xref: /freebsd/sys/geom/raid/g_raid.c (revision f5f7c05209ca2c3748fd8b27c5e80ffad49120eb)
1 /*-
2  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/limits.h>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/bio.h>
38 #include <sys/sbuf.h>
39 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
41 #include <sys/eventhandler.h>
42 #include <vm/uma.h>
43 #include <geom/geom.h>
44 #include <sys/proc.h>
45 #include <sys/kthread.h>
46 #include <sys/sched.h>
47 #include <geom/raid/g_raid.h>
48 #include "g_raid_md_if.h"
49 #include "g_raid_tr_if.h"
50 
51 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
52 
53 SYSCTL_DECL(_kern_geom);
54 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
55 int g_raid_enable = 1;
56 TUNABLE_INT("kern.geom.raid.enable", &g_raid_enable);
57 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RW,
58     &g_raid_enable, 0, "Enable on-disk metadata taste");
59 u_int g_raid_aggressive_spare = 0;
60 TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare);
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW,
62     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug);
65 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0,
66     "Debug level");
67 int g_raid_read_err_thresh = 10;
68 TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh);
69 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW,
70     &g_raid_read_err_thresh, 0,
71     "Number of read errors equated to disk failure");
72 u_int g_raid_start_timeout = 30;
73 TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout);
74 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW,
75     &g_raid_start_timeout, 0,
76     "Time to wait for all array components");
77 static u_int g_raid_clean_time = 5;
78 TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time);
79 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW,
80     &g_raid_clean_time, 0, "Mark volume as clean when idling");
81 static u_int g_raid_disconnect_on_failure = 1;
82 TUNABLE_INT("kern.geom.raid.disconnect_on_failure",
83     &g_raid_disconnect_on_failure);
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW,
85     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
86 static u_int g_raid_name_format = 0;
87 TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format);
88 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW,
89     &g_raid_name_format, 0, "Providers name format.");
90 static u_int g_raid_idle_threshold = 1000000;
91 TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold);
92 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW,
93     &g_raid_idle_threshold, 1000000,
94     "Time in microseconds to consider a volume idle.");
95 
96 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
97 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
98 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
99 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
100 } while (0)
101 
102 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
103     LIST_HEAD_INITIALIZER(g_raid_md_classes);
104 
105 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
106     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
107 
108 LIST_HEAD(, g_raid_volume) g_raid_volumes =
109     LIST_HEAD_INITIALIZER(g_raid_volumes);
110 
111 static eventhandler_tag g_raid_post_sync = NULL;
112 static int g_raid_started = 0;
113 static int g_raid_shutdown = 0;
114 
115 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
116     struct g_geom *gp);
117 static g_taste_t g_raid_taste;
118 static void g_raid_init(struct g_class *mp);
119 static void g_raid_fini(struct g_class *mp);
120 
121 struct g_class g_raid_class = {
122 	.name = G_RAID_CLASS_NAME,
123 	.version = G_VERSION,
124 	.ctlreq = g_raid_ctl,
125 	.taste = g_raid_taste,
126 	.destroy_geom = g_raid_destroy_geom,
127 	.init = g_raid_init,
128 	.fini = g_raid_fini
129 };
130 
131 static void g_raid_destroy_provider(struct g_raid_volume *vol);
132 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
133 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
134 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
135 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
136 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
137     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
138 static void g_raid_start(struct bio *bp);
139 static void g_raid_start_request(struct bio *bp);
140 static void g_raid_disk_done(struct bio *bp);
141 static void g_raid_poll(struct g_raid_softc *sc);
142 
143 static const char *
144 g_raid_node_event2str(int event)
145 {
146 
147 	switch (event) {
148 	case G_RAID_NODE_E_WAKE:
149 		return ("WAKE");
150 	case G_RAID_NODE_E_START:
151 		return ("START");
152 	default:
153 		return ("INVALID");
154 	}
155 }
156 
157 const char *
158 g_raid_disk_state2str(int state)
159 {
160 
161 	switch (state) {
162 	case G_RAID_DISK_S_NONE:
163 		return ("NONE");
164 	case G_RAID_DISK_S_OFFLINE:
165 		return ("OFFLINE");
166 	case G_RAID_DISK_S_DISABLED:
167 		return ("DISABLED");
168 	case G_RAID_DISK_S_FAILED:
169 		return ("FAILED");
170 	case G_RAID_DISK_S_STALE_FAILED:
171 		return ("STALE_FAILED");
172 	case G_RAID_DISK_S_SPARE:
173 		return ("SPARE");
174 	case G_RAID_DISK_S_STALE:
175 		return ("STALE");
176 	case G_RAID_DISK_S_ACTIVE:
177 		return ("ACTIVE");
178 	default:
179 		return ("INVALID");
180 	}
181 }
182 
183 static const char *
184 g_raid_disk_event2str(int event)
185 {
186 
187 	switch (event) {
188 	case G_RAID_DISK_E_DISCONNECTED:
189 		return ("DISCONNECTED");
190 	default:
191 		return ("INVALID");
192 	}
193 }
194 
195 const char *
196 g_raid_subdisk_state2str(int state)
197 {
198 
199 	switch (state) {
200 	case G_RAID_SUBDISK_S_NONE:
201 		return ("NONE");
202 	case G_RAID_SUBDISK_S_FAILED:
203 		return ("FAILED");
204 	case G_RAID_SUBDISK_S_NEW:
205 		return ("NEW");
206 	case G_RAID_SUBDISK_S_REBUILD:
207 		return ("REBUILD");
208 	case G_RAID_SUBDISK_S_UNINITIALIZED:
209 		return ("UNINITIALIZED");
210 	case G_RAID_SUBDISK_S_STALE:
211 		return ("STALE");
212 	case G_RAID_SUBDISK_S_RESYNC:
213 		return ("RESYNC");
214 	case G_RAID_SUBDISK_S_ACTIVE:
215 		return ("ACTIVE");
216 	default:
217 		return ("INVALID");
218 	}
219 }
220 
221 static const char *
222 g_raid_subdisk_event2str(int event)
223 {
224 
225 	switch (event) {
226 	case G_RAID_SUBDISK_E_NEW:
227 		return ("NEW");
228 	case G_RAID_SUBDISK_E_FAILED:
229 		return ("FAILED");
230 	case G_RAID_SUBDISK_E_DISCONNECTED:
231 		return ("DISCONNECTED");
232 	default:
233 		return ("INVALID");
234 	}
235 }
236 
237 const char *
238 g_raid_volume_state2str(int state)
239 {
240 
241 	switch (state) {
242 	case G_RAID_VOLUME_S_STARTING:
243 		return ("STARTING");
244 	case G_RAID_VOLUME_S_BROKEN:
245 		return ("BROKEN");
246 	case G_RAID_VOLUME_S_DEGRADED:
247 		return ("DEGRADED");
248 	case G_RAID_VOLUME_S_SUBOPTIMAL:
249 		return ("SUBOPTIMAL");
250 	case G_RAID_VOLUME_S_OPTIMAL:
251 		return ("OPTIMAL");
252 	case G_RAID_VOLUME_S_UNSUPPORTED:
253 		return ("UNSUPPORTED");
254 	case G_RAID_VOLUME_S_STOPPED:
255 		return ("STOPPED");
256 	default:
257 		return ("INVALID");
258 	}
259 }
260 
261 static const char *
262 g_raid_volume_event2str(int event)
263 {
264 
265 	switch (event) {
266 	case G_RAID_VOLUME_E_UP:
267 		return ("UP");
268 	case G_RAID_VOLUME_E_DOWN:
269 		return ("DOWN");
270 	case G_RAID_VOLUME_E_START:
271 		return ("START");
272 	case G_RAID_VOLUME_E_STARTMD:
273 		return ("STARTMD");
274 	default:
275 		return ("INVALID");
276 	}
277 }
278 
279 const char *
280 g_raid_volume_level2str(int level, int qual)
281 {
282 
283 	switch (level) {
284 	case G_RAID_VOLUME_RL_RAID0:
285 		return ("RAID0");
286 	case G_RAID_VOLUME_RL_RAID1:
287 		return ("RAID1");
288 	case G_RAID_VOLUME_RL_RAID3:
289 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
290 			return ("RAID3-P0");
291 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
292 			return ("RAID3-PN");
293 		return ("RAID3");
294 	case G_RAID_VOLUME_RL_RAID4:
295 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
296 			return ("RAID4-P0");
297 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
298 			return ("RAID4-PN");
299 		return ("RAID4");
300 	case G_RAID_VOLUME_RL_RAID5:
301 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
302 			return ("RAID5-RA");
303 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
304 			return ("RAID5-RS");
305 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
306 			return ("RAID5-LA");
307 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
308 			return ("RAID5-LS");
309 		return ("RAID5");
310 	case G_RAID_VOLUME_RL_RAID6:
311 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
312 			return ("RAID6-RA");
313 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
314 			return ("RAID6-RS");
315 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
316 			return ("RAID6-LA");
317 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
318 			return ("RAID6-LS");
319 		return ("RAID6");
320 	case G_RAID_VOLUME_RL_RAIDMDF:
321 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
322 			return ("RAIDMDF-RA");
323 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
324 			return ("RAIDMDF-RS");
325 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
326 			return ("RAIDMDF-LA");
327 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
328 			return ("RAIDMDF-LS");
329 		return ("RAIDMDF");
330 	case G_RAID_VOLUME_RL_RAID1E:
331 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
332 			return ("RAID1E-A");
333 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
334 			return ("RAID1E-O");
335 		return ("RAID1E");
336 	case G_RAID_VOLUME_RL_SINGLE:
337 		return ("SINGLE");
338 	case G_RAID_VOLUME_RL_CONCAT:
339 		return ("CONCAT");
340 	case G_RAID_VOLUME_RL_RAID5E:
341 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
342 			return ("RAID5E-RA");
343 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
344 			return ("RAID5E-RS");
345 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
346 			return ("RAID5E-LA");
347 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
348 			return ("RAID5E-LS");
349 		return ("RAID5E");
350 	case G_RAID_VOLUME_RL_RAID5EE:
351 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
352 			return ("RAID5EE-RA");
353 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
354 			return ("RAID5EE-RS");
355 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
356 			return ("RAID5EE-LA");
357 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
358 			return ("RAID5EE-LS");
359 		return ("RAID5EE");
360 	case G_RAID_VOLUME_RL_RAID5R:
361 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
362 			return ("RAID5R-RA");
363 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
364 			return ("RAID5R-RS");
365 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
366 			return ("RAID5R-LA");
367 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
368 			return ("RAID5R-LS");
369 		return ("RAID5E");
370 	default:
371 		return ("UNKNOWN");
372 	}
373 }
374 
375 int
376 g_raid_volume_str2level(const char *str, int *level, int *qual)
377 {
378 
379 	*level = G_RAID_VOLUME_RL_UNKNOWN;
380 	*qual = G_RAID_VOLUME_RLQ_NONE;
381 	if (strcasecmp(str, "RAID0") == 0)
382 		*level = G_RAID_VOLUME_RL_RAID0;
383 	else if (strcasecmp(str, "RAID1") == 0)
384 		*level = G_RAID_VOLUME_RL_RAID1;
385 	else if (strcasecmp(str, "RAID3-P0") == 0) {
386 		*level = G_RAID_VOLUME_RL_RAID3;
387 		*qual = G_RAID_VOLUME_RLQ_R3P0;
388 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
389 		   strcasecmp(str, "RAID3") == 0) {
390 		*level = G_RAID_VOLUME_RL_RAID3;
391 		*qual = G_RAID_VOLUME_RLQ_R3PN;
392 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
393 		*level = G_RAID_VOLUME_RL_RAID4;
394 		*qual = G_RAID_VOLUME_RLQ_R4P0;
395 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
396 		   strcasecmp(str, "RAID4") == 0) {
397 		*level = G_RAID_VOLUME_RL_RAID4;
398 		*qual = G_RAID_VOLUME_RLQ_R4PN;
399 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
400 		*level = G_RAID_VOLUME_RL_RAID5;
401 		*qual = G_RAID_VOLUME_RLQ_R5RA;
402 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
403 		*level = G_RAID_VOLUME_RL_RAID5;
404 		*qual = G_RAID_VOLUME_RLQ_R5RS;
405 	} else if (strcasecmp(str, "RAID5") == 0 ||
406 		   strcasecmp(str, "RAID5-LA") == 0) {
407 		*level = G_RAID_VOLUME_RL_RAID5;
408 		*qual = G_RAID_VOLUME_RLQ_R5LA;
409 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
410 		*level = G_RAID_VOLUME_RL_RAID5;
411 		*qual = G_RAID_VOLUME_RLQ_R5LS;
412 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
413 		*level = G_RAID_VOLUME_RL_RAID6;
414 		*qual = G_RAID_VOLUME_RLQ_R6RA;
415 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
416 		*level = G_RAID_VOLUME_RL_RAID6;
417 		*qual = G_RAID_VOLUME_RLQ_R6RS;
418 	} else if (strcasecmp(str, "RAID6") == 0 ||
419 		   strcasecmp(str, "RAID6-LA") == 0) {
420 		*level = G_RAID_VOLUME_RL_RAID6;
421 		*qual = G_RAID_VOLUME_RLQ_R6LA;
422 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
423 		*level = G_RAID_VOLUME_RL_RAID6;
424 		*qual = G_RAID_VOLUME_RLQ_R6LS;
425 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
426 		*level = G_RAID_VOLUME_RL_RAIDMDF;
427 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
428 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
429 		*level = G_RAID_VOLUME_RL_RAIDMDF;
430 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
431 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
432 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
433 		*level = G_RAID_VOLUME_RL_RAIDMDF;
434 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
435 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
436 		*level = G_RAID_VOLUME_RL_RAIDMDF;
437 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
438 	} else if (strcasecmp(str, "RAID10") == 0 ||
439 		   strcasecmp(str, "RAID1E") == 0 ||
440 		   strcasecmp(str, "RAID1E-A") == 0) {
441 		*level = G_RAID_VOLUME_RL_RAID1E;
442 		*qual = G_RAID_VOLUME_RLQ_R1EA;
443 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
444 		*level = G_RAID_VOLUME_RL_RAID1E;
445 		*qual = G_RAID_VOLUME_RLQ_R1EO;
446 	} else if (strcasecmp(str, "SINGLE") == 0)
447 		*level = G_RAID_VOLUME_RL_SINGLE;
448 	else if (strcasecmp(str, "CONCAT") == 0)
449 		*level = G_RAID_VOLUME_RL_CONCAT;
450 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
451 		*level = G_RAID_VOLUME_RL_RAID5E;
452 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
453 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
454 		*level = G_RAID_VOLUME_RL_RAID5E;
455 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
456 	} else if (strcasecmp(str, "RAID5E") == 0 ||
457 		   strcasecmp(str, "RAID5E-LA") == 0) {
458 		*level = G_RAID_VOLUME_RL_RAID5E;
459 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
460 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
461 		*level = G_RAID_VOLUME_RL_RAID5E;
462 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
463 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
464 		*level = G_RAID_VOLUME_RL_RAID5EE;
465 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
466 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
467 		*level = G_RAID_VOLUME_RL_RAID5EE;
468 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
469 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
470 		   strcasecmp(str, "RAID5EE-LA") == 0) {
471 		*level = G_RAID_VOLUME_RL_RAID5EE;
472 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
473 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
474 		*level = G_RAID_VOLUME_RL_RAID5EE;
475 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
476 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
477 		*level = G_RAID_VOLUME_RL_RAID5R;
478 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
479 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
480 		*level = G_RAID_VOLUME_RL_RAID5R;
481 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
482 	} else if (strcasecmp(str, "RAID5R") == 0 ||
483 		   strcasecmp(str, "RAID5R-LA") == 0) {
484 		*level = G_RAID_VOLUME_RL_RAID5R;
485 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
486 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
487 		*level = G_RAID_VOLUME_RL_RAID5R;
488 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
489 	} else
490 		return (-1);
491 	return (0);
492 }
493 
494 const char *
495 g_raid_get_diskname(struct g_raid_disk *disk)
496 {
497 
498 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
499 		return ("[unknown]");
500 	return (disk->d_consumer->provider->name);
501 }
502 
503 void
504 g_raid_get_disk_info(struct g_raid_disk *disk)
505 {
506 	struct g_consumer *cp = disk->d_consumer;
507 	int error, len;
508 
509 	/* Read kernel dumping information. */
510 	disk->d_kd.offset = 0;
511 	disk->d_kd.length = OFF_MAX;
512 	len = sizeof(disk->d_kd);
513 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
514 	if (error)
515 		disk->d_kd.di.dumper = NULL;
516 	if (disk->d_kd.di.dumper == NULL)
517 		G_RAID_DEBUG1(2, disk->d_softc,
518 		    "Dumping not supported by %s: %d.",
519 		    cp->provider->name, error);
520 
521 	/* Read BIO_DELETE support. */
522 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
523 	if (error)
524 		disk->d_candelete = 0;
525 	if (!disk->d_candelete)
526 		G_RAID_DEBUG1(2, disk->d_softc,
527 		    "BIO_DELETE not supported by %s: %d.",
528 		    cp->provider->name, error);
529 }
530 
531 void
532 g_raid_report_disk_state(struct g_raid_disk *disk)
533 {
534 	struct g_raid_subdisk *sd;
535 	int len, state;
536 	uint32_t s;
537 
538 	if (disk->d_consumer == NULL)
539 		return;
540 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
541 		s = G_STATE_ACTIVE; /* XXX */
542 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
543 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
544 		s = G_STATE_FAILED;
545 	} else {
546 		state = G_RAID_SUBDISK_S_ACTIVE;
547 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
548 			if (sd->sd_state < state)
549 				state = sd->sd_state;
550 		}
551 		if (state == G_RAID_SUBDISK_S_FAILED)
552 			s = G_STATE_FAILED;
553 		else if (state == G_RAID_SUBDISK_S_NEW ||
554 		    state == G_RAID_SUBDISK_S_REBUILD)
555 			s = G_STATE_REBUILD;
556 		else if (state == G_RAID_SUBDISK_S_STALE ||
557 		    state == G_RAID_SUBDISK_S_RESYNC)
558 			s = G_STATE_RESYNC;
559 		else
560 			s = G_STATE_ACTIVE;
561 	}
562 	len = sizeof(s);
563 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
564 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
565 	    g_raid_get_diskname(disk), s);
566 }
567 
568 void
569 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
570 {
571 
572 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
573 	    g_raid_get_diskname(disk),
574 	    g_raid_disk_state2str(disk->d_state),
575 	    g_raid_disk_state2str(state));
576 	disk->d_state = state;
577 	g_raid_report_disk_state(disk);
578 }
579 
580 void
581 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
582 {
583 
584 	G_RAID_DEBUG1(0, sd->sd_softc,
585 	    "Subdisk %s:%d-%s state changed from %s to %s.",
586 	    sd->sd_volume->v_name, sd->sd_pos,
587 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
588 	    g_raid_subdisk_state2str(sd->sd_state),
589 	    g_raid_subdisk_state2str(state));
590 	sd->sd_state = state;
591 	if (sd->sd_disk)
592 		g_raid_report_disk_state(sd->sd_disk);
593 }
594 
595 void
596 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
597 {
598 
599 	G_RAID_DEBUG1(0, vol->v_softc,
600 	    "Volume %s state changed from %s to %s.",
601 	    vol->v_name,
602 	    g_raid_volume_state2str(vol->v_state),
603 	    g_raid_volume_state2str(state));
604 	vol->v_state = state;
605 }
606 
607 /*
608  * --- Events handling functions ---
609  * Events in geom_raid are used to maintain subdisks and volumes status
610  * from one thread to simplify locking.
611  */
612 static void
613 g_raid_event_free(struct g_raid_event *ep)
614 {
615 
616 	free(ep, M_RAID);
617 }
618 
619 int
620 g_raid_event_send(void *arg, int event, int flags)
621 {
622 	struct g_raid_softc *sc;
623 	struct g_raid_event *ep;
624 	int error;
625 
626 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
627 		sc = ((struct g_raid_volume *)arg)->v_softc;
628 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
629 		sc = ((struct g_raid_disk *)arg)->d_softc;
630 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
631 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
632 	} else {
633 		sc = arg;
634 	}
635 	ep = malloc(sizeof(*ep), M_RAID,
636 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
637 	if (ep == NULL)
638 		return (ENOMEM);
639 	ep->e_tgt = arg;
640 	ep->e_event = event;
641 	ep->e_flags = flags;
642 	ep->e_error = 0;
643 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
644 	mtx_lock(&sc->sc_queue_mtx);
645 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
646 	mtx_unlock(&sc->sc_queue_mtx);
647 	wakeup(sc);
648 
649 	if ((flags & G_RAID_EVENT_WAIT) == 0)
650 		return (0);
651 
652 	sx_assert(&sc->sc_lock, SX_XLOCKED);
653 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
654 	sx_xunlock(&sc->sc_lock);
655 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
656 		mtx_lock(&sc->sc_queue_mtx);
657 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
658 		    hz * 5);
659 	}
660 	error = ep->e_error;
661 	g_raid_event_free(ep);
662 	sx_xlock(&sc->sc_lock);
663 	return (error);
664 }
665 
666 static void
667 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
668 {
669 	struct g_raid_event *ep, *tmpep;
670 
671 	sx_assert(&sc->sc_lock, SX_XLOCKED);
672 
673 	mtx_lock(&sc->sc_queue_mtx);
674 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
675 		if (ep->e_tgt != tgt)
676 			continue;
677 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
678 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
679 			g_raid_event_free(ep);
680 		else {
681 			ep->e_error = ECANCELED;
682 			wakeup(ep);
683 		}
684 	}
685 	mtx_unlock(&sc->sc_queue_mtx);
686 }
687 
688 static int
689 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
690 {
691 	struct g_raid_event *ep;
692 	int	res = 0;
693 
694 	sx_assert(&sc->sc_lock, SX_XLOCKED);
695 
696 	mtx_lock(&sc->sc_queue_mtx);
697 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
698 		if (ep->e_tgt != tgt)
699 			continue;
700 		res = 1;
701 		break;
702 	}
703 	mtx_unlock(&sc->sc_queue_mtx);
704 	return (res);
705 }
706 
707 /*
708  * Return the number of disks in given state.
709  * If state is equal to -1, count all connected disks.
710  */
711 u_int
712 g_raid_ndisks(struct g_raid_softc *sc, int state)
713 {
714 	struct g_raid_disk *disk;
715 	u_int n;
716 
717 	sx_assert(&sc->sc_lock, SX_LOCKED);
718 
719 	n = 0;
720 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
721 		if (disk->d_state == state || state == -1)
722 			n++;
723 	}
724 	return (n);
725 }
726 
727 /*
728  * Return the number of subdisks in given state.
729  * If state is equal to -1, count all connected disks.
730  */
731 u_int
732 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
733 {
734 	struct g_raid_subdisk *subdisk;
735 	struct g_raid_softc *sc;
736 	u_int i, n ;
737 
738 	sc = vol->v_softc;
739 	sx_assert(&sc->sc_lock, SX_LOCKED);
740 
741 	n = 0;
742 	for (i = 0; i < vol->v_disks_count; i++) {
743 		subdisk = &vol->v_subdisks[i];
744 		if ((state == -1 &&
745 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
746 		    subdisk->sd_state == state)
747 			n++;
748 	}
749 	return (n);
750 }
751 
752 /*
753  * Return the first subdisk in given state.
754  * If state is equal to -1, then the first connected disks.
755  */
756 struct g_raid_subdisk *
757 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
758 {
759 	struct g_raid_subdisk *sd;
760 	struct g_raid_softc *sc;
761 	u_int i;
762 
763 	sc = vol->v_softc;
764 	sx_assert(&sc->sc_lock, SX_LOCKED);
765 
766 	for (i = 0; i < vol->v_disks_count; i++) {
767 		sd = &vol->v_subdisks[i];
768 		if ((state == -1 &&
769 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
770 		    sd->sd_state == state)
771 			return (sd);
772 	}
773 	return (NULL);
774 }
775 
776 struct g_consumer *
777 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
778 {
779 	struct g_consumer *cp;
780 	struct g_provider *pp;
781 
782 	g_topology_assert();
783 
784 	if (strncmp(name, "/dev/", 5) == 0)
785 		name += 5;
786 	pp = g_provider_by_name(name);
787 	if (pp == NULL)
788 		return (NULL);
789 	cp = g_new_consumer(sc->sc_geom);
790 	if (g_attach(cp, pp) != 0) {
791 		g_destroy_consumer(cp);
792 		return (NULL);
793 	}
794 	if (g_access(cp, 1, 1, 1) != 0) {
795 		g_detach(cp);
796 		g_destroy_consumer(cp);
797 		return (NULL);
798 	}
799 	return (cp);
800 }
801 
802 static u_int
803 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
804 {
805 	struct bio *bp;
806 	u_int nreqs = 0;
807 
808 	mtx_lock(&sc->sc_queue_mtx);
809 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
810 		if (bp->bio_from == cp)
811 			nreqs++;
812 	}
813 	mtx_unlock(&sc->sc_queue_mtx);
814 	return (nreqs);
815 }
816 
817 u_int
818 g_raid_nopens(struct g_raid_softc *sc)
819 {
820 	struct g_raid_volume *vol;
821 	u_int opens;
822 
823 	opens = 0;
824 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
825 		if (vol->v_provider_open != 0)
826 			opens++;
827 	}
828 	return (opens);
829 }
830 
831 static int
832 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
833 {
834 
835 	if (cp->index > 0) {
836 		G_RAID_DEBUG1(2, sc,
837 		    "I/O requests for %s exist, can't destroy it now.",
838 		    cp->provider->name);
839 		return (1);
840 	}
841 	if (g_raid_nrequests(sc, cp) > 0) {
842 		G_RAID_DEBUG1(2, sc,
843 		    "I/O requests for %s in queue, can't destroy it now.",
844 		    cp->provider->name);
845 		return (1);
846 	}
847 	return (0);
848 }
849 
850 static void
851 g_raid_destroy_consumer(void *arg, int flags __unused)
852 {
853 	struct g_consumer *cp;
854 
855 	g_topology_assert();
856 
857 	cp = arg;
858 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
859 	g_detach(cp);
860 	g_destroy_consumer(cp);
861 }
862 
863 void
864 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
865 {
866 	struct g_provider *pp;
867 	int retaste_wait;
868 
869 	g_topology_assert_not();
870 
871 	g_topology_lock();
872 	cp->private = NULL;
873 	if (g_raid_consumer_is_busy(sc, cp))
874 		goto out;
875 	pp = cp->provider;
876 	retaste_wait = 0;
877 	if (cp->acw == 1) {
878 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
879 			retaste_wait = 1;
880 	}
881 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
882 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
883 	if (retaste_wait) {
884 		/*
885 		 * After retaste event was send (inside g_access()), we can send
886 		 * event to detach and destroy consumer.
887 		 * A class, which has consumer to the given provider connected
888 		 * will not receive retaste event for the provider.
889 		 * This is the way how I ignore retaste events when I close
890 		 * consumers opened for write: I detach and destroy consumer
891 		 * after retaste event is sent.
892 		 */
893 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
894 		goto out;
895 	}
896 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
897 	g_detach(cp);
898 	g_destroy_consumer(cp);
899 out:
900 	g_topology_unlock();
901 }
902 
903 static void
904 g_raid_orphan(struct g_consumer *cp)
905 {
906 	struct g_raid_disk *disk;
907 
908 	g_topology_assert();
909 
910 	disk = cp->private;
911 	if (disk == NULL)
912 		return;
913 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
914 	    G_RAID_EVENT_DISK);
915 }
916 
917 static void
918 g_raid_clean(struct g_raid_volume *vol, int acw)
919 {
920 	struct g_raid_softc *sc;
921 	int timeout;
922 
923 	sc = vol->v_softc;
924 	g_topology_assert_not();
925 	sx_assert(&sc->sc_lock, SX_XLOCKED);
926 
927 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
928 //		return;
929 	if (!vol->v_dirty)
930 		return;
931 	if (vol->v_writes > 0)
932 		return;
933 	if (acw > 0 || (acw == -1 &&
934 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
935 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
936 		if (!g_raid_shutdown && timeout > 0)
937 			return;
938 	}
939 	vol->v_dirty = 0;
940 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
941 	    vol->v_name);
942 	g_raid_write_metadata(sc, vol, NULL, NULL);
943 }
944 
945 static void
946 g_raid_dirty(struct g_raid_volume *vol)
947 {
948 	struct g_raid_softc *sc;
949 
950 	sc = vol->v_softc;
951 	g_topology_assert_not();
952 	sx_assert(&sc->sc_lock, SX_XLOCKED);
953 
954 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
955 //		return;
956 	vol->v_dirty = 1;
957 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
958 	    vol->v_name);
959 	g_raid_write_metadata(sc, vol, NULL, NULL);
960 }
961 
962 void
963 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
964 {
965 	struct g_raid_softc *sc;
966 	struct g_raid_volume *vol;
967 	struct g_raid_subdisk *sd;
968 	struct bio_queue_head queue;
969 	struct bio *cbp;
970 	int i;
971 
972 	vol = tr->tro_volume;
973 	sc = vol->v_softc;
974 
975 	/*
976 	 * Allocate all bios before sending any request, so we can return
977 	 * ENOMEM in nice and clean way.
978 	 */
979 	bioq_init(&queue);
980 	for (i = 0; i < vol->v_disks_count; i++) {
981 		sd = &vol->v_subdisks[i];
982 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
983 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
984 			continue;
985 		cbp = g_clone_bio(bp);
986 		if (cbp == NULL)
987 			goto failure;
988 		cbp->bio_caller1 = sd;
989 		bioq_insert_tail(&queue, cbp);
990 	}
991 	for (cbp = bioq_first(&queue); cbp != NULL;
992 	    cbp = bioq_first(&queue)) {
993 		bioq_remove(&queue, cbp);
994 		sd = cbp->bio_caller1;
995 		cbp->bio_caller1 = NULL;
996 		g_raid_subdisk_iostart(sd, cbp);
997 	}
998 	return;
999 failure:
1000 	for (cbp = bioq_first(&queue); cbp != NULL;
1001 	    cbp = bioq_first(&queue)) {
1002 		bioq_remove(&queue, cbp);
1003 		g_destroy_bio(cbp);
1004 	}
1005 	if (bp->bio_error == 0)
1006 		bp->bio_error = ENOMEM;
1007 	g_raid_iodone(bp, bp->bio_error);
1008 }
1009 
1010 static void
1011 g_raid_tr_kerneldump_common_done(struct bio *bp)
1012 {
1013 
1014 	bp->bio_flags |= BIO_DONE;
1015 }
1016 
1017 int
1018 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1019     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1020 {
1021 	struct g_raid_softc *sc;
1022 	struct g_raid_volume *vol;
1023 	struct bio bp;
1024 
1025 	vol = tr->tro_volume;
1026 	sc = vol->v_softc;
1027 
1028 	bzero(&bp, sizeof(bp));
1029 	bp.bio_cmd = BIO_WRITE;
1030 	bp.bio_done = g_raid_tr_kerneldump_common_done;
1031 	bp.bio_attribute = NULL;
1032 	bp.bio_offset = offset;
1033 	bp.bio_length = length;
1034 	bp.bio_data = virtual;
1035 	bp.bio_to = vol->v_provider;
1036 
1037 	g_raid_start(&bp);
1038 	while (!(bp.bio_flags & BIO_DONE)) {
1039 		G_RAID_DEBUG1(4, sc, "Poll...");
1040 		g_raid_poll(sc);
1041 		DELAY(10);
1042 	}
1043 
1044 	return (bp.bio_error != 0 ? EIO : 0);
1045 }
1046 
1047 static int
1048 g_raid_dump(void *arg,
1049     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1050 {
1051 	struct g_raid_volume *vol;
1052 	int error;
1053 
1054 	vol = (struct g_raid_volume *)arg;
1055 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1056 	    (long long unsigned)offset, (long long unsigned)length);
1057 
1058 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1059 	    virtual, physical, offset, length);
1060 	return (error);
1061 }
1062 
1063 static void
1064 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1065 {
1066 	struct g_kerneldump *gkd;
1067 	struct g_provider *pp;
1068 	struct g_raid_volume *vol;
1069 
1070 	gkd = (struct g_kerneldump*)bp->bio_data;
1071 	pp = bp->bio_to;
1072 	vol = pp->private;
1073 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1074 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1075 	gkd->di.dumper = g_raid_dump;
1076 	gkd->di.priv = vol;
1077 	gkd->di.blocksize = vol->v_sectorsize;
1078 	gkd->di.maxiosize = DFLTPHYS;
1079 	gkd->di.mediaoffset = gkd->offset;
1080 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1081 		gkd->length = vol->v_mediasize - gkd->offset;
1082 	gkd->di.mediasize = gkd->length;
1083 	g_io_deliver(bp, 0);
1084 }
1085 
1086 static void
1087 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1088 {
1089 	struct g_provider *pp;
1090 	struct g_raid_volume *vol;
1091 	struct g_raid_subdisk *sd;
1092 	int *val;
1093 	int i;
1094 
1095 	val = (int *)bp->bio_data;
1096 	pp = bp->bio_to;
1097 	vol = pp->private;
1098 	*val = 0;
1099 	for (i = 0; i < vol->v_disks_count; i++) {
1100 		sd = &vol->v_subdisks[i];
1101 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1102 			continue;
1103 		if (sd->sd_disk->d_candelete) {
1104 			*val = 1;
1105 			break;
1106 		}
1107 	}
1108 	g_io_deliver(bp, 0);
1109 }
1110 
1111 static void
1112 g_raid_start(struct bio *bp)
1113 {
1114 	struct g_raid_softc *sc;
1115 
1116 	sc = bp->bio_to->geom->softc;
1117 	/*
1118 	 * If sc == NULL or there are no valid disks, provider's error
1119 	 * should be set and g_raid_start() should not be called at all.
1120 	 */
1121 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1122 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1123 //	    bp->bio_to->error, bp->bio_to->name));
1124 	G_RAID_LOGREQ(3, bp, "Request received.");
1125 
1126 	switch (bp->bio_cmd) {
1127 	case BIO_READ:
1128 	case BIO_WRITE:
1129 	case BIO_DELETE:
1130 	case BIO_FLUSH:
1131 		break;
1132 	case BIO_GETATTR:
1133 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1134 			g_raid_candelete(sc, bp);
1135 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1136 			g_raid_kerneldump(sc, bp);
1137 		else
1138 			g_io_deliver(bp, EOPNOTSUPP);
1139 		return;
1140 	default:
1141 		g_io_deliver(bp, EOPNOTSUPP);
1142 		return;
1143 	}
1144 	mtx_lock(&sc->sc_queue_mtx);
1145 	bioq_disksort(&sc->sc_queue, bp);
1146 	mtx_unlock(&sc->sc_queue_mtx);
1147 	if (!dumping) {
1148 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1149 		wakeup(sc);
1150 	}
1151 }
1152 
1153 static int
1154 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1155 {
1156 	/*
1157 	 * 5 cases:
1158 	 * (1) bp entirely below NO
1159 	 * (2) bp entirely above NO
1160 	 * (3) bp start below, but end in range YES
1161 	 * (4) bp entirely within YES
1162 	 * (5) bp starts within, ends above YES
1163 	 *
1164 	 * lock range 10-19 (offset 10 length 10)
1165 	 * (1) 1-5: first if kicks it out
1166 	 * (2) 30-35: second if kicks it out
1167 	 * (3) 5-15: passes both ifs
1168 	 * (4) 12-14: passes both ifs
1169 	 * (5) 19-20: passes both
1170 	 */
1171 	off_t lend = lstart + len - 1;
1172 	off_t bstart = bp->bio_offset;
1173 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1174 
1175 	if (bend < lstart)
1176 		return (0);
1177 	if (lend < bstart)
1178 		return (0);
1179 	return (1);
1180 }
1181 
1182 static int
1183 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1184 {
1185 	struct g_raid_lock *lp;
1186 
1187 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1188 
1189 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1190 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1191 			return (1);
1192 	}
1193 	return (0);
1194 }
1195 
1196 static void
1197 g_raid_start_request(struct bio *bp)
1198 {
1199 	struct g_raid_softc *sc;
1200 	struct g_raid_volume *vol;
1201 
1202 	sc = bp->bio_to->geom->softc;
1203 	sx_assert(&sc->sc_lock, SX_LOCKED);
1204 	vol = bp->bio_to->private;
1205 
1206 	/*
1207 	 * Check to see if this item is in a locked range.  If so,
1208 	 * queue it to our locked queue and return.  We'll requeue
1209 	 * it when the range is unlocked.  Internal I/O for the
1210 	 * rebuild/rescan/recovery process is excluded from this
1211 	 * check so we can actually do the recovery.
1212 	 */
1213 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1214 	    g_raid_is_in_locked_range(vol, bp)) {
1215 		G_RAID_LOGREQ(3, bp, "Defer request.");
1216 		bioq_insert_tail(&vol->v_locked, bp);
1217 		return;
1218 	}
1219 
1220 	/*
1221 	 * If we're actually going to do the write/delete, then
1222 	 * update the idle stats for the volume.
1223 	 */
1224 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1225 		if (!vol->v_dirty)
1226 			g_raid_dirty(vol);
1227 		vol->v_writes++;
1228 	}
1229 
1230 	/*
1231 	 * Put request onto inflight queue, so we can check if new
1232 	 * synchronization requests don't collide with it.  Then tell
1233 	 * the transformation layer to start the I/O.
1234 	 */
1235 	bioq_insert_tail(&vol->v_inflight, bp);
1236 	G_RAID_LOGREQ(4, bp, "Request started");
1237 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1238 }
1239 
1240 static void
1241 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1242 {
1243 	off_t off, len;
1244 	struct bio *nbp;
1245 	struct g_raid_lock *lp;
1246 
1247 	vol->v_pending_lock = 0;
1248 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1249 		if (lp->l_pending) {
1250 			off = lp->l_offset;
1251 			len = lp->l_length;
1252 			lp->l_pending = 0;
1253 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1254 				if (g_raid_bio_overlaps(nbp, off, len))
1255 					lp->l_pending++;
1256 			}
1257 			if (lp->l_pending) {
1258 				vol->v_pending_lock = 1;
1259 				G_RAID_DEBUG1(4, vol->v_softc,
1260 				    "Deferred lock(%jd, %jd) has %d pending",
1261 				    (intmax_t)off, (intmax_t)(off + len),
1262 				    lp->l_pending);
1263 				continue;
1264 			}
1265 			G_RAID_DEBUG1(4, vol->v_softc,
1266 			    "Deferred lock of %jd to %jd completed",
1267 			    (intmax_t)off, (intmax_t)(off + len));
1268 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1269 		}
1270 	}
1271 }
1272 
1273 void
1274 g_raid_iodone(struct bio *bp, int error)
1275 {
1276 	struct g_raid_softc *sc;
1277 	struct g_raid_volume *vol;
1278 
1279 	sc = bp->bio_to->geom->softc;
1280 	sx_assert(&sc->sc_lock, SX_LOCKED);
1281 	vol = bp->bio_to->private;
1282 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1283 
1284 	/* Update stats if we done write/delete. */
1285 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1286 		vol->v_writes--;
1287 		vol->v_last_write = time_uptime;
1288 	}
1289 
1290 	bioq_remove(&vol->v_inflight, bp);
1291 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1292 		g_raid_finish_with_locked_ranges(vol, bp);
1293 	getmicrouptime(&vol->v_last_done);
1294 	g_io_deliver(bp, error);
1295 }
1296 
1297 int
1298 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1299     struct bio *ignore, void *argp)
1300 {
1301 	struct g_raid_softc *sc;
1302 	struct g_raid_lock *lp;
1303 	struct bio *bp;
1304 
1305 	sc = vol->v_softc;
1306 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1307 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1308 	lp->l_offset = off;
1309 	lp->l_length = len;
1310 	lp->l_callback_arg = argp;
1311 
1312 	lp->l_pending = 0;
1313 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1314 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1315 			lp->l_pending++;
1316 	}
1317 
1318 	/*
1319 	 * If there are any writes that are pending, we return EBUSY.  All
1320 	 * callers will have to wait until all pending writes clear.
1321 	 */
1322 	if (lp->l_pending > 0) {
1323 		vol->v_pending_lock = 1;
1324 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1325 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1326 		return (EBUSY);
1327 	}
1328 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1329 	    (intmax_t)off, (intmax_t)(off+len));
1330 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1331 	return (0);
1332 }
1333 
1334 int
1335 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1336 {
1337 	struct g_raid_lock *lp;
1338 	struct g_raid_softc *sc;
1339 	struct bio *bp;
1340 
1341 	sc = vol->v_softc;
1342 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1343 		if (lp->l_offset == off && lp->l_length == len) {
1344 			LIST_REMOVE(lp, l_next);
1345 			/* XXX
1346 			 * Right now we just put them all back on the queue
1347 			 * and hope for the best.  We hope this because any
1348 			 * locked ranges will go right back on this list
1349 			 * when the worker thread runs.
1350 			 * XXX
1351 			 */
1352 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1353 			    (intmax_t)lp->l_offset,
1354 			    (intmax_t)(lp->l_offset+lp->l_length));
1355 			mtx_lock(&sc->sc_queue_mtx);
1356 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1357 				bioq_disksort(&sc->sc_queue, bp);
1358 			mtx_unlock(&sc->sc_queue_mtx);
1359 			free(lp, M_RAID);
1360 			return (0);
1361 		}
1362 	}
1363 	return (EINVAL);
1364 }
1365 
1366 void
1367 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1368 {
1369 	struct g_consumer *cp;
1370 	struct g_raid_disk *disk, *tdisk;
1371 
1372 	bp->bio_caller1 = sd;
1373 
1374 	/*
1375 	 * Make sure that the disk is present. Generally it is a task of
1376 	 * transformation layers to not send requests to absent disks, but
1377 	 * it is better to be safe and report situation then sorry.
1378 	 */
1379 	if (sd->sd_disk == NULL) {
1380 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1381 nodisk:
1382 		bp->bio_from = NULL;
1383 		bp->bio_to = NULL;
1384 		bp->bio_error = ENXIO;
1385 		g_raid_disk_done(bp);
1386 		return;
1387 	}
1388 	disk = sd->sd_disk;
1389 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1390 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1391 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1392 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1393 		goto nodisk;
1394 	}
1395 
1396 	cp = disk->d_consumer;
1397 	bp->bio_from = cp;
1398 	bp->bio_to = cp->provider;
1399 	cp->index++;
1400 
1401 	/* Update average disks load. */
1402 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1403 		if (tdisk->d_consumer == NULL)
1404 			tdisk->d_load = 0;
1405 		else
1406 			tdisk->d_load = (tdisk->d_consumer->index *
1407 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1408 	}
1409 
1410 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1411 	if (dumping) {
1412 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1413 		if (bp->bio_cmd == BIO_WRITE) {
1414 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1415 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1416 		} else
1417 			bp->bio_error = EOPNOTSUPP;
1418 		g_raid_disk_done(bp);
1419 	} else {
1420 		bp->bio_done = g_raid_disk_done;
1421 		bp->bio_offset += sd->sd_offset;
1422 		G_RAID_LOGREQ(3, bp, "Sending request.");
1423 		g_io_request(bp, cp);
1424 	}
1425 }
1426 
1427 int
1428 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1429     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1430 {
1431 
1432 	if (sd->sd_disk == NULL)
1433 		return (ENXIO);
1434 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1435 		return (EOPNOTSUPP);
1436 	return (dump_write(&sd->sd_disk->d_kd.di,
1437 	    virtual, physical,
1438 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1439 	    length));
1440 }
1441 
1442 static void
1443 g_raid_disk_done(struct bio *bp)
1444 {
1445 	struct g_raid_softc *sc;
1446 	struct g_raid_subdisk *sd;
1447 
1448 	sd = bp->bio_caller1;
1449 	sc = sd->sd_softc;
1450 	mtx_lock(&sc->sc_queue_mtx);
1451 	bioq_disksort(&sc->sc_queue, bp);
1452 	mtx_unlock(&sc->sc_queue_mtx);
1453 	if (!dumping)
1454 		wakeup(sc);
1455 }
1456 
1457 static void
1458 g_raid_disk_done_request(struct bio *bp)
1459 {
1460 	struct g_raid_softc *sc;
1461 	struct g_raid_disk *disk;
1462 	struct g_raid_subdisk *sd;
1463 	struct g_raid_volume *vol;
1464 
1465 	g_topology_assert_not();
1466 
1467 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1468 	sd = bp->bio_caller1;
1469 	sc = sd->sd_softc;
1470 	vol = sd->sd_volume;
1471 	if (bp->bio_from != NULL) {
1472 		bp->bio_from->index--;
1473 		disk = bp->bio_from->private;
1474 		if (disk == NULL)
1475 			g_raid_kill_consumer(sc, bp->bio_from);
1476 	}
1477 	bp->bio_offset -= sd->sd_offset;
1478 
1479 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1480 }
1481 
1482 static void
1483 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1484 {
1485 
1486 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1487 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1488 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1489 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1490 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1491 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1492 	else
1493 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1494 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1495 		KASSERT(ep->e_error == 0,
1496 		    ("Error cannot be handled."));
1497 		g_raid_event_free(ep);
1498 	} else {
1499 		ep->e_flags |= G_RAID_EVENT_DONE;
1500 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1501 		mtx_lock(&sc->sc_queue_mtx);
1502 		wakeup(ep);
1503 		mtx_unlock(&sc->sc_queue_mtx);
1504 	}
1505 }
1506 
1507 /*
1508  * Worker thread.
1509  */
1510 static void
1511 g_raid_worker(void *arg)
1512 {
1513 	struct g_raid_softc *sc;
1514 	struct g_raid_event *ep;
1515 	struct g_raid_volume *vol;
1516 	struct bio *bp;
1517 	struct timeval now, t;
1518 	int timeout, rv;
1519 
1520 	sc = arg;
1521 	thread_lock(curthread);
1522 	sched_prio(curthread, PRIBIO);
1523 	thread_unlock(curthread);
1524 
1525 	sx_xlock(&sc->sc_lock);
1526 	for (;;) {
1527 		mtx_lock(&sc->sc_queue_mtx);
1528 		/*
1529 		 * First take a look at events.
1530 		 * This is important to handle events before any I/O requests.
1531 		 */
1532 		bp = NULL;
1533 		vol = NULL;
1534 		rv = 0;
1535 		ep = TAILQ_FIRST(&sc->sc_events);
1536 		if (ep != NULL)
1537 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1538 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1539 			;
1540 		else {
1541 			getmicrouptime(&now);
1542 			t = now;
1543 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1544 				if (bioq_first(&vol->v_inflight) == NULL &&
1545 				    vol->v_tr &&
1546 				    timevalcmp(&vol->v_last_done, &t, < ))
1547 					t = vol->v_last_done;
1548 			}
1549 			timevalsub(&t, &now);
1550 			timeout = g_raid_idle_threshold +
1551 			    t.tv_sec * 1000000 + t.tv_usec;
1552 			if (timeout > 0) {
1553 				/*
1554 				 * Two steps to avoid overflows at HZ=1000
1555 				 * and idle timeouts > 2.1s.  Some rounding
1556 				 * errors can occur, but they are < 1tick,
1557 				 * which is deemed to be close enough for
1558 				 * this purpose.
1559 				 */
1560 				int micpertic = 1000000 / hz;
1561 				timeout = (timeout + micpertic - 1) / micpertic;
1562 				sx_xunlock(&sc->sc_lock);
1563 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1564 				    PRIBIO | PDROP, "-", timeout);
1565 				sx_xlock(&sc->sc_lock);
1566 				goto process;
1567 			} else
1568 				rv = EWOULDBLOCK;
1569 		}
1570 		mtx_unlock(&sc->sc_queue_mtx);
1571 process:
1572 		if (ep != NULL) {
1573 			g_raid_handle_event(sc, ep);
1574 		} else if (bp != NULL) {
1575 			if (bp->bio_to != NULL &&
1576 			    bp->bio_to->geom == sc->sc_geom)
1577 				g_raid_start_request(bp);
1578 			else
1579 				g_raid_disk_done_request(bp);
1580 		} else if (rv == EWOULDBLOCK) {
1581 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1582 				g_raid_clean(vol, -1);
1583 				if (bioq_first(&vol->v_inflight) == NULL &&
1584 				    vol->v_tr) {
1585 					t.tv_sec = g_raid_idle_threshold / 1000000;
1586 					t.tv_usec = g_raid_idle_threshold % 1000000;
1587 					timevaladd(&t, &vol->v_last_done);
1588 					getmicrouptime(&now);
1589 					if (timevalcmp(&t, &now, <= )) {
1590 						G_RAID_TR_IDLE(vol->v_tr);
1591 						vol->v_last_done = now;
1592 					}
1593 				}
1594 			}
1595 		}
1596 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1597 			g_raid_destroy_node(sc, 1);	/* May not return. */
1598 	}
1599 }
1600 
1601 static void
1602 g_raid_poll(struct g_raid_softc *sc)
1603 {
1604 	struct g_raid_event *ep;
1605 	struct bio *bp;
1606 
1607 	sx_xlock(&sc->sc_lock);
1608 	mtx_lock(&sc->sc_queue_mtx);
1609 	/*
1610 	 * First take a look at events.
1611 	 * This is important to handle events before any I/O requests.
1612 	 */
1613 	ep = TAILQ_FIRST(&sc->sc_events);
1614 	if (ep != NULL) {
1615 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1616 		mtx_unlock(&sc->sc_queue_mtx);
1617 		g_raid_handle_event(sc, ep);
1618 		goto out;
1619 	}
1620 	bp = bioq_takefirst(&sc->sc_queue);
1621 	if (bp != NULL) {
1622 		mtx_unlock(&sc->sc_queue_mtx);
1623 		if (bp->bio_from == NULL ||
1624 		    bp->bio_from->geom != sc->sc_geom)
1625 			g_raid_start_request(bp);
1626 		else
1627 			g_raid_disk_done_request(bp);
1628 	}
1629 out:
1630 	sx_xunlock(&sc->sc_lock);
1631 }
1632 
1633 static void
1634 g_raid_launch_provider(struct g_raid_volume *vol)
1635 {
1636 	struct g_raid_disk *disk;
1637 	struct g_raid_softc *sc;
1638 	struct g_provider *pp;
1639 	char name[G_RAID_MAX_VOLUMENAME];
1640 	off_t off;
1641 
1642 	sc = vol->v_softc;
1643 	sx_assert(&sc->sc_lock, SX_LOCKED);
1644 
1645 	g_topology_lock();
1646 	/* Try to name provider with volume name. */
1647 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1648 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1649 	    g_provider_by_name(name) != NULL) {
1650 		/* Otherwise use sequential volume number. */
1651 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1652 	}
1653 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1654 	pp->private = vol;
1655 	pp->mediasize = vol->v_mediasize;
1656 	pp->sectorsize = vol->v_sectorsize;
1657 	pp->stripesize = 0;
1658 	pp->stripeoffset = 0;
1659 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1660 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1661 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1662 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1663 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1664 		    disk->d_consumer != NULL &&
1665 		    disk->d_consumer->provider != NULL) {
1666 			pp->stripesize = disk->d_consumer->provider->stripesize;
1667 			off = disk->d_consumer->provider->stripeoffset;
1668 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1669 			if (off > 0)
1670 				pp->stripeoffset %= off;
1671 		}
1672 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1673 			pp->stripesize *= (vol->v_disks_count - 1);
1674 			pp->stripeoffset *= (vol->v_disks_count - 1);
1675 		}
1676 	} else
1677 		pp->stripesize = vol->v_strip_size;
1678 	vol->v_provider = pp;
1679 	g_error_provider(pp, 0);
1680 	g_topology_unlock();
1681 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1682 	    pp->name, vol->v_name);
1683 }
1684 
1685 static void
1686 g_raid_destroy_provider(struct g_raid_volume *vol)
1687 {
1688 	struct g_raid_softc *sc;
1689 	struct g_provider *pp;
1690 	struct bio *bp, *tmp;
1691 
1692 	g_topology_assert_not();
1693 	sc = vol->v_softc;
1694 	pp = vol->v_provider;
1695 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1696 
1697 	g_topology_lock();
1698 	g_error_provider(pp, ENXIO);
1699 	mtx_lock(&sc->sc_queue_mtx);
1700 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1701 		if (bp->bio_to != pp)
1702 			continue;
1703 		bioq_remove(&sc->sc_queue, bp);
1704 		g_io_deliver(bp, ENXIO);
1705 	}
1706 	mtx_unlock(&sc->sc_queue_mtx);
1707 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1708 	    pp->name, vol->v_name);
1709 	g_wither_provider(pp, ENXIO);
1710 	g_topology_unlock();
1711 	vol->v_provider = NULL;
1712 }
1713 
1714 /*
1715  * Update device state.
1716  */
1717 static int
1718 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1719 {
1720 	struct g_raid_softc *sc;
1721 
1722 	sc = vol->v_softc;
1723 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1724 
1725 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1726 	    g_raid_volume_event2str(event),
1727 	    vol->v_name);
1728 	switch (event) {
1729 	case G_RAID_VOLUME_E_DOWN:
1730 		if (vol->v_provider != NULL)
1731 			g_raid_destroy_provider(vol);
1732 		break;
1733 	case G_RAID_VOLUME_E_UP:
1734 		if (vol->v_provider == NULL)
1735 			g_raid_launch_provider(vol);
1736 		break;
1737 	case G_RAID_VOLUME_E_START:
1738 		if (vol->v_tr)
1739 			G_RAID_TR_START(vol->v_tr);
1740 		return (0);
1741 	default:
1742 		if (sc->sc_md)
1743 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1744 		return (0);
1745 	}
1746 
1747 	/* Manage root mount release. */
1748 	if (vol->v_starting) {
1749 		vol->v_starting = 0;
1750 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1751 		root_mount_rel(vol->v_rootmount);
1752 		vol->v_rootmount = NULL;
1753 	}
1754 	if (vol->v_stopping && vol->v_provider_open == 0)
1755 		g_raid_destroy_volume(vol);
1756 	return (0);
1757 }
1758 
1759 /*
1760  * Update subdisk state.
1761  */
1762 static int
1763 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1764 {
1765 	struct g_raid_softc *sc;
1766 	struct g_raid_volume *vol;
1767 
1768 	sc = sd->sd_softc;
1769 	vol = sd->sd_volume;
1770 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1771 
1772 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1773 	    g_raid_subdisk_event2str(event),
1774 	    vol->v_name, sd->sd_pos,
1775 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1776 	if (vol->v_tr)
1777 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1778 
1779 	return (0);
1780 }
1781 
1782 /*
1783  * Update disk state.
1784  */
1785 static int
1786 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1787 {
1788 	struct g_raid_softc *sc;
1789 
1790 	sc = disk->d_softc;
1791 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1792 
1793 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1794 	    g_raid_disk_event2str(event),
1795 	    g_raid_get_diskname(disk));
1796 
1797 	if (sc->sc_md)
1798 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1799 	return (0);
1800 }
1801 
1802 /*
1803  * Node event.
1804  */
1805 static int
1806 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1807 {
1808 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1809 
1810 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1811 	    g_raid_node_event2str(event));
1812 
1813 	if (event == G_RAID_NODE_E_WAKE)
1814 		return (0);
1815 	if (sc->sc_md)
1816 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1817 	return (0);
1818 }
1819 
1820 static int
1821 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1822 {
1823 	struct g_raid_volume *vol;
1824 	struct g_raid_softc *sc;
1825 	int dcw, opens, error = 0;
1826 
1827 	g_topology_assert();
1828 	sc = pp->geom->softc;
1829 	vol = pp->private;
1830 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1831 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1832 
1833 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1834 	    acr, acw, ace);
1835 	dcw = pp->acw + acw;
1836 
1837 	g_topology_unlock();
1838 	sx_xlock(&sc->sc_lock);
1839 	/* Deny new opens while dying. */
1840 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1841 		error = ENXIO;
1842 		goto out;
1843 	}
1844 	if (dcw == 0)
1845 		g_raid_clean(vol, dcw);
1846 	vol->v_provider_open += acr + acw + ace;
1847 	/* Handle delayed node destruction. */
1848 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1849 	    vol->v_provider_open == 0) {
1850 		/* Count open volumes. */
1851 		opens = g_raid_nopens(sc);
1852 		if (opens == 0) {
1853 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1854 			/* Wake up worker to make it selfdestruct. */
1855 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1856 		}
1857 	}
1858 	/* Handle open volume destruction. */
1859 	if (vol->v_stopping && vol->v_provider_open == 0)
1860 		g_raid_destroy_volume(vol);
1861 out:
1862 	sx_xunlock(&sc->sc_lock);
1863 	g_topology_lock();
1864 	return (error);
1865 }
1866 
1867 struct g_raid_softc *
1868 g_raid_create_node(struct g_class *mp,
1869     const char *name, struct g_raid_md_object *md)
1870 {
1871 	struct g_raid_softc *sc;
1872 	struct g_geom *gp;
1873 	int error;
1874 
1875 	g_topology_assert();
1876 	G_RAID_DEBUG(1, "Creating array %s.", name);
1877 
1878 	gp = g_new_geomf(mp, "%s", name);
1879 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1880 	gp->start = g_raid_start;
1881 	gp->orphan = g_raid_orphan;
1882 	gp->access = g_raid_access;
1883 	gp->dumpconf = g_raid_dumpconf;
1884 
1885 	sc->sc_md = md;
1886 	sc->sc_geom = gp;
1887 	sc->sc_flags = 0;
1888 	TAILQ_INIT(&sc->sc_volumes);
1889 	TAILQ_INIT(&sc->sc_disks);
1890 	sx_init(&sc->sc_lock, "graid:lock");
1891 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1892 	TAILQ_INIT(&sc->sc_events);
1893 	bioq_init(&sc->sc_queue);
1894 	gp->softc = sc;
1895 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1896 	    "g_raid %s", name);
1897 	if (error != 0) {
1898 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1899 		mtx_destroy(&sc->sc_queue_mtx);
1900 		sx_destroy(&sc->sc_lock);
1901 		g_destroy_geom(sc->sc_geom);
1902 		free(sc, M_RAID);
1903 		return (NULL);
1904 	}
1905 
1906 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1907 	return (sc);
1908 }
1909 
1910 struct g_raid_volume *
1911 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1912 {
1913 	struct g_raid_volume	*vol, *vol1;
1914 	int i;
1915 
1916 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1917 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1918 	vol->v_softc = sc;
1919 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1920 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1921 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1922 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1923 	vol->v_rotate_parity = 1;
1924 	bioq_init(&vol->v_inflight);
1925 	bioq_init(&vol->v_locked);
1926 	LIST_INIT(&vol->v_locks);
1927 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1928 		vol->v_subdisks[i].sd_softc = sc;
1929 		vol->v_subdisks[i].sd_volume = vol;
1930 		vol->v_subdisks[i].sd_pos = i;
1931 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1932 	}
1933 
1934 	/* Find free ID for this volume. */
1935 	g_topology_lock();
1936 	vol1 = vol;
1937 	if (id >= 0) {
1938 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1939 			if (vol1->v_global_id == id)
1940 				break;
1941 		}
1942 	}
1943 	if (vol1 != NULL) {
1944 		for (id = 0; ; id++) {
1945 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1946 				if (vol1->v_global_id == id)
1947 					break;
1948 			}
1949 			if (vol1 == NULL)
1950 				break;
1951 		}
1952 	}
1953 	vol->v_global_id = id;
1954 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1955 	g_topology_unlock();
1956 
1957 	/* Delay root mounting. */
1958 	vol->v_rootmount = root_mount_hold("GRAID");
1959 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1960 	vol->v_starting = 1;
1961 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1962 	return (vol);
1963 }
1964 
1965 struct g_raid_disk *
1966 g_raid_create_disk(struct g_raid_softc *sc)
1967 {
1968 	struct g_raid_disk	*disk;
1969 
1970 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1971 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1972 	disk->d_softc = sc;
1973 	disk->d_state = G_RAID_DISK_S_NONE;
1974 	TAILQ_INIT(&disk->d_subdisks);
1975 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1976 	return (disk);
1977 }
1978 
1979 int g_raid_start_volume(struct g_raid_volume *vol)
1980 {
1981 	struct g_raid_tr_class *class;
1982 	struct g_raid_tr_object *obj;
1983 	int status;
1984 
1985 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1986 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1987 		if (!class->trc_enable)
1988 			continue;
1989 		G_RAID_DEBUG1(2, vol->v_softc,
1990 		    "Tasting volume %s for %s transformation.",
1991 		    vol->v_name, class->name);
1992 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1993 		    M_WAITOK);
1994 		obj->tro_class = class;
1995 		obj->tro_volume = vol;
1996 		status = G_RAID_TR_TASTE(obj, vol);
1997 		if (status != G_RAID_TR_TASTE_FAIL)
1998 			break;
1999 		kobj_delete((kobj_t)obj, M_RAID);
2000 	}
2001 	if (class == NULL) {
2002 		G_RAID_DEBUG1(0, vol->v_softc,
2003 		    "No transformation module found for %s.",
2004 		    vol->v_name);
2005 		vol->v_tr = NULL;
2006 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2007 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2008 		    G_RAID_EVENT_VOLUME);
2009 		return (-1);
2010 	}
2011 	G_RAID_DEBUG1(2, vol->v_softc,
2012 	    "Transformation module %s chosen for %s.",
2013 	    class->name, vol->v_name);
2014 	vol->v_tr = obj;
2015 	return (0);
2016 }
2017 
2018 int
2019 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2020 {
2021 	struct g_raid_volume *vol, *tmpv;
2022 	struct g_raid_disk *disk, *tmpd;
2023 	int error = 0;
2024 
2025 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2026 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2027 		if (g_raid_destroy_volume(vol))
2028 			error = EBUSY;
2029 	}
2030 	if (error)
2031 		return (error);
2032 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2033 		if (g_raid_destroy_disk(disk))
2034 			error = EBUSY;
2035 	}
2036 	if (error)
2037 		return (error);
2038 	if (sc->sc_md) {
2039 		G_RAID_MD_FREE(sc->sc_md);
2040 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2041 		sc->sc_md = NULL;
2042 	}
2043 	if (sc->sc_geom != NULL) {
2044 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2045 		g_topology_lock();
2046 		sc->sc_geom->softc = NULL;
2047 		g_wither_geom(sc->sc_geom, ENXIO);
2048 		g_topology_unlock();
2049 		sc->sc_geom = NULL;
2050 	} else
2051 		G_RAID_DEBUG(1, "Array destroyed.");
2052 	if (worker) {
2053 		g_raid_event_cancel(sc, sc);
2054 		mtx_destroy(&sc->sc_queue_mtx);
2055 		sx_xunlock(&sc->sc_lock);
2056 		sx_destroy(&sc->sc_lock);
2057 		wakeup(&sc->sc_stopping);
2058 		free(sc, M_RAID);
2059 		curthread->td_pflags &= ~TDP_GEOM;
2060 		G_RAID_DEBUG(1, "Thread exiting.");
2061 		kproc_exit(0);
2062 	} else {
2063 		/* Wake up worker to make it selfdestruct. */
2064 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2065 	}
2066 	return (0);
2067 }
2068 
2069 int
2070 g_raid_destroy_volume(struct g_raid_volume *vol)
2071 {
2072 	struct g_raid_softc *sc;
2073 	struct g_raid_disk *disk;
2074 	int i;
2075 
2076 	sc = vol->v_softc;
2077 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2078 	vol->v_stopping = 1;
2079 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2080 		if (vol->v_tr) {
2081 			G_RAID_TR_STOP(vol->v_tr);
2082 			return (EBUSY);
2083 		} else
2084 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2085 	}
2086 	if (g_raid_event_check(sc, vol) != 0)
2087 		return (EBUSY);
2088 	if (vol->v_provider != NULL)
2089 		return (EBUSY);
2090 	if (vol->v_provider_open != 0)
2091 		return (EBUSY);
2092 	if (vol->v_tr) {
2093 		G_RAID_TR_FREE(vol->v_tr);
2094 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2095 		vol->v_tr = NULL;
2096 	}
2097 	if (vol->v_rootmount)
2098 		root_mount_rel(vol->v_rootmount);
2099 	g_topology_lock();
2100 	LIST_REMOVE(vol, v_global_next);
2101 	g_topology_unlock();
2102 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2103 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2104 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2105 		disk = vol->v_subdisks[i].sd_disk;
2106 		if (disk == NULL)
2107 			continue;
2108 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2109 	}
2110 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2111 	if (sc->sc_md)
2112 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2113 	g_raid_event_cancel(sc, vol);
2114 	free(vol, M_RAID);
2115 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2116 		/* Wake up worker to let it selfdestruct. */
2117 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2118 	}
2119 	return (0);
2120 }
2121 
2122 int
2123 g_raid_destroy_disk(struct g_raid_disk *disk)
2124 {
2125 	struct g_raid_softc *sc;
2126 	struct g_raid_subdisk *sd, *tmp;
2127 
2128 	sc = disk->d_softc;
2129 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2130 	if (disk->d_consumer) {
2131 		g_raid_kill_consumer(sc, disk->d_consumer);
2132 		disk->d_consumer = NULL;
2133 	}
2134 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2135 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2136 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2137 		    G_RAID_EVENT_SUBDISK);
2138 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2139 		sd->sd_disk = NULL;
2140 	}
2141 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2142 	if (sc->sc_md)
2143 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2144 	g_raid_event_cancel(sc, disk);
2145 	free(disk, M_RAID);
2146 	return (0);
2147 }
2148 
2149 int
2150 g_raid_destroy(struct g_raid_softc *sc, int how)
2151 {
2152 	int opens;
2153 
2154 	g_topology_assert_not();
2155 	if (sc == NULL)
2156 		return (ENXIO);
2157 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2158 
2159 	/* Count open volumes. */
2160 	opens = g_raid_nopens(sc);
2161 
2162 	/* React on some opened volumes. */
2163 	if (opens > 0) {
2164 		switch (how) {
2165 		case G_RAID_DESTROY_SOFT:
2166 			G_RAID_DEBUG1(1, sc,
2167 			    "%d volumes are still open.",
2168 			    opens);
2169 			return (EBUSY);
2170 		case G_RAID_DESTROY_DELAYED:
2171 			G_RAID_DEBUG1(1, sc,
2172 			    "Array will be destroyed on last close.");
2173 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2174 			return (EBUSY);
2175 		case G_RAID_DESTROY_HARD:
2176 			G_RAID_DEBUG1(1, sc,
2177 			    "%d volumes are still open.",
2178 			    opens);
2179 		}
2180 	}
2181 
2182 	/* Mark node for destruction. */
2183 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2184 	/* Wake up worker to let it selfdestruct. */
2185 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2186 	/* Sleep until node destroyed. */
2187 	sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2188 	    PRIBIO | PDROP, "r:destroy", 0);
2189 	return (0);
2190 }
2191 
2192 static void
2193 g_raid_taste_orphan(struct g_consumer *cp)
2194 {
2195 
2196 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2197 	    cp->provider->name));
2198 }
2199 
2200 static struct g_geom *
2201 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2202 {
2203 	struct g_consumer *cp;
2204 	struct g_geom *gp, *geom;
2205 	struct g_raid_md_class *class;
2206 	struct g_raid_md_object *obj;
2207 	int status;
2208 
2209 	g_topology_assert();
2210 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2211 	if (!g_raid_enable)
2212 		return (NULL);
2213 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2214 
2215 	gp = g_new_geomf(mp, "raid:taste");
2216 	/*
2217 	 * This orphan function should be never called.
2218 	 */
2219 	gp->orphan = g_raid_taste_orphan;
2220 	cp = g_new_consumer(gp);
2221 	g_attach(cp, pp);
2222 
2223 	geom = NULL;
2224 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2225 		if (!class->mdc_enable)
2226 			continue;
2227 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2228 		    pp->name, class->name);
2229 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2230 		    M_WAITOK);
2231 		obj->mdo_class = class;
2232 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2233 		if (status != G_RAID_MD_TASTE_NEW)
2234 			kobj_delete((kobj_t)obj, M_RAID);
2235 		if (status != G_RAID_MD_TASTE_FAIL)
2236 			break;
2237 	}
2238 
2239 	g_detach(cp);
2240 	g_destroy_consumer(cp);
2241 	g_destroy_geom(gp);
2242 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2243 	return (geom);
2244 }
2245 
2246 int
2247 g_raid_create_node_format(const char *format, struct gctl_req *req,
2248     struct g_geom **gp)
2249 {
2250 	struct g_raid_md_class *class;
2251 	struct g_raid_md_object *obj;
2252 	int status;
2253 
2254 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2255 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2256 		if (strcasecmp(class->name, format) == 0)
2257 			break;
2258 	}
2259 	if (class == NULL) {
2260 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2261 		return (G_RAID_MD_TASTE_FAIL);
2262 	}
2263 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2264 	    M_WAITOK);
2265 	obj->mdo_class = class;
2266 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2267 	if (status != G_RAID_MD_TASTE_NEW)
2268 		kobj_delete((kobj_t)obj, M_RAID);
2269 	return (status);
2270 }
2271 
2272 static int
2273 g_raid_destroy_geom(struct gctl_req *req __unused,
2274     struct g_class *mp __unused, struct g_geom *gp)
2275 {
2276 	struct g_raid_softc *sc;
2277 	int error;
2278 
2279 	g_topology_unlock();
2280 	sc = gp->softc;
2281 	sx_xlock(&sc->sc_lock);
2282 	g_cancel_event(sc);
2283 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2284 	if (error != 0)
2285 		sx_xunlock(&sc->sc_lock);
2286 	g_topology_lock();
2287 	return (error);
2288 }
2289 
2290 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2291     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2292 {
2293 
2294 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2295 		return;
2296 	if (sc->sc_md)
2297 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2298 }
2299 
2300 void g_raid_fail_disk(struct g_raid_softc *sc,
2301     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2302 {
2303 
2304 	if (disk == NULL)
2305 		disk = sd->sd_disk;
2306 	if (disk == NULL) {
2307 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2308 		return;
2309 	}
2310 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2311 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2312 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2313 		return;
2314 	}
2315 	if (sc->sc_md)
2316 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2317 }
2318 
2319 static void
2320 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2321     struct g_consumer *cp, struct g_provider *pp)
2322 {
2323 	struct g_raid_softc *sc;
2324 	struct g_raid_volume *vol;
2325 	struct g_raid_subdisk *sd;
2326 	struct g_raid_disk *disk;
2327 	int i, s;
2328 
2329 	g_topology_assert();
2330 
2331 	sc = gp->softc;
2332 	if (sc == NULL)
2333 		return;
2334 	if (pp != NULL) {
2335 		vol = pp->private;
2336 		g_topology_unlock();
2337 		sx_xlock(&sc->sc_lock);
2338 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2339 		    vol->v_name);
2340 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2341 		    g_raid_volume_level2str(vol->v_raid_level,
2342 		    vol->v_raid_level_qualifier));
2343 		sbuf_printf(sb,
2344 		    "%s<Transformation>%s</Transformation>\n", indent,
2345 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2346 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2347 		    vol->v_disks_count);
2348 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2349 		    vol->v_strip_size);
2350 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2351 		    g_raid_volume_state2str(vol->v_state));
2352 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2353 		    vol->v_dirty ? "Yes" : "No");
2354 		sbuf_printf(sb, "%s<Subdisks>", indent);
2355 		for (i = 0; i < vol->v_disks_count; i++) {
2356 			sd = &vol->v_subdisks[i];
2357 			if (sd->sd_disk != NULL &&
2358 			    sd->sd_disk->d_consumer != NULL) {
2359 				sbuf_printf(sb, "%s ",
2360 				    g_raid_get_diskname(sd->sd_disk));
2361 			} else {
2362 				sbuf_printf(sb, "NONE ");
2363 			}
2364 			sbuf_printf(sb, "(%s",
2365 			    g_raid_subdisk_state2str(sd->sd_state));
2366 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2367 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2368 				sbuf_printf(sb, " %d%%",
2369 				    (int)(sd->sd_rebuild_pos * 100 /
2370 				     sd->sd_size));
2371 			}
2372 			sbuf_printf(sb, ")");
2373 			if (i + 1 < vol->v_disks_count)
2374 				sbuf_printf(sb, ", ");
2375 		}
2376 		sbuf_printf(sb, "</Subdisks>\n");
2377 		sx_xunlock(&sc->sc_lock);
2378 		g_topology_lock();
2379 	} else if (cp != NULL) {
2380 		disk = cp->private;
2381 		if (disk == NULL)
2382 			return;
2383 		g_topology_unlock();
2384 		sx_xlock(&sc->sc_lock);
2385 		sbuf_printf(sb, "%s<State>%s", indent,
2386 		    g_raid_disk_state2str(disk->d_state));
2387 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2388 			sbuf_printf(sb, " (");
2389 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2390 				sbuf_printf(sb, "%s",
2391 				    g_raid_subdisk_state2str(sd->sd_state));
2392 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2393 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2394 					sbuf_printf(sb, " %d%%",
2395 					    (int)(sd->sd_rebuild_pos * 100 /
2396 					     sd->sd_size));
2397 				}
2398 				if (TAILQ_NEXT(sd, sd_next))
2399 					sbuf_printf(sb, ", ");
2400 			}
2401 			sbuf_printf(sb, ")");
2402 		}
2403 		sbuf_printf(sb, "</State>\n");
2404 		sbuf_printf(sb, "%s<Subdisks>", indent);
2405 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2406 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2407 			    sd->sd_volume->v_global_id,
2408 			    sd->sd_volume->v_name,
2409 			    sd->sd_pos, sd->sd_offset);
2410 			if (TAILQ_NEXT(sd, sd_next))
2411 				sbuf_printf(sb, ", ");
2412 		}
2413 		sbuf_printf(sb, "</Subdisks>\n");
2414 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2415 		    disk->d_read_errs);
2416 		sx_xunlock(&sc->sc_lock);
2417 		g_topology_lock();
2418 	} else {
2419 		g_topology_unlock();
2420 		sx_xlock(&sc->sc_lock);
2421 		if (sc->sc_md) {
2422 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2423 			    sc->sc_md->mdo_class->name);
2424 		}
2425 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2426 			s = 0xff;
2427 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2428 				if (vol->v_state < s)
2429 					s = vol->v_state;
2430 			}
2431 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2432 			    g_raid_volume_state2str(s));
2433 		}
2434 		sx_xunlock(&sc->sc_lock);
2435 		g_topology_lock();
2436 	}
2437 }
2438 
2439 static void
2440 g_raid_shutdown_post_sync(void *arg, int howto)
2441 {
2442 	struct g_class *mp;
2443 	struct g_geom *gp, *gp2;
2444 	struct g_raid_softc *sc;
2445 	struct g_raid_volume *vol;
2446 	int error;
2447 
2448 	mp = arg;
2449 	DROP_GIANT();
2450 	g_topology_lock();
2451 	g_raid_shutdown = 1;
2452 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2453 		if ((sc = gp->softc) == NULL)
2454 			continue;
2455 		g_topology_unlock();
2456 		sx_xlock(&sc->sc_lock);
2457 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2458 			g_raid_clean(vol, -1);
2459 		g_cancel_event(sc);
2460 		error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2461 		if (error != 0)
2462 			sx_xunlock(&sc->sc_lock);
2463 		g_topology_lock();
2464 	}
2465 	g_topology_unlock();
2466 	PICKUP_GIANT();
2467 }
2468 
2469 static void
2470 g_raid_init(struct g_class *mp)
2471 {
2472 
2473 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2474 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2475 	if (g_raid_post_sync == NULL)
2476 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2477 	g_raid_started = 1;
2478 }
2479 
2480 static void
2481 g_raid_fini(struct g_class *mp)
2482 {
2483 
2484 	if (g_raid_post_sync != NULL)
2485 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2486 	g_raid_started = 0;
2487 }
2488 
2489 int
2490 g_raid_md_modevent(module_t mod, int type, void *arg)
2491 {
2492 	struct g_raid_md_class *class, *c, *nc;
2493 	int error;
2494 
2495 	error = 0;
2496 	class = arg;
2497 	switch (type) {
2498 	case MOD_LOAD:
2499 		c = LIST_FIRST(&g_raid_md_classes);
2500 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2501 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2502 		else {
2503 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2504 			    nc->mdc_priority < class->mdc_priority)
2505 				c = nc;
2506 			LIST_INSERT_AFTER(c, class, mdc_list);
2507 		}
2508 		if (g_raid_started)
2509 			g_retaste(&g_raid_class);
2510 		break;
2511 	case MOD_UNLOAD:
2512 		LIST_REMOVE(class, mdc_list);
2513 		break;
2514 	default:
2515 		error = EOPNOTSUPP;
2516 		break;
2517 	}
2518 
2519 	return (error);
2520 }
2521 
2522 int
2523 g_raid_tr_modevent(module_t mod, int type, void *arg)
2524 {
2525 	struct g_raid_tr_class *class, *c, *nc;
2526 	int error;
2527 
2528 	error = 0;
2529 	class = arg;
2530 	switch (type) {
2531 	case MOD_LOAD:
2532 		c = LIST_FIRST(&g_raid_tr_classes);
2533 		if (c == NULL || c->trc_priority > class->trc_priority)
2534 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2535 		else {
2536 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2537 			    nc->trc_priority < class->trc_priority)
2538 				c = nc;
2539 			LIST_INSERT_AFTER(c, class, trc_list);
2540 		}
2541 		break;
2542 	case MOD_UNLOAD:
2543 		LIST_REMOVE(class, trc_list);
2544 		break;
2545 	default:
2546 		error = EOPNOTSUPP;
2547 		break;
2548 	}
2549 
2550 	return (error);
2551 }
2552 
2553 /*
2554  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2555  * to reduce module priority, allowing submodules to register them first.
2556  */
2557 static moduledata_t g_raid_mod = {
2558 	"g_raid",
2559 	g_modevent,
2560 	&g_raid_class
2561 };
2562 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2563 MODULE_VERSION(geom_raid, 0);
2564