xref: /freebsd/sys/geom/raid/md_jmicron.c (revision 6486b015fc84e96725fef22b0e3363351399ae83)
1 /*-
2  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
3  * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include <sys/param.h>
32 #include <sys/bio.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/kobj.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/systm.h>
41 #include <sys/taskqueue.h>
42 #include <geom/geom.h>
43 #include "geom/raid/g_raid.h"
44 #include "g_raid_md_if.h"
45 
46 static MALLOC_DEFINE(M_MD_JMICRON, "md_jmicron_data", "GEOM_RAID JMicron metadata");
47 
48 #define	JMICRON_MAX_DISKS	8
49 #define	JMICRON_MAX_SPARE	2
50 
51 struct jmicron_raid_conf {
52     u_int8_t		signature[2];
53 #define	JMICRON_MAGIC		"JM"
54 
55     u_int16_t		version;
56 #define	JMICRON_VERSION		0x0001
57 
58     u_int16_t		checksum;
59     u_int8_t		filler_1[10];
60     u_int32_t		disk_id;
61     u_int32_t		offset;
62     u_int32_t		disk_sectors_high;
63     u_int16_t		disk_sectors_low;
64     u_int8_t		filler_2[2];
65     u_int8_t		name[16];
66     u_int8_t		type;
67 #define	JMICRON_T_RAID0		0
68 #define	JMICRON_T_RAID1		1
69 #define	JMICRON_T_RAID01	2
70 #define	JMICRON_T_CONCAT	3
71 #define	JMICRON_T_RAID5		5
72 
73     u_int8_t		stripe_shift;
74     u_int16_t		flags;
75 #define	JMICRON_F_READY		0x0001
76 #define	JMICRON_F_BOOTABLE	0x0002
77 #define	JMICRON_F_BADSEC	0x0004
78 #define	JMICRON_F_ACTIVE	0x0010
79 #define	JMICRON_F_UNSYNC	0x0020
80 #define	JMICRON_F_NEWEST	0x0040
81 
82     u_int8_t		filler_3[4];
83     u_int32_t		spare[JMICRON_MAX_SPARE];
84     u_int32_t		disks[JMICRON_MAX_DISKS];
85 #define	JMICRON_DISK_MASK	0xFFFFFFF0
86 #define	JMICRON_SEG_MASK	0x0000000F
87     u_int8_t		filler_4[32];
88     u_int8_t		filler_5[384];
89 };
90 
91 struct g_raid_md_jmicron_perdisk {
92 	struct jmicron_raid_conf	*pd_meta;
93 	int				 pd_disk_pos;
94 	int				 pd_disk_id;
95 	off_t				 pd_disk_size;
96 };
97 
98 struct g_raid_md_jmicron_object {
99 	struct g_raid_md_object	 mdio_base;
100 	uint32_t		 mdio_config_id;
101 	struct jmicron_raid_conf	*mdio_meta;
102 	struct callout		 mdio_start_co;	/* STARTING state timer. */
103 	int			 mdio_total_disks;
104 	int			 mdio_disks_present;
105 	int			 mdio_started;
106 	int			 mdio_incomplete;
107 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
108 };
109 
110 static g_raid_md_create_t g_raid_md_create_jmicron;
111 static g_raid_md_taste_t g_raid_md_taste_jmicron;
112 static g_raid_md_event_t g_raid_md_event_jmicron;
113 static g_raid_md_ctl_t g_raid_md_ctl_jmicron;
114 static g_raid_md_write_t g_raid_md_write_jmicron;
115 static g_raid_md_fail_disk_t g_raid_md_fail_disk_jmicron;
116 static g_raid_md_free_disk_t g_raid_md_free_disk_jmicron;
117 static g_raid_md_free_t g_raid_md_free_jmicron;
118 
119 static kobj_method_t g_raid_md_jmicron_methods[] = {
120 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_jmicron),
121 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_jmicron),
122 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_jmicron),
123 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_jmicron),
124 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_jmicron),
125 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_jmicron),
126 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_jmicron),
127 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_jmicron),
128 	{ 0, 0 }
129 };
130 
131 static struct g_raid_md_class g_raid_md_jmicron_class = {
132 	"JMicron",
133 	g_raid_md_jmicron_methods,
134 	sizeof(struct g_raid_md_jmicron_object),
135 	.mdc_priority = 100
136 };
137 
138 static void
139 g_raid_md_jmicron_print(struct jmicron_raid_conf *meta)
140 {
141 	int k;
142 
143 	if (g_raid_debug < 1)
144 		return;
145 
146 	printf("********* ATA JMicron RAID Metadata *********\n");
147 	printf("signature           <%c%c>\n", meta->signature[0], meta->signature[1]);
148 	printf("version             %04x\n", meta->version);
149 	printf("checksum            0x%04x\n", meta->checksum);
150 	printf("disk_id             0x%08x\n", meta->disk_id);
151 	printf("offset              0x%08x\n", meta->offset);
152 	printf("disk_sectors_high   0x%08x\n", meta->disk_sectors_high);
153 	printf("disk_sectors_low    0x%04x\n", meta->disk_sectors_low);
154 	printf("name                <%.16s>\n", meta->name);
155 	printf("type                %d\n", meta->type);
156 	printf("stripe_shift        %d\n", meta->stripe_shift);
157 	printf("flags               %04x\n", meta->flags);
158 	printf("spare              ");
159 	for (k = 0; k < JMICRON_MAX_SPARE; k++)
160 		printf(" 0x%08x", meta->spare[k]);
161 	printf("\n");
162 	printf("disks              ");
163 	for (k = 0; k < JMICRON_MAX_DISKS; k++)
164 		printf(" 0x%08x", meta->disks[k]);
165 	printf("\n");
166 	printf("=================================================\n");
167 }
168 
169 static struct jmicron_raid_conf *
170 jmicron_meta_copy(struct jmicron_raid_conf *meta)
171 {
172 	struct jmicron_raid_conf *nmeta;
173 
174 	nmeta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK);
175 	memcpy(nmeta, meta, sizeof(*meta));
176 	return (nmeta);
177 }
178 
179 static int
180 jmicron_meta_total_disks(struct jmicron_raid_conf *meta)
181 {
182 	int pos;
183 
184 	for (pos = 0; pos < JMICRON_MAX_DISKS; pos++) {
185 		if (meta->disks[pos] == 0)
186 			break;
187 	}
188 	return (pos);
189 }
190 
191 static int
192 jmicron_meta_total_spare(struct jmicron_raid_conf *meta)
193 {
194 	int pos, n;
195 
196 	n = 0;
197 	for (pos = 0; pos < JMICRON_MAX_SPARE; pos++) {
198 		if (meta->spare[pos] != 0)
199 			n++;
200 	}
201 	return (n);
202 }
203 
204 /*
205  * Generate fake Configuration ID based on disk IDs.
206  * Note: it will change after each disk set change.
207  */
208 static uint32_t
209 jmicron_meta_config_id(struct jmicron_raid_conf *meta)
210 {
211 	int pos;
212 	uint32_t config_id;
213 
214 	config_id = 0;
215 	for (pos = 0; pos < JMICRON_MAX_DISKS; pos++)
216 		config_id += meta->disks[pos] << pos;
217 	return (config_id);
218 }
219 
220 static void
221 jmicron_meta_get_name(struct jmicron_raid_conf *meta, char *buf)
222 {
223 	int i;
224 
225 	strncpy(buf, meta->name, 16);
226 	buf[16] = 0;
227 	for (i = 15; i >= 0; i--) {
228 		if (buf[i] > 0x20)
229 			break;
230 		buf[i] = 0;
231 	}
232 }
233 
234 static void
235 jmicron_meta_put_name(struct jmicron_raid_conf *meta, char *buf)
236 {
237 
238 	memset(meta->name, 0x20, 16);
239 	memcpy(meta->name, buf, MIN(strlen(buf), 16));
240 }
241 
242 static int
243 jmicron_meta_find_disk(struct jmicron_raid_conf *meta, uint32_t id)
244 {
245 	int pos;
246 
247 	id &= JMICRON_DISK_MASK;
248 	for (pos = 0; pos < JMICRON_MAX_DISKS; pos++) {
249 		if ((meta->disks[pos] & JMICRON_DISK_MASK) == id)
250 			return (pos);
251 	}
252 	for (pos = 0; pos < JMICRON_MAX_SPARE; pos++) {
253 		if ((meta->spare[pos] & JMICRON_DISK_MASK) == id)
254 			return (-3);
255 	}
256 	return (-1);
257 }
258 
259 static struct jmicron_raid_conf *
260 jmicron_meta_read(struct g_consumer *cp)
261 {
262 	struct g_provider *pp;
263 	struct jmicron_raid_conf *meta;
264 	char *buf;
265 	int error, i;
266 	uint16_t checksum, *ptr;
267 
268 	pp = cp->provider;
269 
270 	/* Read the anchor sector. */
271 	buf = g_read_data(cp,
272 	    pp->mediasize - pp->sectorsize, pp->sectorsize, &error);
273 	if (buf == NULL) {
274 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
275 		    pp->name, error);
276 		return (NULL);
277 	}
278 	meta = (struct jmicron_raid_conf *)buf;
279 
280 	/* Check if this is an JMicron RAID struct */
281 	if (strncmp(meta->signature, JMICRON_MAGIC, strlen(JMICRON_MAGIC))) {
282 		G_RAID_DEBUG(1, "JMicron signature check failed on %s", pp->name);
283 		g_free(buf);
284 		return (NULL);
285 	}
286 	meta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK);
287 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
288 	g_free(buf);
289 
290 	/* Check metadata checksum. */
291 	for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 64; i++)
292 		checksum += *ptr++;
293 	if (checksum != 0) {
294 		G_RAID_DEBUG(1, "JMicron checksum check failed on %s", pp->name);
295 		free(meta, M_MD_JMICRON);
296 		return (NULL);
297 	}
298 
299 	return (meta);
300 }
301 
302 static int
303 jmicron_meta_write(struct g_consumer *cp, struct jmicron_raid_conf *meta)
304 {
305 	struct g_provider *pp;
306 	char *buf;
307 	int error, i;
308 	uint16_t checksum, *ptr;
309 
310 	pp = cp->provider;
311 
312 	/* Recalculate checksum for case if metadata were changed. */
313 	meta->checksum = 0;
314 	for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 64; i++)
315 		checksum += *ptr++;
316 	meta->checksum -= checksum;
317 
318 	/* Create and fill buffer. */
319 	buf = malloc(pp->sectorsize, M_MD_JMICRON, M_WAITOK | M_ZERO);
320 	memcpy(buf, meta, sizeof(*meta));
321 
322 	error = g_write_data(cp,
323 	    pp->mediasize - pp->sectorsize, buf, pp->sectorsize);
324 	if (error != 0) {
325 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
326 		    pp->name, error);
327 	}
328 
329 	free(buf, M_MD_JMICRON);
330 	return (error);
331 }
332 
333 static int
334 jmicron_meta_erase(struct g_consumer *cp)
335 {
336 	struct g_provider *pp;
337 	char *buf;
338 	int error;
339 
340 	pp = cp->provider;
341 	buf = malloc(pp->sectorsize, M_MD_JMICRON, M_WAITOK | M_ZERO);
342 	error = g_write_data(cp,
343 	    pp->mediasize - pp->sectorsize, buf, pp->sectorsize);
344 	if (error != 0) {
345 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
346 		    pp->name, error);
347 	}
348 	free(buf, M_MD_JMICRON);
349 	return (error);
350 }
351 
352 static struct g_raid_disk *
353 g_raid_md_jmicron_get_disk(struct g_raid_softc *sc, int id)
354 {
355 	struct g_raid_disk	*disk;
356 	struct g_raid_md_jmicron_perdisk *pd;
357 
358 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
359 		pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
360 		if (pd->pd_disk_pos == id)
361 			break;
362 	}
363 	return (disk);
364 }
365 
366 static int
367 g_raid_md_jmicron_supported(int level, int qual, int disks, int force)
368 {
369 
370 	if (disks > 8)
371 		return (0);
372 	switch (level) {
373 	case G_RAID_VOLUME_RL_RAID0:
374 		if (disks < 1)
375 			return (0);
376 		if (!force && (disks < 2 || disks > 6))
377 			return (0);
378 		break;
379 	case G_RAID_VOLUME_RL_RAID1:
380 		if (disks < 1)
381 			return (0);
382 		if (!force && (disks != 2))
383 			return (0);
384 		break;
385 	case G_RAID_VOLUME_RL_RAID1E:
386 		if (disks < 2)
387 			return (0);
388 		if (!force && (disks != 4))
389 			return (0);
390 		break;
391 	case G_RAID_VOLUME_RL_SINGLE:
392 		if (disks != 1)
393 			return (0);
394 		if (!force)
395 			return (0);
396 		break;
397 	case G_RAID_VOLUME_RL_CONCAT:
398 		if (disks < 2)
399 			return (0);
400 		break;
401 	case G_RAID_VOLUME_RL_RAID5:
402 		if (disks < 3)
403 			return (0);
404 		if (qual != G_RAID_VOLUME_RLQ_R5LA)
405 			return (0);
406 		if (!force)
407 			return (0);
408 		break;
409 	default:
410 		return (0);
411 	}
412 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
413 		return (0);
414 	return (1);
415 }
416 
417 static int
418 g_raid_md_jmicron_start_disk(struct g_raid_disk *disk)
419 {
420 	struct g_raid_softc *sc;
421 	struct g_raid_subdisk *sd, *tmpsd;
422 	struct g_raid_disk *olddisk, *tmpdisk;
423 	struct g_raid_md_object *md;
424 	struct g_raid_md_jmicron_object *mdi;
425 	struct g_raid_md_jmicron_perdisk *pd, *oldpd;
426 	struct jmicron_raid_conf *meta;
427 	int disk_pos, resurrection = 0;
428 
429 	sc = disk->d_softc;
430 	md = sc->sc_md;
431 	mdi = (struct g_raid_md_jmicron_object *)md;
432 	meta = mdi->mdio_meta;
433 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
434 	olddisk = NULL;
435 
436 	/* Find disk position in metadata by it's serial. */
437 	if (pd->pd_meta != NULL)
438 		disk_pos = jmicron_meta_find_disk(meta, pd->pd_disk_id);
439 	else
440 		disk_pos = -1;
441 	if (disk_pos < 0) {
442 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
443 		/* If we are in the start process, that's all for now. */
444 		if (!mdi->mdio_started)
445 			goto nofit;
446 		/*
447 		 * If we have already started - try to get use of the disk.
448 		 * Try to replace OFFLINE disks first, then FAILED.
449 		 */
450 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
451 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
452 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
453 				continue;
454 			/* Make sure this disk is big enough. */
455 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
456 				if (sd->sd_offset + sd->sd_size + 512 >
457 				    pd->pd_disk_size) {
458 					G_RAID_DEBUG1(1, sc,
459 					    "Disk too small (%ju < %ju)",
460 					    pd->pd_disk_size,
461 					    sd->sd_offset + sd->sd_size + 512);
462 					break;
463 				}
464 			}
465 			if (sd != NULL)
466 				continue;
467 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
468 				olddisk = tmpdisk;
469 				break;
470 			} else if (olddisk == NULL)
471 				olddisk = tmpdisk;
472 		}
473 		if (olddisk == NULL) {
474 nofit:
475 			if (disk_pos == -3 || pd->pd_disk_pos == -3) {
476 				g_raid_change_disk_state(disk,
477 				    G_RAID_DISK_S_SPARE);
478 				return (1);
479 			} else {
480 				g_raid_change_disk_state(disk,
481 				    G_RAID_DISK_S_STALE);
482 				return (0);
483 			}
484 		}
485 		oldpd = (struct g_raid_md_jmicron_perdisk *)olddisk->d_md_data;
486 		disk_pos = oldpd->pd_disk_pos;
487 		resurrection = 1;
488 	}
489 
490 	if (olddisk == NULL) {
491 		/* Find placeholder by position. */
492 		olddisk = g_raid_md_jmicron_get_disk(sc, disk_pos);
493 		if (olddisk == NULL)
494 			panic("No disk at position %d!", disk_pos);
495 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
496 			G_RAID_DEBUG1(1, sc, "More then one disk for pos %d",
497 			    disk_pos);
498 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
499 			return (0);
500 		}
501 		oldpd = (struct g_raid_md_jmicron_perdisk *)olddisk->d_md_data;
502 	}
503 
504 	/* Replace failed disk or placeholder with new disk. */
505 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
506 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
507 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
508 		sd->sd_disk = disk;
509 	}
510 	oldpd->pd_disk_pos = -2;
511 	pd->pd_disk_pos = disk_pos;
512 	/* Update global metadata just in case. */
513 	meta->disks[disk_pos] = pd->pd_disk_id;
514 
515 	/* If it was placeholder -- destroy it. */
516 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
517 		g_raid_destroy_disk(olddisk);
518 	} else {
519 		/* Otherwise, make it STALE_FAILED. */
520 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
521 	}
522 
523 	/* Welcome the new disk. */
524 	g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
525 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
526 
527 		/*
528 		 * Different disks may have different sizes/offsets,
529 		 * especially in concat mode. Update.
530 		 */
531 		if (!resurrection) {
532 			sd->sd_offset =
533 			    (off_t)pd->pd_meta->offset * 16 * 512; //ZZZ
534 			sd->sd_size =
535 			    (((off_t)pd->pd_meta->disk_sectors_high << 16) +
536 			      pd->pd_meta->disk_sectors_low) * 512;
537 		}
538 
539 		if (resurrection) {
540 			/* Stale disk, almost same as new. */
541 			g_raid_change_subdisk_state(sd,
542 			    G_RAID_SUBDISK_S_NEW);
543 		} else if ((meta->flags & JMICRON_F_BADSEC) != 0 &&
544 		    (pd->pd_meta->flags & JMICRON_F_BADSEC) == 0) {
545 			/* Cold-inserted or rebuilding disk. */
546 			g_raid_change_subdisk_state(sd,
547 			    G_RAID_SUBDISK_S_NEW);
548 		} else if (pd->pd_meta->flags & JMICRON_F_UNSYNC) {
549 			/* Dirty or resyncing disk.. */
550 			g_raid_change_subdisk_state(sd,
551 			    G_RAID_SUBDISK_S_STALE);
552 		} else {
553 			/* Up to date disk. */
554 			g_raid_change_subdisk_state(sd,
555 			    G_RAID_SUBDISK_S_ACTIVE);
556 		}
557 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
558 		    G_RAID_EVENT_SUBDISK);
559 	}
560 
561 	/* Update status of our need for spare. */
562 	if (mdi->mdio_started) {
563 		mdi->mdio_incomplete =
564 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
565 		     mdi->mdio_total_disks);
566 	}
567 
568 	return (resurrection);
569 }
570 
571 static void
572 g_disk_md_jmicron_retaste(void *arg, int pending)
573 {
574 
575 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
576 	g_retaste(&g_raid_class);
577 	free(arg, M_MD_JMICRON);
578 }
579 
580 static void
581 g_raid_md_jmicron_refill(struct g_raid_softc *sc)
582 {
583 	struct g_raid_md_object *md;
584 	struct g_raid_md_jmicron_object *mdi;
585 	struct g_raid_disk *disk;
586 	struct task *task;
587 	int update, na;
588 
589 	md = sc->sc_md;
590 	mdi = (struct g_raid_md_jmicron_object *)md;
591 	update = 0;
592 	do {
593 		/* Make sure we miss anything. */
594 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
595 		if (na == mdi->mdio_total_disks)
596 			break;
597 
598 		G_RAID_DEBUG1(1, md->mdo_softc,
599 		    "Array is not complete (%d of %d), "
600 		    "trying to refill.", na, mdi->mdio_total_disks);
601 
602 		/* Try to get use some of STALE disks. */
603 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
604 			if (disk->d_state == G_RAID_DISK_S_STALE) {
605 				update += g_raid_md_jmicron_start_disk(disk);
606 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
607 					break;
608 			}
609 		}
610 		if (disk != NULL)
611 			continue;
612 
613 		/* Try to get use some of SPARE disks. */
614 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
615 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
616 				update += g_raid_md_jmicron_start_disk(disk);
617 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
618 					break;
619 			}
620 		}
621 	} while (disk != NULL);
622 
623 	/* Write new metadata if we changed something. */
624 	if (update)
625 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
626 
627 	/* Update status of our need for spare. */
628 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
629 	    mdi->mdio_total_disks);
630 
631 	/* Request retaste hoping to find spare. */
632 	if (mdi->mdio_incomplete) {
633 		task = malloc(sizeof(struct task),
634 		    M_MD_JMICRON, M_WAITOK | M_ZERO);
635 		TASK_INIT(task, 0, g_disk_md_jmicron_retaste, task);
636 		taskqueue_enqueue(taskqueue_swi, task);
637 	}
638 }
639 
640 static void
641 g_raid_md_jmicron_start(struct g_raid_softc *sc)
642 {
643 	struct g_raid_md_object *md;
644 	struct g_raid_md_jmicron_object *mdi;
645 	struct g_raid_md_jmicron_perdisk *pd;
646 	struct jmicron_raid_conf *meta;
647 	struct g_raid_volume *vol;
648 	struct g_raid_subdisk *sd;
649 	struct g_raid_disk *disk;
650 	off_t size;
651 	int j, disk_pos;
652 	char buf[17];
653 
654 	md = sc->sc_md;
655 	mdi = (struct g_raid_md_jmicron_object *)md;
656 	meta = mdi->mdio_meta;
657 
658 	/* Create volumes and subdisks. */
659 	jmicron_meta_get_name(meta, buf);
660 	vol = g_raid_create_volume(sc, buf, -1);
661 	size = ((off_t)meta->disk_sectors_high << 16) + meta->disk_sectors_low;
662 	size *= 512; //ZZZ
663 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
664 	if (meta->type == JMICRON_T_RAID0) {
665 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
666 		vol->v_mediasize = size * mdi->mdio_total_disks;
667 	} else if (meta->type == JMICRON_T_RAID1) {
668 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
669 		vol->v_mediasize = size;
670 	} else if (meta->type == JMICRON_T_RAID01) {
671 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
672 		vol->v_mediasize = size * mdi->mdio_total_disks / 2;
673 	} else if (meta->type == JMICRON_T_CONCAT) {
674 		if (mdi->mdio_total_disks == 1)
675 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
676 		else
677 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
678 		vol->v_mediasize = 0;
679 	} else if (meta->type == JMICRON_T_RAID5) {
680 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
681 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
682 		vol->v_mediasize = size * (mdi->mdio_total_disks - 1);
683 	} else {
684 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
685 		vol->v_mediasize = 0;
686 	}
687 	vol->v_strip_size = 1024 << meta->stripe_shift; //ZZZ
688 	vol->v_disks_count = mdi->mdio_total_disks;
689 	vol->v_sectorsize = 512; //ZZZ
690 	for (j = 0; j < vol->v_disks_count; j++) {
691 		sd = &vol->v_subdisks[j];
692 		sd->sd_offset = (off_t)meta->offset * 16 * 512; //ZZZ
693 		sd->sd_size = size;
694 	}
695 	g_raid_start_volume(vol);
696 
697 	/* Create disk placeholders to store data for later writing. */
698 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
699 		pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
700 		pd->pd_disk_pos = disk_pos;
701 		pd->pd_disk_id = meta->disks[disk_pos];
702 		disk = g_raid_create_disk(sc);
703 		disk->d_md_data = (void *)pd;
704 		disk->d_state = G_RAID_DISK_S_OFFLINE;
705 		sd = &vol->v_subdisks[disk_pos];
706 		sd->sd_disk = disk;
707 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
708 	}
709 
710 	/* Make all disks found till the moment take their places. */
711 	do {
712 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
713 			if (disk->d_state == G_RAID_DISK_S_NONE) {
714 				g_raid_md_jmicron_start_disk(disk);
715 				break;
716 			}
717 		}
718 	} while (disk != NULL);
719 
720 	mdi->mdio_started = 1;
721 	G_RAID_DEBUG1(0, sc, "Array started.");
722 	g_raid_md_write_jmicron(md, NULL, NULL, NULL);
723 
724 	/* Pickup any STALE/SPARE disks to refill array if needed. */
725 	g_raid_md_jmicron_refill(sc);
726 
727 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
728 
729 	callout_stop(&mdi->mdio_start_co);
730 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
731 	root_mount_rel(mdi->mdio_rootmount);
732 	mdi->mdio_rootmount = NULL;
733 }
734 
735 static void
736 g_raid_md_jmicron_new_disk(struct g_raid_disk *disk)
737 {
738 	struct g_raid_softc *sc;
739 	struct g_raid_md_object *md;
740 	struct g_raid_md_jmicron_object *mdi;
741 	struct jmicron_raid_conf *pdmeta;
742 	struct g_raid_md_jmicron_perdisk *pd;
743 
744 	sc = disk->d_softc;
745 	md = sc->sc_md;
746 	mdi = (struct g_raid_md_jmicron_object *)md;
747 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
748 	pdmeta = pd->pd_meta;
749 
750 	if (mdi->mdio_started) {
751 		if (g_raid_md_jmicron_start_disk(disk))
752 			g_raid_md_write_jmicron(md, NULL, NULL, NULL);
753 	} else {
754 		/*
755 		 * If we haven't started yet - update common metadata
756 		 * to get subdisks details, avoiding data from spare disks.
757 		 */
758 		if (mdi->mdio_meta == NULL ||
759 		    jmicron_meta_find_disk(mdi->mdio_meta,
760 		     mdi->mdio_meta->disk_id) == -3) {
761 			if (mdi->mdio_meta != NULL)
762 				free(mdi->mdio_meta, M_MD_JMICRON);
763 			mdi->mdio_meta = jmicron_meta_copy(pdmeta);
764 			mdi->mdio_total_disks = jmicron_meta_total_disks(pdmeta);
765 		}
766 		mdi->mdio_meta->flags |= pdmeta->flags & JMICRON_F_BADSEC;
767 
768 		mdi->mdio_disks_present++;
769 		G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d+%d up)",
770 		    mdi->mdio_disks_present,
771 		    mdi->mdio_total_disks,
772 		    jmicron_meta_total_spare(mdi->mdio_meta));
773 
774 		/* If we collected all needed disks - start array. */
775 		if (mdi->mdio_disks_present == mdi->mdio_total_disks +
776 		    jmicron_meta_total_spare(mdi->mdio_meta))
777 			g_raid_md_jmicron_start(sc);
778 	}
779 }
780 
781 static void
782 g_raid_jmicron_go(void *arg)
783 {
784 	struct g_raid_softc *sc;
785 	struct g_raid_md_object *md;
786 	struct g_raid_md_jmicron_object *mdi;
787 
788 	sc = arg;
789 	md = sc->sc_md;
790 	mdi = (struct g_raid_md_jmicron_object *)md;
791 	if (!mdi->mdio_started) {
792 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
793 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
794 	}
795 }
796 
797 static int
798 g_raid_md_create_jmicron(struct g_raid_md_object *md, struct g_class *mp,
799     struct g_geom **gp)
800 {
801 	struct g_raid_softc *sc;
802 	struct g_raid_md_jmicron_object *mdi;
803 	char name[16];
804 
805 	mdi = (struct g_raid_md_jmicron_object *)md;
806 	mdi->mdio_config_id = arc4random();
807 	snprintf(name, sizeof(name), "JMicron-%08x", mdi->mdio_config_id);
808 	sc = g_raid_create_node(mp, name, md);
809 	if (sc == NULL)
810 		return (G_RAID_MD_TASTE_FAIL);
811 	md->mdo_softc = sc;
812 	*gp = sc->sc_geom;
813 	return (G_RAID_MD_TASTE_NEW);
814 }
815 
816 static int
817 g_raid_md_taste_jmicron(struct g_raid_md_object *md, struct g_class *mp,
818                               struct g_consumer *cp, struct g_geom **gp)
819 {
820 	struct g_consumer *rcp;
821 	struct g_provider *pp;
822 	struct g_raid_md_jmicron_object *mdi, *mdi1;
823 	struct g_raid_softc *sc;
824 	struct g_raid_disk *disk;
825 	struct jmicron_raid_conf *meta;
826 	struct g_raid_md_jmicron_perdisk *pd;
827 	struct g_geom *geom;
828 	int error, disk_pos, result, spare, len;
829 	char name[16];
830 	uint16_t vendor;
831 
832 	G_RAID_DEBUG(1, "Tasting JMicron on %s", cp->provider->name);
833 	mdi = (struct g_raid_md_jmicron_object *)md;
834 	pp = cp->provider;
835 
836 	/* Read metadata from device. */
837 	meta = NULL;
838 	vendor = 0xffff;
839 	if (g_access(cp, 1, 0, 0) != 0)
840 		return (G_RAID_MD_TASTE_FAIL);
841 	g_topology_unlock();
842 	len = 2;
843 	if (pp->geom->rank == 1)
844 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
845 	meta = jmicron_meta_read(cp);
846 	g_topology_lock();
847 	g_access(cp, -1, 0, 0);
848 	if (meta == NULL) {
849 		if (g_raid_aggressive_spare) {
850 			if (vendor == 0x197b) {
851 				G_RAID_DEBUG(1,
852 				    "No JMicron metadata, forcing spare.");
853 				spare = 2;
854 				goto search;
855 			} else {
856 				G_RAID_DEBUG(1,
857 				    "JMicron vendor mismatch 0x%04x != 0x197b",
858 				    vendor);
859 			}
860 		}
861 		return (G_RAID_MD_TASTE_FAIL);
862 	}
863 
864 	/* Check this disk position in obtained metadata. */
865 	disk_pos = jmicron_meta_find_disk(meta, meta->disk_id);
866 	if (disk_pos == -1) {
867 		G_RAID_DEBUG(1, "JMicron disk_id %08x not found",
868 		    meta->disk_id);
869 		goto fail1;
870 	}
871 
872 	/* Metadata valid. Print it. */
873 	g_raid_md_jmicron_print(meta);
874 	G_RAID_DEBUG(1, "JMicron disk position %d", disk_pos);
875 	spare = (disk_pos == -2) ? 1 : 0;
876 
877 search:
878 	/* Search for matching node. */
879 	sc = NULL;
880 	mdi1 = NULL;
881 	LIST_FOREACH(geom, &mp->geom, geom) {
882 		sc = geom->softc;
883 		if (sc == NULL)
884 			continue;
885 		if (sc->sc_stopping != 0)
886 			continue;
887 		if (sc->sc_md->mdo_class != md->mdo_class)
888 			continue;
889 		mdi1 = (struct g_raid_md_jmicron_object *)sc->sc_md;
890 		if (spare == 2) {
891 			if (mdi1->mdio_incomplete)
892 				break;
893 		} else {
894 			if (mdi1->mdio_config_id ==
895 			    jmicron_meta_config_id(meta))
896 				break;
897 		}
898 	}
899 
900 	/* Found matching node. */
901 	if (geom != NULL) {
902 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
903 		result = G_RAID_MD_TASTE_EXISTING;
904 
905 	} else if (spare) { /* Not found needy node -- left for later. */
906 		G_RAID_DEBUG(1, "Spare is not needed at this time");
907 		goto fail1;
908 
909 	} else { /* Not found matching node -- create one. */
910 		result = G_RAID_MD_TASTE_NEW;
911 		mdi->mdio_config_id = jmicron_meta_config_id(meta);
912 		snprintf(name, sizeof(name), "JMicron-%08x",
913 		    mdi->mdio_config_id);
914 		sc = g_raid_create_node(mp, name, md);
915 		md->mdo_softc = sc;
916 		geom = sc->sc_geom;
917 		callout_init(&mdi->mdio_start_co, 1);
918 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
919 		    g_raid_jmicron_go, sc);
920 		mdi->mdio_rootmount = root_mount_hold("GRAID-JMicron");
921 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
922 	}
923 
924 	rcp = g_new_consumer(geom);
925 	g_attach(rcp, pp);
926 	if (g_access(rcp, 1, 1, 1) != 0)
927 		; //goto fail1;
928 
929 	g_topology_unlock();
930 	sx_xlock(&sc->sc_lock);
931 
932 	pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
933 	pd->pd_meta = meta;
934 	if (spare == 2) {
935 		pd->pd_disk_pos = -3;
936 		pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK;
937 	} else {
938 		pd->pd_disk_pos = -1;
939 		pd->pd_disk_id = meta->disk_id;
940 	}
941 	pd->pd_disk_size = pp->mediasize;
942 	disk = g_raid_create_disk(sc);
943 	disk->d_md_data = (void *)pd;
944 	disk->d_consumer = rcp;
945 	rcp->private = disk;
946 
947 	/* Read kernel dumping information. */
948 	disk->d_kd.offset = 0;
949 	disk->d_kd.length = OFF_MAX;
950 	len = sizeof(disk->d_kd);
951 	error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd);
952 	if (disk->d_kd.di.dumper == NULL)
953 		G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.",
954 		    rcp->provider->name, error);
955 
956 	g_raid_md_jmicron_new_disk(disk);
957 
958 	sx_xunlock(&sc->sc_lock);
959 	g_topology_lock();
960 	*gp = geom;
961 	return (result);
962 fail1:
963 	free(meta, M_MD_JMICRON);
964 	return (G_RAID_MD_TASTE_FAIL);
965 }
966 
967 static int
968 g_raid_md_event_jmicron(struct g_raid_md_object *md,
969     struct g_raid_disk *disk, u_int event)
970 {
971 	struct g_raid_softc *sc;
972 	struct g_raid_subdisk *sd;
973 	struct g_raid_md_jmicron_object *mdi;
974 	struct g_raid_md_jmicron_perdisk *pd;
975 
976 	sc = md->mdo_softc;
977 	mdi = (struct g_raid_md_jmicron_object *)md;
978 	if (disk == NULL) {
979 		switch (event) {
980 		case G_RAID_NODE_E_START:
981 			if (!mdi->mdio_started)
982 				g_raid_md_jmicron_start(sc);
983 			return (0);
984 		}
985 		return (-1);
986 	}
987 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
988 	switch (event) {
989 	case G_RAID_DISK_E_DISCONNECTED:
990 		/* If disk was assigned, just update statuses. */
991 		if (pd->pd_disk_pos >= 0) {
992 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
993 			if (disk->d_consumer) {
994 				g_raid_kill_consumer(sc, disk->d_consumer);
995 				disk->d_consumer = NULL;
996 			}
997 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
998 				g_raid_change_subdisk_state(sd,
999 				    G_RAID_SUBDISK_S_NONE);
1000 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1001 				    G_RAID_EVENT_SUBDISK);
1002 			}
1003 		} else {
1004 			/* Otherwise -- delete. */
1005 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1006 			g_raid_destroy_disk(disk);
1007 		}
1008 
1009 		/* Write updated metadata to all disks. */
1010 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1011 
1012 		/* Check if anything left except placeholders. */
1013 		if (g_raid_ndisks(sc, -1) ==
1014 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1015 			g_raid_destroy_node(sc, 0);
1016 		else
1017 			g_raid_md_jmicron_refill(sc);
1018 		return (0);
1019 	}
1020 	return (-2);
1021 }
1022 
1023 static int
1024 g_raid_md_ctl_jmicron(struct g_raid_md_object *md,
1025     struct gctl_req *req)
1026 {
1027 	struct g_raid_softc *sc;
1028 	struct g_raid_volume *vol;
1029 	struct g_raid_subdisk *sd;
1030 	struct g_raid_disk *disk;
1031 	struct g_raid_md_jmicron_object *mdi;
1032 	struct g_raid_md_jmicron_perdisk *pd;
1033 	struct g_consumer *cp;
1034 	struct g_provider *pp;
1035 	char arg[16];
1036 	const char *verb, *volname, *levelname, *diskname;
1037 	int *nargs, *force;
1038 	off_t size, sectorsize, strip;
1039 	intmax_t *sizearg, *striparg;
1040 	int numdisks, i, len, level, qual, update;
1041 	int error;
1042 
1043 	sc = md->mdo_softc;
1044 	mdi = (struct g_raid_md_jmicron_object *)md;
1045 	verb = gctl_get_param(req, "verb", NULL);
1046 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1047 	error = 0;
1048 	if (strcmp(verb, "label") == 0) {
1049 
1050 		if (*nargs < 4) {
1051 			gctl_error(req, "Invalid number of arguments.");
1052 			return (-1);
1053 		}
1054 		volname = gctl_get_asciiparam(req, "arg1");
1055 		if (volname == NULL) {
1056 			gctl_error(req, "No volume name.");
1057 			return (-2);
1058 		}
1059 		levelname = gctl_get_asciiparam(req, "arg2");
1060 		if (levelname == NULL) {
1061 			gctl_error(req, "No RAID level.");
1062 			return (-3);
1063 		}
1064 		if (strcasecmp(levelname, "RAID5") == 0)
1065 			levelname = "RAID5-LA";
1066 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1067 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1068 			return (-4);
1069 		}
1070 		numdisks = *nargs - 3;
1071 		force = gctl_get_paraml(req, "force", sizeof(*force));
1072 		if (!g_raid_md_jmicron_supported(level, qual, numdisks,
1073 		    force ? *force : 0)) {
1074 			gctl_error(req, "Unsupported RAID level "
1075 			    "(0x%02x/0x%02x), or number of disks (%d).",
1076 			    level, qual, numdisks);
1077 			return (-5);
1078 		}
1079 
1080 		/* Search for disks, connect them and probe. */
1081 		size = 0x7fffffffffffffffllu;
1082 		sectorsize = 0;
1083 		for (i = 0; i < numdisks; i++) {
1084 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1085 			diskname = gctl_get_asciiparam(req, arg);
1086 			if (diskname == NULL) {
1087 				gctl_error(req, "No disk name (%s).", arg);
1088 				error = -6;
1089 				break;
1090 			}
1091 			if (strcmp(diskname, "NONE") == 0) {
1092 				cp = NULL;
1093 				pp = NULL;
1094 			} else {
1095 				g_topology_lock();
1096 				cp = g_raid_open_consumer(sc, diskname);
1097 				if (cp == NULL) {
1098 					gctl_error(req, "Can't open '%s'.",
1099 					    diskname);
1100 					g_topology_unlock();
1101 					error = -7;
1102 					break;
1103 				}
1104 				pp = cp->provider;
1105 			}
1106 			pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
1107 			pd->pd_disk_pos = i;
1108 			pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK;
1109 			disk = g_raid_create_disk(sc);
1110 			disk->d_md_data = (void *)pd;
1111 			disk->d_consumer = cp;
1112 			if (cp == NULL)
1113 				continue;
1114 			cp->private = disk;
1115 			g_topology_unlock();
1116 
1117 			/* Read kernel dumping information. */
1118 			disk->d_kd.offset = 0;
1119 			disk->d_kd.length = OFF_MAX;
1120 			len = sizeof(disk->d_kd);
1121 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1122 			if (disk->d_kd.di.dumper == NULL)
1123 				G_RAID_DEBUG1(2, sc,
1124 				    "Dumping not supported by %s.",
1125 				    cp->provider->name);
1126 
1127 			pd->pd_disk_size = pp->mediasize;
1128 			if (size > pp->mediasize)
1129 				size = pp->mediasize;
1130 			if (sectorsize < pp->sectorsize)
1131 				sectorsize = pp->sectorsize;
1132 		}
1133 		if (error != 0)
1134 			return (error);
1135 
1136 		if (sectorsize <= 0) {
1137 			gctl_error(req, "Can't get sector size.");
1138 			return (-8);
1139 		}
1140 
1141 		/* Reserve space for metadata. */
1142 		size -= sectorsize;
1143 
1144 		/* Handle size argument. */
1145 		len = sizeof(*sizearg);
1146 		sizearg = gctl_get_param(req, "size", &len);
1147 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1148 		    *sizearg > 0) {
1149 			if (*sizearg > size) {
1150 				gctl_error(req, "Size too big %lld > %lld.",
1151 				    (long long)*sizearg, (long long)size);
1152 				return (-9);
1153 			}
1154 			size = *sizearg;
1155 		}
1156 
1157 		/* Handle strip argument. */
1158 		strip = 131072;
1159 		len = sizeof(*striparg);
1160 		striparg = gctl_get_param(req, "strip", &len);
1161 		if (striparg != NULL && len == sizeof(*striparg) &&
1162 		    *striparg > 0) {
1163 			if (*striparg < sectorsize) {
1164 				gctl_error(req, "Strip size too small.");
1165 				return (-10);
1166 			}
1167 			if (*striparg % sectorsize != 0) {
1168 				gctl_error(req, "Incorrect strip size.");
1169 				return (-11);
1170 			}
1171 			if (strip > 65535 * sectorsize) {
1172 				gctl_error(req, "Strip size too big.");
1173 				return (-12);
1174 			}
1175 			strip = *striparg;
1176 		}
1177 
1178 		/* Round size down to strip or sector. */
1179 		if (level == G_RAID_VOLUME_RL_RAID1)
1180 			size -= (size % sectorsize);
1181 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1182 		    (numdisks & 1) != 0)
1183 			size -= (size % (2 * strip));
1184 		else
1185 			size -= (size % strip);
1186 		if (size <= 0) {
1187 			gctl_error(req, "Size too small.");
1188 			return (-13);
1189 		}
1190 		if (size > 0xffffffffffffllu * sectorsize) {
1191 			gctl_error(req, "Size too big.");
1192 			return (-14);
1193 		}
1194 
1195 		/* We have all we need, create things: volume, ... */
1196 		mdi->mdio_total_disks = numdisks;
1197 		mdi->mdio_started = 1;
1198 		vol = g_raid_create_volume(sc, volname, -1);
1199 		vol->v_md_data = (void *)(intptr_t)0;
1200 		vol->v_raid_level = level;
1201 		vol->v_raid_level_qualifier = qual;
1202 		vol->v_strip_size = strip;
1203 		vol->v_disks_count = numdisks;
1204 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1205 		    level == G_RAID_VOLUME_RL_CONCAT ||
1206 		    level == G_RAID_VOLUME_RL_SINGLE)
1207 			vol->v_mediasize = size * numdisks;
1208 		else if (level == G_RAID_VOLUME_RL_RAID1)
1209 			vol->v_mediasize = size;
1210 		else if (level == G_RAID_VOLUME_RL_RAID5)
1211 			vol->v_mediasize = size * (numdisks - 1);
1212 		else { /* RAID1E */
1213 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1214 			    strip;
1215 		}
1216 		vol->v_sectorsize = sectorsize;
1217 		g_raid_start_volume(vol);
1218 
1219 		/* , and subdisks. */
1220 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1221 			pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1222 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1223 			sd->sd_disk = disk;
1224 			sd->sd_offset = 0;
1225 			sd->sd_size = size;
1226 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1227 			if (sd->sd_disk->d_consumer != NULL) {
1228 				g_raid_change_disk_state(disk,
1229 				    G_RAID_DISK_S_ACTIVE);
1230 				g_raid_change_subdisk_state(sd,
1231 				    G_RAID_SUBDISK_S_ACTIVE);
1232 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1233 				    G_RAID_EVENT_SUBDISK);
1234 			} else {
1235 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1236 			}
1237 		}
1238 
1239 		/* Write metadata based on created entities. */
1240 		G_RAID_DEBUG1(0, sc, "Array started.");
1241 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1242 
1243 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1244 		g_raid_md_jmicron_refill(sc);
1245 
1246 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1247 		    G_RAID_EVENT_VOLUME);
1248 		return (0);
1249 	}
1250 	if (strcmp(verb, "delete") == 0) {
1251 
1252 		/* Check if some volume is still open. */
1253 		force = gctl_get_paraml(req, "force", sizeof(*force));
1254 		if (force != NULL && *force == 0 &&
1255 		    g_raid_nopens(sc) != 0) {
1256 			gctl_error(req, "Some volume is still open.");
1257 			return (-4);
1258 		}
1259 
1260 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1261 			if (disk->d_consumer)
1262 				jmicron_meta_erase(disk->d_consumer);
1263 		}
1264 		g_raid_destroy_node(sc, 0);
1265 		return (0);
1266 	}
1267 	if (strcmp(verb, "remove") == 0 ||
1268 	    strcmp(verb, "fail") == 0) {
1269 		if (*nargs < 2) {
1270 			gctl_error(req, "Invalid number of arguments.");
1271 			return (-1);
1272 		}
1273 		for (i = 1; i < *nargs; i++) {
1274 			snprintf(arg, sizeof(arg), "arg%d", i);
1275 			diskname = gctl_get_asciiparam(req, arg);
1276 			if (diskname == NULL) {
1277 				gctl_error(req, "No disk name (%s).", arg);
1278 				error = -2;
1279 				break;
1280 			}
1281 			if (strncmp(diskname, "/dev/", 5) == 0)
1282 				diskname += 5;
1283 
1284 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1285 				if (disk->d_consumer != NULL &&
1286 				    disk->d_consumer->provider != NULL &&
1287 				    strcmp(disk->d_consumer->provider->name,
1288 				     diskname) == 0)
1289 					break;
1290 			}
1291 			if (disk == NULL) {
1292 				gctl_error(req, "Disk '%s' not found.",
1293 				    diskname);
1294 				error = -3;
1295 				break;
1296 			}
1297 
1298 			if (strcmp(verb, "fail") == 0) {
1299 				g_raid_md_fail_disk_jmicron(md, NULL, disk);
1300 				continue;
1301 			}
1302 
1303 			pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1304 
1305 			/* Erase metadata on deleting disk. */
1306 			jmicron_meta_erase(disk->d_consumer);
1307 
1308 			/* If disk was assigned, just update statuses. */
1309 			if (pd->pd_disk_pos >= 0) {
1310 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1311 				g_raid_kill_consumer(sc, disk->d_consumer);
1312 				disk->d_consumer = NULL;
1313 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1314 					g_raid_change_subdisk_state(sd,
1315 					    G_RAID_SUBDISK_S_NONE);
1316 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1317 					    G_RAID_EVENT_SUBDISK);
1318 				}
1319 			} else {
1320 				/* Otherwise -- delete. */
1321 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1322 				g_raid_destroy_disk(disk);
1323 			}
1324 		}
1325 
1326 		/* Write updated metadata to remaining disks. */
1327 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1328 
1329 		/* Check if anything left except placeholders. */
1330 		if (g_raid_ndisks(sc, -1) ==
1331 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1332 			g_raid_destroy_node(sc, 0);
1333 		else
1334 			g_raid_md_jmicron_refill(sc);
1335 		return (error);
1336 	}
1337 	if (strcmp(verb, "insert") == 0) {
1338 		if (*nargs < 2) {
1339 			gctl_error(req, "Invalid number of arguments.");
1340 			return (-1);
1341 		}
1342 		update = 0;
1343 		for (i = 1; i < *nargs; i++) {
1344 			/* Get disk name. */
1345 			snprintf(arg, sizeof(arg), "arg%d", i);
1346 			diskname = gctl_get_asciiparam(req, arg);
1347 			if (diskname == NULL) {
1348 				gctl_error(req, "No disk name (%s).", arg);
1349 				error = -3;
1350 				break;
1351 			}
1352 
1353 			/* Try to find provider with specified name. */
1354 			g_topology_lock();
1355 			cp = g_raid_open_consumer(sc, diskname);
1356 			if (cp == NULL) {
1357 				gctl_error(req, "Can't open disk '%s'.",
1358 				    diskname);
1359 				g_topology_unlock();
1360 				error = -4;
1361 				break;
1362 			}
1363 			pp = cp->provider;
1364 
1365 			pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
1366 			pd->pd_disk_pos = -3;
1367 			pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK;
1368 			pd->pd_disk_size = pp->mediasize;
1369 
1370 			disk = g_raid_create_disk(sc);
1371 			disk->d_consumer = cp;
1372 			disk->d_md_data = (void *)pd;
1373 			cp->private = disk;
1374 			g_topology_unlock();
1375 
1376 			/* Read kernel dumping information. */
1377 			disk->d_kd.offset = 0;
1378 			disk->d_kd.length = OFF_MAX;
1379 			len = sizeof(disk->d_kd);
1380 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1381 			if (disk->d_kd.di.dumper == NULL)
1382 				G_RAID_DEBUG1(2, sc,
1383 				    "Dumping not supported by %s.",
1384 				    cp->provider->name);
1385 
1386 			/* Welcome the "new" disk. */
1387 			update += g_raid_md_jmicron_start_disk(disk);
1388 			if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1389 			    disk->d_state != G_RAID_DISK_S_SPARE) {
1390 				gctl_error(req, "Disk '%s' doesn't fit.",
1391 				    diskname);
1392 				g_raid_destroy_disk(disk);
1393 				error = -8;
1394 				break;
1395 			}
1396 		}
1397 
1398 		/* Write new metadata if we changed something. */
1399 		if (update)
1400 			g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1401 		return (error);
1402 	}
1403 	gctl_error(req, "Command '%s' is not supported.", verb);
1404 	return (-100);
1405 }
1406 
1407 static int
1408 g_raid_md_write_jmicron(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1409     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1410 {
1411 	struct g_raid_softc *sc;
1412 	struct g_raid_volume *vol;
1413 	struct g_raid_subdisk *sd;
1414 	struct g_raid_disk *disk;
1415 	struct g_raid_md_jmicron_object *mdi;
1416 	struct g_raid_md_jmicron_perdisk *pd;
1417 	struct jmicron_raid_conf *meta;
1418 	int i, spares;
1419 
1420 	sc = md->mdo_softc;
1421 	mdi = (struct g_raid_md_jmicron_object *)md;
1422 
1423 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1424 		return (0);
1425 
1426 	/* There is only one volume. */
1427 	vol = TAILQ_FIRST(&sc->sc_volumes);
1428 
1429 	/* Fill global fields. */
1430 	meta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK | M_ZERO);
1431 	strncpy(meta->signature, JMICRON_MAGIC, 2);
1432 	meta->version = JMICRON_VERSION;
1433 	jmicron_meta_put_name(meta, vol->v_name);
1434 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
1435 		meta->type = JMICRON_T_RAID0;
1436 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1437 		meta->type = JMICRON_T_RAID1;
1438 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1439 		meta->type = JMICRON_T_RAID01;
1440 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1441 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1442 		meta->type = JMICRON_T_CONCAT;
1443 	else
1444 		meta->type = JMICRON_T_RAID5;
1445 	meta->stripe_shift = fls(vol->v_strip_size / 2048);
1446 	meta->flags = JMICRON_F_READY | JMICRON_F_BOOTABLE;
1447 	for (i = 0; i < vol->v_disks_count; i++) {
1448 		sd = &vol->v_subdisks[i];
1449 		if (sd->sd_disk == NULL || sd->sd_disk->d_md_data == NULL)
1450 			meta->disks[i] = 0xffffffff;
1451 		else {
1452 			pd = (struct g_raid_md_jmicron_perdisk *)
1453 			    sd->sd_disk->d_md_data;
1454 			meta->disks[i] = pd->pd_disk_id;
1455 		}
1456 		if (sd->sd_state < G_RAID_SUBDISK_S_STALE)
1457 			meta->flags |= JMICRON_F_BADSEC;
1458 		if (vol->v_dirty)
1459 			meta->flags |= JMICRON_F_UNSYNC;
1460 	}
1461 
1462 	/* Put spares to their slots. */
1463 	spares = 0;
1464 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1465 		pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1466 		if (disk->d_state != G_RAID_DISK_S_SPARE)
1467 			continue;
1468 		meta->spare[spares] = pd->pd_disk_id;
1469 		if (++spares >= 2)
1470 			break;
1471 	}
1472 
1473 	/* We are done. Print meta data and store them to disks. */
1474 	if (mdi->mdio_meta != NULL)
1475 		free(mdi->mdio_meta, M_MD_JMICRON);
1476 	mdi->mdio_meta = meta;
1477 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1478 		pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1479 		if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1480 		    disk->d_state != G_RAID_DISK_S_SPARE)
1481 			continue;
1482 		if (pd->pd_meta != NULL) {
1483 			free(pd->pd_meta, M_MD_JMICRON);
1484 			pd->pd_meta = NULL;
1485 		}
1486 		pd->pd_meta = jmicron_meta_copy(meta);
1487 		pd->pd_meta->disk_id = pd->pd_disk_id;
1488 		if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1489 			pd->pd_meta->offset =
1490 			    (sd->sd_offset / 512) / 16;
1491 			pd->pd_meta->disk_sectors_high =
1492 			    (sd->sd_size / 512) >> 16;
1493 			pd->pd_meta->disk_sectors_low =
1494 			    (sd->sd_size / 512) & 0xffff;
1495 			if (sd->sd_state < G_RAID_SUBDISK_S_STALE)
1496 				pd->pd_meta->flags &= ~JMICRON_F_BADSEC;
1497 			else if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE)
1498 				pd->pd_meta->flags |= JMICRON_F_UNSYNC;
1499 		}
1500 		G_RAID_DEBUG(1, "Writing JMicron metadata to %s",
1501 		    g_raid_get_diskname(disk));
1502 		g_raid_md_jmicron_print(pd->pd_meta);
1503 		jmicron_meta_write(disk->d_consumer, pd->pd_meta);
1504 	}
1505 	return (0);
1506 }
1507 
1508 static int
1509 g_raid_md_fail_disk_jmicron(struct g_raid_md_object *md,
1510     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1511 {
1512 	struct g_raid_softc *sc;
1513 	struct g_raid_md_jmicron_perdisk *pd;
1514 	struct g_raid_subdisk *sd;
1515 
1516 	sc = md->mdo_softc;
1517 	pd = (struct g_raid_md_jmicron_perdisk *)tdisk->d_md_data;
1518 
1519 	/* We can't fail disk that is not a part of array now. */
1520 	if (pd->pd_disk_pos < 0)
1521 		return (-1);
1522 
1523 	if (tdisk->d_consumer)
1524 		jmicron_meta_erase(tdisk->d_consumer);
1525 
1526 	/* Change states. */
1527 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1528 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1529 		g_raid_change_subdisk_state(sd,
1530 		    G_RAID_SUBDISK_S_FAILED);
1531 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1532 		    G_RAID_EVENT_SUBDISK);
1533 	}
1534 
1535 	/* Write updated metadata to remaining disks. */
1536 	g_raid_md_write_jmicron(md, NULL, NULL, tdisk);
1537 
1538 	/* Check if anything left except placeholders. */
1539 	if (g_raid_ndisks(sc, -1) ==
1540 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1541 		g_raid_destroy_node(sc, 0);
1542 	else
1543 		g_raid_md_jmicron_refill(sc);
1544 	return (0);
1545 }
1546 
1547 static int
1548 g_raid_md_free_disk_jmicron(struct g_raid_md_object *md,
1549     struct g_raid_disk *disk)
1550 {
1551 	struct g_raid_md_jmicron_perdisk *pd;
1552 
1553 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1554 	if (pd->pd_meta != NULL) {
1555 		free(pd->pd_meta, M_MD_JMICRON);
1556 		pd->pd_meta = NULL;
1557 	}
1558 	free(pd, M_MD_JMICRON);
1559 	disk->d_md_data = NULL;
1560 	return (0);
1561 }
1562 
1563 static int
1564 g_raid_md_free_jmicron(struct g_raid_md_object *md)
1565 {
1566 	struct g_raid_md_jmicron_object *mdi;
1567 
1568 	mdi = (struct g_raid_md_jmicron_object *)md;
1569 	if (!mdi->mdio_started) {
1570 		mdi->mdio_started = 0;
1571 		callout_stop(&mdi->mdio_start_co);
1572 		G_RAID_DEBUG1(1, md->mdo_softc,
1573 		    "root_mount_rel %p", mdi->mdio_rootmount);
1574 		root_mount_rel(mdi->mdio_rootmount);
1575 		mdi->mdio_rootmount = NULL;
1576 	}
1577 	if (mdi->mdio_meta != NULL) {
1578 		free(mdi->mdio_meta, M_MD_JMICRON);
1579 		mdi->mdio_meta = NULL;
1580 	}
1581 	return (0);
1582 }
1583 
1584 G_RAID_MD_DECLARE(g_raid_md_jmicron);
1585