthread, PRIBIO);
1505 	thread_unlock(curthread);
1506 
1507 	sx_xlock(&sc->sc_lock);
1508 	for (;;) {
1509 		mtx_lock(&sc->sc_queue_mtx);
1510 		/*
1511 		 * First take a look at events.
1512 		 * This is important to handle events before any I/O requests.
1513 		 */
1514 		bp = NULL;
1515 		vol = NULL;
1516 		rv = 0;
1517 		ep = TAILQ_FIRST(&sc->sc_events);
1518 		if (ep != NULL)
1519 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1520 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1521 			;
1522 		else {
1523 			getmicrouptime(&now);
1524 			t = now;
1525 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1526 				if (bioq_first(&vol->v_inflight) == NULL &&
1527 				    vol->v_tr &&
1528 				    timevalcmp(&vol->v_last_done, &t, < ))
1529 					t = vol->v_last_done;
1530 			}
1531 			timevalsub(&t, &now);
1532 			timeout = g_raid_idle_threshold +
1533 			    t.tv_sec * 1000000 + t.tv_usec;
1534 			if (timeout > 0) {
1535 				/*
1536 				 * Two steps to avoid overflows at HZ=1000
1537 				 * and idle timeouts > 2.1s.  Some rounding
1538 				 * errors can occur, but they are < 1tick,
1539 				 * which is deemed to be close enough for
1540 				 * this purpose.
1541 				 */
1542 				int micpertic = 1000000 / hz;
1543 				timeout = (timeout + micpertic - 1) / micpertic;
1544 				sx_xunlock(&sc->sc_lock);
1545 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1546 				    PRIBIO | PDROP, "-", timeout);
1547 				sx_xlock(&sc->sc_lock);
1548 				goto process;
1549 			} else
1550 				rv = EWOULDBLOCK;
1551 		}
1552 		mtx_unlock(&sc->sc_queue_mtx);
1553 process:
1554 		if (ep != NULL) {
1555 			g_raid_handle_event(sc, ep);
1556 		} else if (bp != NULL) {
1557 			if (bp->bio_to != NULL &&
1558 			    bp->bio_to->geom == sc->sc_geom)
1559 				g_raid_start_request(bp);
1560 			else
1561 				g_raid_disk_done_request(bp);
1562 		} else if (rv == EWOULDBLOCK) {
1563 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1564 				g_raid_clean(vol, -1);
1565 				if (bioq_first(&vol->v_inflight) == NULL &&
1566 				    vol->v_tr) {
1567 					t.tv_sec = g_raid_idle_threshold / 1000000;
1568 					t.tv_usec = g_raid_idle_threshold % 1000000;
1569 					timevaladd(&t, &vol->v_last_done);
1570 					getmicrouptime(&now);
1571 					if (timevalcmp(&t, &now, <= )) {
1572 						G_RAID_TR_IDLE(vol->v_tr);
1573 						vol->v_last_done = now;
1574 					}
1575 				}
1576 			}
1577 		}
1578 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1579 			g_raid_destroy_node(sc, 1);	/* May not return. */
1580 	}
1581 }
1582 
1583 static void
1584 g_raid_poll(struct g_raid_softc *sc)
1585 {
1586 	struct g_raid_event *ep;
1587 	struct bio *bp;
1588 
1589 	sx_xlock(&sc->sc_lock);
1590 	mtx_lock(&sc->sc_queue_mtx);
1591 	/*
1592 	 * First take a look at events.
1593 	 * This is important to handle events before any I/O requests.
1594 	 */
1595 	ep = TAILQ_FIRST(&sc->sc_events);
1596 	if (ep != NULL) {
1597 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1598 		mtx_unlock(&sc->sc_queue_mtx);
1599 		g_raid_handle_event(sc, ep);
1600 		goto out;
1601 	}
1602 	bp = bioq_takefirst(&sc->sc_queue);
1603 	if (bp != NULL) {
1604 		mtx_unlock(&sc->sc_queue_mtx);
1605 		if (bp->bio_from == NULL ||
1606 		    bp->bio_from->geom != sc->sc_geom)
1607 			g_raid_start_request(bp);
1608 		else
1609 			g_raid_disk_done_request(bp);
1610 	}
1611 out:
1612 	sx_xunlock(&sc->sc_lock);
1613 }
1614 
1615 static void
1616 g_raid_launch_provider(struct g_raid_volume *vol)
1617 {
1618 	struct g_raid_disk *disk;
1619 	struct g_raid_subdisk *sd;
1620 	struct g_raid_softc *sc;
1621 	struct g_provider *pp;
1622 	char name[G_RAID_MAX_VOLUMENAME];
1623 	off_t off;
1624 	int i;
1625 
1626 	sc = vol->v_softc;
1627 	sx_assert(&sc->sc_lock, SX_LOCKED);
1628 
1629 	g_topology_lock();
1630 	/* Try to name provider with volume name. */
1631 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1632 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1633 	    g_provider_by_name(name) != NULL) {
1634 		/* Otherwise use sequential volume number. */
1635 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1636 	}
1637 
1638 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1639 	pp->flags |= G_PF_DIRECT_RECEIVE;
1640 	if (vol->v_tr->tro_class->trc_accept_unmapped) {
1641 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1642 		for (i = 0; i < vol->v_disks_count; i++) {
1643 			sd = &vol->v_subdisks[i];
1644 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1645 				continue;
1646 			if ((sd->sd_disk->d_consumer->provider->flags &
1647 			    G_PF_ACCEPT_UNMAPPED) == 0)
1648 				pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
1649 		}
1650 	}
1651 	pp->private = vol;
1652 	pp->mediasize = vol->v_mediasize;
1653 	pp->sectorsize = vol->v_sectorsize;
1654 	pp->stripesize = 0;
1655 	pp->stripeoffset = 0;
1656 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1657 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1658 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1659 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1660 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1661 		    disk->d_consumer != NULL &&
1662 		    disk->d_consumer->provider != NULL) {
1663 			pp->stripesize = disk->d_consumer->provider->stripesize;
1664 			off = disk->d_consumer->provider->stripeoffset;
1665 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1666 			if (off > 0)
1667 				pp->stripeoffset %= off;
1668 		}
1669 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1670 			pp->stripesize *= (vol->v_disks_count - 1);
1671 			pp->stripeoffset *= (vol->v_disks_count - 1);
1672 		}
1673 	} else
1674 		pp->stripesize = vol->v_strip_size;
1675 	vol->v_provider = pp;
1676 	g_error_provider(pp, 0);
1677 	g_topology_unlock();
1678 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1679 	    pp->name, vol->v_name);
1680 }
1681 
1682 static void
1683 g_raid_destroy_provider(struct g_raid_volume *vol)
1684 {
1685 	struct g_raid_softc *sc;
1686 	struct g_provider *pp;
1687 	struct bio *bp, *tmp;
1688 
1689 	g_topology_assert_not();
1690 	sc = vol->v_softc;
1691 	pp = vol->v_provider;
1692 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1693 
1694 	g_topology_lock();
1695 	g_error_provider(pp, ENXIO);
1696 	mtx_lock(&sc->sc_queue_mtx);
1697 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1698 		if (bp->bio_to != pp)
1699 			continue;
1700 		bioq_remove(&sc->sc_queue, bp);
1701 		g_io_deliver(bp, ENXIO);
1702 	}
1703 	mtx_unlock(&sc->sc_queue_mtx);
1704 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1705 	    pp->name, vol->v_name);
1706 	g_wither_provider(pp, ENXIO);
1707 	g_topology_unlock();
1708 	vol->v_provider = NULL;
1709 }
1710 
1711 /*
1712  * Update device state.
1713  */
1714 static int
1715 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1716 {
1717 	struct g_raid_softc *sc;
1718 
1719 	sc = vol->v_softc;
1720 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1721 
1722 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1723 	    g_raid_volume_event2str(event),
1724 	    vol->v_name);
1725 	switch (event) {
1726 	case G_RAID_VOLUME_E_DOWN:
1727 		if (vol->v_provider != NULL)
1728 			g_raid_destroy_provider(vol);
1729 		break;
1730 	case G_RAID_VOLUME_E_UP:
1731 		if (vol->v_provider == NULL)
1732 			g_raid_launch_provider(vol);
1733 		break;
1734 	case G_RAID_VOLUME_E_START:
1735 		if (vol->v_tr)
1736 			G_RAID_TR_START(vol->v_tr);
1737 		return (0);
1738 	default:
1739 		if (sc->sc_md)
1740 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1741 		return (0);
1742 	}
1743 
1744 	/* Manage root mount release. */
1745 	if (vol->v_starting) {
1746 		vol->v_starting = 0;
1747 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1748 		root_mount_rel(vol->v_rootmount);
1749 		vol->v_rootmount = NULL;
1750 	}
1751 	if (vol->v_stopping && vol->v_provider_open == 0)
1752 		g_raid_destroy_volume(vol);
1753 	return (0);
1754 }
1755 
1756 /*
1757  * Update subdisk state.
1758  */
1759 static int
1760 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1761 {
1762 	struct g_raid_softc *sc;
1763 	struct g_raid_volume *vol;
1764 
1765 	sc = sd->sd_softc;
1766 	vol = sd->sd_volume;
1767 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1768 
1769 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1770 	    g_raid_subdisk_event2str(event),
1771 	    vol->v_name, sd->sd_pos,
1772 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1773 	if (vol->v_tr)
1774 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1775 
1776 	return (0);
1777 }
1778 
1779 /*
1780  * Update disk state.
1781  */
1782 static int
1783 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1784 {
1785 	struct g_raid_softc *sc;
1786 
1787 	sc = disk->d_softc;
1788 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1789 
1790 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1791 	    g_raid_disk_event2str(event),
1792 	    g_raid_get_diskname(disk));
1793 
1794 	if (sc->sc_md)
1795 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1796 	return (0);
1797 }
1798 
1799 /*
1800  * Node event.
1801  */
1802 static int
1803 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1804 {
1805 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1806 
1807 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1808 	    g_raid_node_event2str(event));
1809 
1810 	if (event == G_RAID_NODE_E_WAKE)
1811 		return (0);
1812 	if (sc->sc_md)
1813 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1814 	return (0);
1815 }
1816 
1817 static int
1818 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1819 {
1820 	struct g_raid_volume *vol;
1821 	struct g_raid_softc *sc;
1822 	int dcw, opens, error = 0;
1823 
1824 	g_topology_assert();
1825 	sc = pp->geom->softc;
1826 	vol = pp->private;
1827 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1828 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1829 
1830 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1831 	    acr, acw, ace);
1832 	dcw = pp->acw + acw;
1833 
1834 	g_topology_unlock();
1835 	sx_xlock(&sc->sc_lock);
1836 	/* Deny new opens while dying. */
1837 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1838 		error = ENXIO;
1839 		goto out;
1840 	}
1841 	/* Deny write opens for read-only volumes. */
1842 	if (vol->v_read_only && acw > 0) {
1843 		error = EROFS;
1844 		goto out;
1845 	}
1846 	if (dcw == 0)
1847 		g_raid_clean(vol, dcw);
1848 	vol->v_provider_open += acr + acw + ace;
1849 	/* Handle delayed node destruction. */
1850 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1851 	    vol->v_provider_open == 0) {
1852 		/* Count open volumes. */
1853 		opens = g_raid_nopens(sc);
1854 		if (opens == 0) {
1855 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1856 			/* Wake up worker to make it selfdestruct. */
1857 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1858 		}
1859 	}
1860 	/* Handle open volume destruction. */
1861 	if (vol->v_stopping && vol->v_provider_open == 0)
1862 		g_raid_destroy_volume(vol);
1863 out:
1864 	sx_xunlock(&sc->sc_lock);
1865 	g_topology_lock();
1866 	return (error);
1867 }
1868 
1869 struct g_raid_softc *
1870 g_raid_create_node(struct g_class *mp,
1871     const char *name, struct g_raid_md_object *md)
1872 {
1873 	struct g_raid_softc *sc;
1874 	struct g_geom *gp;
1875 	int error;
1876 
1877 	g_topology_assert();
1878 	G_RAID_DEBUG(1, "Creating array %s.", name);
1879 
1880 	gp = g_new_geomf(mp, "%s", name);
1881 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1882 	gp->start = g_raid_start;
1883 	gp->orphan = g_raid_orphan;
1884 	gp->access = g_raid_access;
1885 	gp->dumpconf = g_raid_dumpconf;
1886 
1887 	sc->sc_md = md;
1888 	sc->sc_geom = gp;
1889 	sc->sc_flags = 0;
1890 	TAILQ_INIT(&sc->sc_volumes);
1891 	TAILQ_INIT(&sc->sc_disks);
1892 	sx_init(&sc->sc_lock, "graid:lock");
1893 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1894 	TAILQ_INIT(&sc->sc_events);
1895 	bioq_init(&sc->sc_queue);
1896 	gp->softc = sc;
1897 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1898 	    "g_raid %s", name);
1899 	if (error != 0) {
1900 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1901 		mtx_destroy(&sc->sc_queue_mtx);
1902 		sx_destroy(&sc->sc_lock);
1903 		g_destroy_geom(sc->sc_geom);
1904 		free(sc, M_RAID);
1905 		return (NULL);
1906 	}
1907 
1908 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1909 	return (sc);
1910 }
1911 
1912 struct g_raid_volume *
1913 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1914 {
1915 	struct g_raid_volume	*vol, *vol1;
1916 	int i;
1917 
1918 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1919 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1920 	vol->v_softc = sc;
1921 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1922 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1923 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1924 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1925 	vol->v_rotate_parity = 1;
1926 	bioq_init(&vol->v_inflight);
1927 	bioq_init(&vol->v_locked);
1928 	LIST_INIT(&vol->v_locks);
1929 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1930 		vol->v_subdisks[i].sd_softc = sc;
1931 		vol->v_subdisks[i].sd_volume = vol;
1932 		vol->v_subdisks[i].sd_pos = i;
1933 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1934 	}
1935 
1936 	/* Find free ID for this volume. */
1937 	g_topology_lock();
1938 	vol1 = vol;
1939 	if (id >= 0) {
1940 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1941 			if (vol1->v_global_id == id)
1942 				break;
1943 		}
1944 	}
1945 	if (vol1 != NULL) {
1946 		for (id = 0; ; id++) {
1947 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1948 				if (vol1->v_global_id == id)
1949 					break;
1950 			}
1951 			if (vol1 == NULL)
1952 				break;
1953 		}
1954 	}
1955 	vol->v_global_id = id;
1956 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1957 	g_topology_unlock();
1958 
1959 	/* Delay root mounting. */
1960 	vol->v_rootmount = root_mount_hold("GRAID");
1961 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1962 	vol->v_starting = 1;
1963 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1964 	return (vol);
1965 }
1966 
1967 struct g_raid_disk *
1968 g_raid_create_disk(struct g_raid_softc *sc)
1969 {
1970 	struct g_raid_disk	*disk;
1971 
1972 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1973 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1974 	disk->d_softc = sc;
1975 	disk->d_state = G_RAID_DISK_S_NONE;
1976 	TAILQ_INIT(&disk->d_subdisks);
1977 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1978 	return (disk);
1979 }
1980 
1981 int g_raid_start_volume(struct g_raid_volume *vol)
1982 {
1983 	struct g_raid_tr_class *class;
1984 	struct g_raid_tr_object *obj;
1985 	int status;
1986 
1987 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1988 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1989 		if (!class->trc_enable)
1990 			continue;
1991 		G_RAID_DEBUG1(2, vol->v_softc,
1992 		    "Tasting volume %s for %s transformation.",
1993 		    vol->v_name, class->name);
1994 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1995 		    M_WAITOK);
1996 		obj->tro_class = class;
1997 		obj->tro_volume = vol;
1998 		status = G_RAID_TR_TASTE(obj, vol);
1999 		if (status != G_RAID_TR_TASTE_FAIL)
2000 			break;
2001 		kobj_delete((kobj_t)obj, M_RAID);
2002 	}
2003 	if (class == NULL) {
2004 		G_RAID_DEBUG1(0, vol->v_softc,
2005 		    "No transformation module found for %s.",
2006 		    vol->v_name);
2007 		vol->v_tr = NULL;
2008 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2009 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2010 		    G_RAID_EVENT_VOLUME);
2011 		return (-1);
2012 	}
2013 	G_RAID_DEBUG1(2, vol->v_softc,
2014 	    "Transformation module %s chosen for %s.",
2015 	    class->name, vol->v_name);
2016 	vol->v_tr = obj;
2017 	return (0);
2018 }
2019 
2020 int
2021 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2022 {
2023 	struct g_raid_volume *vol, *tmpv;
2024 	struct g_raid_disk *disk, *tmpd;
2025 	int error = 0;
2026 
2027 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2028 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2029 		if (g_raid_destroy_volume(vol))
2030 			error = EBUSY;
2031 	}
2032 	if (error)
2033 		return (error);
2034 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2035 		if (g_raid_destroy_disk(disk))
2036 			error = EBUSY;
2037 	}
2038 	if (error)
2039 		return (error);
2040 	if (sc->sc_md) {
2041 		G_RAID_MD_FREE(sc->sc_md);
2042 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2043 		sc->sc_md = NULL;
2044 	}
2045 	if (sc->sc_geom != NULL) {
2046 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2047 		g_topology_lock();
2048 		sc->sc_geom->softc = NULL;
2049 		g_wither_geom(sc->sc_geom, ENXIO);
2050 		g_topology_unlock();
2051 		sc->sc_geom = NULL;
2052 	} else
2053 		G_RAID_DEBUG(1, "Array destroyed.");
2054 	if (worker) {
2055 		g_raid_event_cancel(sc, sc);
2056 		mtx_destroy(&sc->sc_queue_mtx);
2057 		sx_xunlock(&sc->sc_lock);
2058 		sx_destroy(&sc->sc_lock);
2059 		wakeup(&sc->sc_stopping);
2060 		free(sc, M_RAID);
2061 		curthread->td_pflags &= ~TDP_GEOM;
2062 		G_RAID_DEBUG(1, "Thread exiting.");
2063 		kproc_exit(0);
2064 	} else {
2065 		/* Wake up worker to make it selfdestruct. */
2066 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2067 	}
2068 	return (0);
2069 }
2070 
2071 int
2072 g_raid_destroy_volume(struct g_raid_volume *vol)
2073 {
2074 	struct g_raid_softc *sc;
2075 	struct g_raid_disk *disk;
2076 	int i;
2077 
2078 	sc = vol->v_softc;
2079 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2080 	vol->v_stopping = 1;
2081 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2082 		if (vol->v_tr) {
2083 			G_RAID_TR_STOP(vol->v_tr);
2084 			return (EBUSY);
2085 		} else
2086 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2087 	}
2088 	if (g_raid_event_check(sc, vol) != 0)
2089 		return (EBUSY);
2090 	if (vol->v_provider != NULL)
2091 		return (EBUSY);
2092 	if (vol->v_provider_open != 0)
2093 		return (EBUSY);
2094 	if (vol->v_tr) {
2095 		G_RAID_TR_FREE(vol->v_tr);
2096 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2097 		vol->v_tr = NULL;
2098 	}
2099 	if (vol->v_rootmount)
2100 		root_mount_rel(vol->v_rootmount);
2101 	g_topology_lock();
2102 	LIST_REMOVE(vol, v_global_next);
2103 	g_topology_unlock();
2104 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2105 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2106 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2107 		disk = vol->v_subdisks[i].sd_disk;
2108 		if (disk == NULL)
2109 			continue;
2110 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2111 	}
2112 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2113 	if (sc->sc_md)
2114 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2115 	g_raid_event_cancel(sc, vol);
2116 	free(vol, M_RAID);
2117 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2118 		/* Wake up worker to let it selfdestruct. */
2119 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2120 	}
2121 	return (0);
2122 }
2123 
2124 int
2125 g_raid_destroy_disk(struct g_raid_disk *disk)
2126 {
2127 	struct g_raid_softc *sc;
2128 	struct g_raid_subdisk *sd, *tmp;
2129 
2130 	sc = disk->d_softc;
2131 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2132 	if (disk->d_consumer) {
2133 		g_raid_kill_consumer(sc, disk->d_consumer);
2134 		disk->d_consumer = NULL;
2135 	}
2136 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2137 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2138 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2139 		    G_RAID_EVENT_SUBDISK);
2140 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2141 		sd->sd_disk = NULL;
2142 	}
2143 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2144 	if (sc->sc_md)
2145 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2146 	g_raid_event_cancel(sc, disk);
2147 	free(disk, M_RAID);
2148 	return (0);
2149 }
2150 
2151 int
2152 g_raid_destroy(struct g_raid_softc *sc, int how)
2153 {
2154 	int error, opens;
2155 
2156 	g_topology_assert_not();
2157 	if (sc == NULL)
2158 		return (ENXIO);
2159 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2160 
2161 	/* Count open volumes. */
2162 	opens = g_raid_nopens(sc);
2163 
2164 	/* React on some opened volumes. */
2165 	if (opens > 0) {
2166 		switch (how) {
2167 		case G_RAID_DESTROY_SOFT:
2168 			G_RAID_DEBUG1(1, sc,
2169 			    "%d volumes are still open.",
2170 			    opens);
2171 			sx_xunlock(&sc->sc_lock);
2172 			return (EBUSY);
2173 		case G_RAID_DESTROY_DELAYED:
2174 			G_RAID_DEBUG1(1, sc,
2175 			    "Array will be destroyed on last close.");
2176 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2177 			sx_xunlock(&sc->sc_lock);
2178 			return (EBUSY);
2179 		case G_RAID_DESTROY_HARD:
2180 			G_RAID_DEBUG1(1, sc,
2181 			    "%d volumes are still open.",
2182 			    opens);
2183 		}
2184 	}
2185 
2186 	/* Mark node for destruction. */
2187 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2188 	/* Wake up worker to let it selfdestruct. */
2189 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2190 	/* Sleep until node destroyed. */
2191 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2192 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2193 	return (error == EWOULDBLOCK ? EBUSY : 0);
2194 }
2195 
2196 static void
2197 g_raid_taste_orphan(struct g_consumer *cp)
2198 {
2199 
2200 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2201 	    cp->provider->name));
2202 }
2203 
2204 static struct g_geom *
2205 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2206 {
2207 	struct g_consumer *cp;
2208 	struct g_geom *gp, *geom;
2209 	struct g_raid_md_class *class;
2210 	struct g_raid_md_object *obj;
2211 	int status;
2212 
2213 	g_topology_assert();
2214 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2215 	if (!g_raid_enable)
2216 		return (NULL);
2217 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2218 
2219 	geom = NULL;
2220 	status = G_RAID_MD_TASTE_FAIL;
2221 	gp = g_new_geomf(mp, "raid:taste");
2222 	/*
2223 	 * This orphan function should be never called.
2224 	 */
2225 	gp->orphan = g_raid_taste_orphan;
2226 	cp = g_new_consumer(gp);
2227 	cp->flags |= G_CF_DIRECT_RECEIVE;
2228 	g_attach(cp, pp);
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 	g_destroy_consumer(cp);
2252 	g_destroy_geom(gp);
2253 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2254 	return (geom);
2255 }
2256 
2257 int
2258 g_raid_create_node_format(const char *format, struct gctl_req *req,
2259     struct g_geom **gp)
2260 {
2261 	struct g_raid_md_class *class;
2262 	struct g_raid_md_object *obj;
2263 	int status;
2264 
2265 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2266 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2267 		if (strcasecmp(class->name, format) == 0)
2268 			break;
2269 	}
2270 	if (class == NULL) {
2271 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2272 		return (G_RAID_MD_TASTE_FAIL);
2273 	}
2274 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2275 	    M_WAITOK);
2276 	obj->mdo_class = class;
2277 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2278 	if (status != G_RAID_MD_TASTE_NEW)
2279 		kobj_delete((kobj_t)obj, M_RAID);
2280 	return (status);
2281 }
2282 
2283 static int
2284 g_raid_destroy_geom(struct gctl_req *req __unused,
2285     struct g_class *mp __unused, struct g_geom *gp)
2286 {
2287 	struct g_raid_softc *sc;
2288 	int error;
2289 
2290 	g_topology_unlock();
2291 	sc = gp->softc;
2292 	sx_xlock(&sc->sc_lock);
2293 	g_cancel_event(sc);
2294 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2295 	g_topology_lock();
2296 	return (error);
2297 }
2298 
2299 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2300     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2301 {
2302 
2303 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2304 		return;
2305 	if (sc->sc_md)
2306 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2307 }
2308 
2309 void g_raid_fail_disk(struct g_raid_softc *sc,
2310     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2311 {
2312 
2313 	if (disk == NULL)
2314 		disk = sd->sd_disk;
2315 	if (disk == NULL) {
2316 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2317 		return;
2318 	}
2319 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2320 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2321 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2322 		return;
2323 	}
2324 	if (sc->sc_md)
2325 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2326 }
2327 
2328 static void
2329 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2330     struct g_consumer *cp, struct g_provider *pp)
2331 {
2332 	struct g_raid_softc *sc;
2333 	struct g_raid_volume *vol;
2334 	struct g_raid_subdisk *sd;
2335 	struct g_raid_disk *disk;
2336 	int i, s;
2337 
2338 	g_topology_assert();
2339 
2340 	sc = gp->softc;
2341 	if (sc == NULL)
2342 		return;
2343 	if (pp != NULL) {
2344 		vol = pp->private;
2345 		g_topology_unlock();
2346 		sx_xlock(&sc->sc_lock);
2347 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2348 		    sc->sc_md->mdo_class->name,
2349 		    g_raid_volume_level2str(vol->v_raid_level,
2350 		    vol->v_raid_level_qualifier));
2351 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2352 		    vol->v_name);
2353 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2354 		    g_raid_volume_level2str(vol->v_raid_level,
2355 		    vol->v_raid_level_qualifier));
2356 		sbuf_printf(sb,
2357 		    "%s<Transformation>%s</Transformation>\n", indent,
2358 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2359 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2360 		    vol->v_disks_count);
2361 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2362 		    vol->v_strip_size);
2363 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2364 		    g_raid_volume_state2str(vol->v_state));
2365 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2366 		    vol->v_dirty ? "Yes" : "No");
2367 		sbuf_printf(sb, "%s<Subdisks>", indent);
2368 		for (i = 0; i < vol->v_disks_count; i++) {
2369 			sd = &vol->v_subdisks[i];
2370 			if (sd->sd_disk != NULL &&
2371 			    sd->sd_disk->d_consumer != NULL) {
2372 				sbuf_printf(sb, "%s ",
2373 				    g_raid_get_diskname(sd->sd_disk));
2374 			} else {
2375 				sbuf_printf(sb, "NONE ");
2376 			}
2377 			sbuf_printf(sb, "(%s",
2378 			    g_raid_subdisk_state2str(sd->sd_state));
2379 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2380 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2381 				sbuf_printf(sb, " %d%%",
2382 				    (int)(sd->sd_rebuild_pos * 100 /
2383 				     sd->sd_size));
2384 			}
2385 			sbuf_printf(sb, ")");
2386 			if (i + 1 < vol->v_disks_count)
2387 				sbuf_printf(sb, ", ");
2388 		}
2389 		sbuf_printf(sb, "</Subdisks>\n");
2390 		sx_xunlock(&sc->sc_lock);
2391 		g_topology_lock();
2392 	} else if (cp != NULL) {
2393 		disk = cp->private;
2394 		if (disk == NULL)
2395 			return;
2396 		g_topology_unlock();
2397 		sx_xlock(&sc->sc_lock);
2398 		sbuf_printf(sb, "%s<State>%s", indent,
2399 		    g_raid_disk_state2str(disk->d_state));
2400 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2401 			sbuf_printf(sb, " (");
2402 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2403 				sbuf_printf(sb, "%s",
2404 				    g_raid_subdisk_state2str(sd->sd_state));
2405 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2406 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2407 					sbuf_printf(sb, " %d%%",
2408 					    (int)(sd->sd_rebuild_pos * 100 /
2409 					     sd->sd_size));
2410 				}
2411 				if (TAILQ_NEXT(sd, sd_next))
2412 					sbuf_printf(sb, ", ");
2413 			}
2414 			sbuf_printf(sb, ")");
2415 		}
2416 		sbuf_printf(sb, "</State>\n");
2417 		sbuf_printf(sb, "%s<Subdisks>", indent);
2418 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2419 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2420 			    sd->sd_volume->v_global_id,
2421 			    sd->sd_volume->v_name,
2422 			    sd->sd_pos, (uintmax_t)sd->sd_offset);
2423 			if (TAILQ_NEXT(sd, sd_next))
2424 				sbuf_printf(sb, ", ");
2425 		}
2426 		sbuf_printf(sb, "</Subdisks>\n");
2427 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2428 		    disk->d_read_errs);
2429 		sx_xunlock(&sc->sc_lock);
2430 		g_topology_lock();
2431 	} else {
2432 		g_topology_unlock();
2433 		sx_xlock(&sc->sc_lock);
2434 		if (sc->sc_md) {
2435 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2436 			    sc->sc_md->mdo_class->name);
2437 		}
2438 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2439 			s = 0xff;
2440 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2441 				if (vol->v_state < s)
2442 					s = vol->v_state;
2443 			}
2444 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2445 			    g_raid_volume_state2str(s));
2446 		}
2447 		sx_xunlock(&sc->sc_lock);
2448 		g_topology_lock();
2449 	}
2450 }
2451 
2452 static void
2453 g_raid_shutdown_post_sync(void *arg, int howto)
2454 {
2455 	struct g_class *mp;
2456 	struct g_geom *gp, *gp2;
2457 	struct g_raid_softc *sc;
2458 	struct g_raid_volume *vol;
2459 
2460 	mp = arg;
2461 	g_topology_lock();
2462 	g_raid_shutdown = 1;
2463 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2464 		if ((sc = gp->softc) == NULL)
2465 			continue;
2466 		g_topology_unlock();
2467 		sx_xlock(&sc->sc_lock);
2468 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2469 			g_raid_clean(vol, -1);
2470 		g_cancel_event(sc);
2471 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2472 		g_topology_lock();
2473 	}
2474 	g_topology_unlock();
2475 }
2476 
2477 static void
2478 g_raid_init(struct g_class *mp)
2479 {
2480 
2481 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2482 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2483 	if (g_raid_post_sync == NULL)
2484 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2485 	g_raid_started = 1;
2486 }
2487 
2488 static void
2489 g_raid_fini(struct g_class *mp)
2490 {
2491 
2492 	if (g_raid_post_sync != NULL)
2493 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2494 	g_raid_started = 0;
2495 }
2496 
2497 int
2498 g_raid_md_modevent(module_t mod, int type, void *arg)
2499 {
2500 	struct g_raid_md_class *class, *c, *nc;
2501 	int error;
2502 
2503 	error = 0;
2504 	class = arg;
2505 	switch (type) {
2506 	case MOD_LOAD:
2507 		c = LIST_FIRST(&g_raid_md_classes);
2508 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2509 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2510 		else {
2511 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2512 			    nc->mdc_priority < class->mdc_priority)
2513 				c = nc;
2514 			LIST_INSERT_AFTER(c, class, mdc_list);
2515 		}
2516 		if (g_raid_started)
2517 			g_retaste(&g_raid_class);
2518 		break;
2519 	case MOD_UNLOAD:
2520 		LIST_REMOVE(class, mdc_list);
2521 		break;
2522 	default:
2523 		error = EOPNOTSUPP;
2524 		break;
2525 	}
2526 
2527 	return (error);
2528 }
2529 
2530 int
2531 g_raid_tr_modevent(module_t mod, int type, void *arg)
2532 {
2533 	struct g_raid_tr_class *class, *c, *nc;
2534 	int error;
2535 
2536 	error = 0;
2537 	class = arg;
2538 	switch (type) {
2539 	case MOD_LOAD:
2540 		c = LIST_FIRST(&g_raid_tr_classes);
2541 		if (c == NULL || c->trc_priority > class->trc_priority)
2542 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2543 		else {
2544 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2545 			    nc->trc_priority < class->trc_priority)
2546 				c = nc;
2547 			LIST_INSERT_AFTER(c, class, trc_list);
2548 		}
2549 		break;
2550 	case MOD_UNLOAD:
2551 		LIST_REMOVE(class, trc_list);
2552 		break;
2553 	default:
2554 		error = EOPNOTSUPP;
2555 		break;
2556 	}
2557 
2558 	return (error);
2559 }
2560 
2561 /*
2562  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2563  * to reduce module priority, allowing submodules to register them first.
2564  */
2565 static moduledata_t g_raid_mod = {
2566 	"g_raid",
2567 	g_modevent,
2568 	&g_raid_class
2569 };
2570 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2571 MODULE_VERSION(geom_raid, 0);
2572