xref: /freebsd/sys/geom/raid/md_promise.c (revision 8ce070c1b28cd5f33c098da43378d0239091bd00)
1 /*-
2  * Copyright (c) 2011 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/bio.h>
32 #include <sys/endian.h>
33 #include <sys/kernel.h>
34 #include <sys/kobj.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/systm.h>
40 #include <geom/geom.h>
41 #include "geom/raid/g_raid.h"
42 #include "g_raid_md_if.h"
43 
44 static MALLOC_DEFINE(M_MD_PROMISE, "md_promise_data", "GEOM_RAID Promise metadata");
45 
46 #define	PROMISE_MAX_DISKS	8
47 #define	PROMISE_MAX_SUBDISKS	2
48 #define	PROMISE_META_OFFSET	14
49 
50 struct promise_raid_disk {
51 	uint8_t		flags;			/* Subdisk status. */
52 #define PROMISE_F_VALID		0x01
53 #define PROMISE_F_ONLINE	0x02
54 #define PROMISE_F_ASSIGNED	0x04
55 #define PROMISE_F_SPARE		0x08
56 #define PROMISE_F_DUPLICATE	0x10
57 #define PROMISE_F_REDIR		0x20
58 #define PROMISE_F_DOWN		0x40
59 #define PROMISE_F_READY		0x80
60 
61 	uint8_t		number;			/* Position in a volume. */
62 	uint8_t		channel;		/* ATA channel number. */
63 	uint8_t		device;			/* ATA device number. */
64 	uint64_t	id __packed;		/* Subdisk ID. */
65 } __packed;
66 
67 struct promise_raid_conf {
68 	char		promise_id[24];
69 #define PROMISE_MAGIC		"Promise Technology, Inc."
70 #define FREEBSD_MAGIC		"FreeBSD ATA driver RAID "
71 
72 	uint32_t	dummy_0;
73 	uint64_t	magic_0;
74 #define PROMISE_MAGIC0(x)	(((uint64_t)(x.channel) << 48) | \
75 				((uint64_t)(x.device != 0) << 56))
76 	uint16_t	magic_1;
77 	uint32_t	magic_2;
78 	uint8_t		filler1[470];
79 
80 	uint32_t	integrity;
81 #define PROMISE_I_VALID		0x00000080
82 
83 	struct promise_raid_disk	disk;	/* This subdisk info. */
84 	uint32_t	disk_offset;		/* Subdisk offset. */
85 	uint32_t	disk_sectors;		/* Subdisk size */
86 	uint32_t	rebuild_lba;		/* Rebuild position. */
87 	uint16_t	generation;		/* Generation number. */
88 	uint8_t		status;			/* Volume status. */
89 #define PROMISE_S_VALID		0x01
90 #define PROMISE_S_ONLINE	0x02
91 #define PROMISE_S_INITED	0x04
92 #define PROMISE_S_READY		0x08
93 #define PROMISE_S_DEGRADED	0x10
94 #define PROMISE_S_MARKED	0x20
95 #define PROMISE_S_MIGRATING	0x40
96 #define PROMISE_S_FUNCTIONAL	0x80
97 
98 	uint8_t		type;			/* Voluem type. */
99 #define PROMISE_T_RAID0		0x00
100 #define PROMISE_T_RAID1		0x01
101 #define PROMISE_T_RAID3		0x02
102 #define PROMISE_T_RAID5		0x04
103 #define PROMISE_T_SPAN		0x08
104 #define PROMISE_T_JBOD		0x10
105 
106 	uint8_t		total_disks;		/* Disks in this volume. */
107 	uint8_t		stripe_shift;		/* Strip size. */
108 	uint8_t		array_width;		/* Number of RAID0 stripes. */
109 	uint8_t		array_number;		/* Global volume number. */
110 	uint32_t	total_sectors;		/* Volume size. */
111 	uint16_t	cylinders;		/* Volume geometry: C. */
112 	uint8_t		heads;			/* Volume geometry: H. */
113 	uint8_t		sectors;		/* Volume geometry: S. */
114 	uint64_t	volume_id __packed;	/* Volume ID, */
115 	struct promise_raid_disk	disks[PROMISE_MAX_DISKS];
116 						/* Subdisks in this volume. */
117 	char		name[32];		/* Volume label. */
118 
119 	uint32_t	filler2[8];
120 	uint32_t	magic_3;	/* Something related to rebuild. */
121 	uint64_t	rebuild_lba64;	/* Per-volume rebuild position. */
122 	uint32_t	magic_4;
123 	uint32_t	magic_5;
124 	uint32_t	total_sectors_high;
125 	uint32_t	filler3[324];
126 	uint32_t	checksum;
127 } __packed;
128 
129 struct g_raid_md_promise_perdisk {
130 	int		 pd_updated;
131 	int		 pd_subdisks;
132 	struct promise_raid_conf	*pd_meta[PROMISE_MAX_SUBDISKS];
133 };
134 
135 struct g_raid_md_promise_pervolume {
136 	struct promise_raid_conf	*pv_meta;
137 	uint64_t			 pv_id;
138 	uint16_t			 pv_generation;
139 	int				 pv_disks_present;
140 	int				 pv_started;
141 	struct callout			 pv_start_co;	/* STARTING state timer. */
142 };
143 
144 static g_raid_md_create_t g_raid_md_create_promise;
145 static g_raid_md_taste_t g_raid_md_taste_promise;
146 static g_raid_md_event_t g_raid_md_event_promise;
147 static g_raid_md_volume_event_t g_raid_md_volume_event_promise;
148 static g_raid_md_ctl_t g_raid_md_ctl_promise;
149 static g_raid_md_write_t g_raid_md_write_promise;
150 static g_raid_md_fail_disk_t g_raid_md_fail_disk_promise;
151 static g_raid_md_free_disk_t g_raid_md_free_disk_promise;
152 static g_raid_md_free_volume_t g_raid_md_free_volume_promise;
153 static g_raid_md_free_t g_raid_md_free_promise;
154 
155 static kobj_method_t g_raid_md_promise_methods[] = {
156 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_promise),
157 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_promise),
158 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_promise),
159 	KOBJMETHOD(g_raid_md_volume_event,	g_raid_md_volume_event_promise),
160 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_promise),
161 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_promise),
162 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_promise),
163 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_promise),
164 	KOBJMETHOD(g_raid_md_free_volume,	g_raid_md_free_volume_promise),
165 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_promise),
166 	{ 0, 0 }
167 };
168 
169 static struct g_raid_md_class g_raid_md_promise_class = {
170 	"Promise",
171 	g_raid_md_promise_methods,
172 	sizeof(struct g_raid_md_object),
173 	.mdc_priority = 100
174 };
175 
176 
177 static void
178 g_raid_md_promise_print(struct promise_raid_conf *meta)
179 {
180 	int i;
181 
182 	if (g_raid_debug < 1)
183 		return;
184 
185 	printf("********* ATA Promise Metadata *********\n");
186 	printf("promise_id          <%.24s>\n", meta->promise_id);
187 	printf("disk                %02x %02x %02x %02x %016jx\n",
188 	    meta->disk.flags, meta->disk.number, meta->disk.channel,
189 	    meta->disk.device, meta->disk.id);
190 	printf("disk_offset         %u\n", meta->disk_offset);
191 	printf("disk_sectors        %u\n", meta->disk_sectors);
192 	printf("rebuild_lba         %u\n", meta->rebuild_lba);
193 	printf("generation          %u\n", meta->generation);
194 	printf("status              0x%02x\n", meta->status);
195 	printf("type                %u\n", meta->type);
196 	printf("total_disks         %u\n", meta->total_disks);
197 	printf("stripe_shift        %u\n", meta->stripe_shift);
198 	printf("array_width         %u\n", meta->array_width);
199 	printf("array_number        %u\n", meta->array_number);
200 	printf("total_sectors       %u\n", meta->total_sectors);
201 	printf("cylinders           %u\n", meta->cylinders);
202 	printf("heads               %u\n", meta->heads);
203 	printf("sectors             %u\n", meta->sectors);
204 	printf("volume_id           0x%016jx\n", meta->volume_id);
205 	printf("disks:\n");
206 	for (i = 0; i < PROMISE_MAX_DISKS; i++ ) {
207 		printf("                    %02x %02x %02x %02x %016jx\n",
208 		    meta->disks[i].flags, meta->disks[i].number,
209 		    meta->disks[i].channel, meta->disks[i].device,
210 		    meta->disks[i].id);
211 	}
212 	printf("name                <%.32s>\n", meta->name);
213 	printf("magic_3             0x%08x\n", meta->magic_3);
214 	printf("rebuild_lba64       %ju\n", meta->rebuild_lba64);
215 	printf("magic_4             0x%08x\n", meta->magic_4);
216 	printf("magic_5             0x%08x\n", meta->magic_5);
217 	printf("total_sectors_high  0x%08x\n", meta->total_sectors_high);
218 	printf("=================================================\n");
219 }
220 
221 static struct promise_raid_conf *
222 promise_meta_copy(struct promise_raid_conf *meta)
223 {
224 	struct promise_raid_conf *nmeta;
225 
226 	nmeta = malloc(sizeof(*nmeta), M_MD_PROMISE, M_WAITOK);
227 	memcpy(nmeta, meta, sizeof(*nmeta));
228 	return (nmeta);
229 }
230 
231 static int
232 promise_meta_find_disk(struct promise_raid_conf *meta, uint64_t id)
233 {
234 	int pos;
235 
236 	for (pos = 0; pos < meta->total_disks; pos++) {
237 		if (meta->disks[pos].id == id)
238 			return (pos);
239 	}
240 	return (-1);
241 }
242 
243 static int
244 promise_meta_unused_range(struct promise_raid_conf **metaarr, int nsd,
245     uint32_t sectors, uint32_t *off, uint32_t *size)
246 {
247 	uint32_t coff, csize;
248 	int i, j;
249 
250 	sectors -= 131072;
251 	*off = 0;
252 	*size = 0;
253 	coff = 0;
254 	csize = sectors;
255 	i = 0;
256 	while (1) {
257 		for (j = 0; j < nsd; j++) {
258 			if (metaarr[j]->disk_offset >= coff) {
259 				csize = MIN(csize,
260 				    metaarr[j]->disk_offset - coff);
261 			}
262 		}
263 		if (csize > *size) {
264 			*off = coff;
265 			*size = csize;
266 		}
267 		if (i >= nsd)
268 			break;
269 		coff = metaarr[i]->disk_offset + metaarr[i]->disk_sectors;
270 		csize = sectors - coff;
271 		i++;
272 	};
273 	return ((*size > 0) ? 1 : 0);
274 }
275 
276 static int
277 promise_meta_translate_disk(struct g_raid_volume *vol, int md_disk_pos)
278 {
279 	int disk_pos, width;
280 
281 	if (md_disk_pos >= 0 && vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
282 		width = vol->v_disks_count / 2;
283 		disk_pos = (md_disk_pos / width) +
284 		    (md_disk_pos % width) * width;
285 	} else
286 		disk_pos = md_disk_pos;
287 	return (disk_pos);
288 }
289 
290 static void
291 promise_meta_get_name(struct promise_raid_conf *meta, char *buf)
292 {
293 	int i;
294 
295 	strncpy(buf, meta->name, 32);
296 	buf[32] = 0;
297 	for (i = 31; i >= 0; i--) {
298 		if (buf[i] > 0x20)
299 			break;
300 		buf[i] = 0;
301 	}
302 }
303 
304 static void
305 promise_meta_put_name(struct promise_raid_conf *meta, char *buf)
306 {
307 
308 	memset(meta->name, 0x20, 32);
309 	memcpy(meta->name, buf, MIN(strlen(buf), 32));
310 }
311 
312 static int
313 promise_meta_read(struct g_consumer *cp, struct promise_raid_conf **metaarr)
314 {
315 	struct g_provider *pp;
316 	struct promise_raid_conf *meta;
317 	char *buf;
318 	int error, i, subdisks;
319 	uint32_t checksum, *ptr;
320 
321 	pp = cp->provider;
322 	subdisks = 0;
323 next:
324 	/* Read metadata block. */
325 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize *
326 	    (63 - subdisks * PROMISE_META_OFFSET),
327 	    pp->sectorsize * 4, &error);
328 	if (buf == NULL) {
329 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
330 		    pp->name, error);
331 		return (subdisks);
332 	}
333 	meta = (struct promise_raid_conf *)buf;
334 
335 	/* Check if this is an Promise RAID struct */
336 	if (strncmp(meta->promise_id, PROMISE_MAGIC, strlen(PROMISE_MAGIC)) &&
337 	    strncmp(meta->promise_id, FREEBSD_MAGIC, strlen(FREEBSD_MAGIC))) {
338 		if (subdisks == 0)
339 			G_RAID_DEBUG(1,
340 			    "Promise signature check failed on %s", pp->name);
341 		g_free(buf);
342 		return (subdisks);
343 	}
344 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK);
345 	memcpy(meta, buf, MIN(sizeof(*meta), pp->sectorsize * 4));
346 	g_free(buf);
347 
348 	/* Check metadata checksum. */
349 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
350 		checksum += *ptr++;
351 	if (checksum != meta->checksum) {
352 		G_RAID_DEBUG(1, "Promise checksum check failed on %s", pp->name);
353 		free(meta, M_MD_PROMISE);
354 		return (subdisks);
355 	}
356 
357 	if ((meta->integrity & PROMISE_I_VALID) == 0) {
358 		G_RAID_DEBUG(1, "Promise metadata is invalid on %s", pp->name);
359 		free(meta, M_MD_PROMISE);
360 		return (subdisks);
361 	}
362 
363 	if (meta->total_disks > PROMISE_MAX_DISKS) {
364 		G_RAID_DEBUG(1, "Wrong number of disks on %s (%d)",
365 		    pp->name, meta->total_disks);
366 		free(meta, M_MD_PROMISE);
367 		return (subdisks);
368 	}
369 
370 	/* Save this part and look for next. */
371 	*metaarr = meta;
372 	metaarr++;
373 	subdisks++;
374 	if (subdisks < PROMISE_MAX_SUBDISKS)
375 		goto next;
376 
377 	return (subdisks);
378 }
379 
380 static int
381 promise_meta_write(struct g_consumer *cp,
382     struct promise_raid_conf **metaarr, int nsd)
383 {
384 	struct g_provider *pp;
385 	struct promise_raid_conf *meta;
386 	char *buf;
387 	int error, i, subdisk, fake;
388 	uint32_t checksum, *ptr, off, size;
389 
390 	pp = cp->provider;
391 	subdisk = 0;
392 	fake = 0;
393 next:
394 	buf = malloc(pp->sectorsize * 4, M_MD_PROMISE, M_WAITOK | M_ZERO);
395 	meta = NULL;
396 	if (subdisk < nsd) {
397 		meta = metaarr[subdisk];
398 	} else if (!fake && promise_meta_unused_range(metaarr, nsd,
399 	    cp->provider->mediasize / cp->provider->sectorsize,
400 	    &off, &size)) {
401 		/* Optionally add record for unused space. */
402 		meta = (struct promise_raid_conf *)buf;
403 		memcpy(&meta->promise_id[0], PROMISE_MAGIC,
404 		    sizeof(PROMISE_MAGIC) - 1);
405 		meta->dummy_0 = 0x00020000;
406 		meta->integrity = PROMISE_I_VALID;
407 		meta->disk.flags = PROMISE_F_ONLINE | PROMISE_F_VALID;
408 		meta->disk.number = 0xff;
409 		arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
410 		meta->disk_offset = off;
411 		meta->disk_sectors = size;
412 		meta->rebuild_lba = UINT32_MAX;
413 		fake = 1;
414 	}
415 	if (meta != NULL) {
416 		/* Recalculate checksum for case if metadata were changed. */
417 		meta->checksum = 0;
418 		for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
419 			checksum += *ptr++;
420 		meta->checksum = checksum;
421 		memcpy(buf, meta, MIN(pp->sectorsize * 4, sizeof(*meta)));
422 	}
423 	error = g_write_data(cp, pp->mediasize - pp->sectorsize *
424 	    (63 - subdisk * PROMISE_META_OFFSET),
425 	    buf, pp->sectorsize * 4);
426 	if (error != 0) {
427 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
428 		    pp->name, error);
429 	}
430 	free(buf, M_MD_PROMISE);
431 
432 	subdisk++;
433 	if (subdisk < PROMISE_MAX_SUBDISKS)
434 		goto next;
435 
436 	return (error);
437 }
438 
439 static int
440 promise_meta_erase(struct g_consumer *cp)
441 {
442 	struct g_provider *pp;
443 	char *buf;
444 	int error, subdisk;
445 
446 	pp = cp->provider;
447 	buf = malloc(4 * pp->sectorsize, M_MD_PROMISE, M_WAITOK | M_ZERO);
448 	for (subdisk = 0; subdisk < PROMISE_MAX_SUBDISKS; subdisk++) {
449 		error = g_write_data(cp, pp->mediasize - pp->sectorsize *
450 		    (63 - subdisk * PROMISE_META_OFFSET),
451 		    buf, 4 * pp->sectorsize);
452 		if (error != 0) {
453 			G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
454 			    pp->name, error);
455 		}
456 	}
457 	free(buf, M_MD_PROMISE);
458 	return (error);
459 }
460 
461 static int
462 promise_meta_write_spare(struct g_consumer *cp)
463 {
464 	struct promise_raid_conf *meta;
465 	int error;
466 
467 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
468 	memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC) - 1);
469 	meta->dummy_0 = 0x00020000;
470 	meta->integrity = PROMISE_I_VALID;
471 	meta->disk.flags = PROMISE_F_SPARE | PROMISE_F_ONLINE | PROMISE_F_VALID;
472 	meta->disk.number = 0xff;
473 	arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
474 	meta->disk_sectors = cp->provider->mediasize / cp->provider->sectorsize;
475 	meta->disk_sectors -= 131072;
476 	meta->rebuild_lba = UINT32_MAX;
477 	error = promise_meta_write(cp, &meta, 1);
478 	free(meta, M_MD_PROMISE);
479 	return (error);
480 }
481 
482 static struct g_raid_volume *
483 g_raid_md_promise_get_volume(struct g_raid_softc *sc, uint64_t id)
484 {
485 	struct g_raid_volume	*vol;
486 	struct g_raid_md_promise_pervolume *pv;
487 
488 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
489 		pv = vol->v_md_data;
490 		if (pv->pv_id == id)
491 			break;
492 	}
493 	return (vol);
494 }
495 
496 static int
497 g_raid_md_promise_purge_volumes(struct g_raid_softc *sc)
498 {
499 	struct g_raid_volume	*vol, *tvol;
500 	struct g_raid_md_promise_pervolume *pv;
501 	int i, res;
502 
503 	res = 0;
504 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tvol) {
505 		pv = vol->v_md_data;
506 		if (!pv->pv_started || vol->v_stopping)
507 			continue;
508 		for (i = 0; i < vol->v_disks_count; i++) {
509 			if (vol->v_subdisks[i].sd_state != G_RAID_SUBDISK_S_NONE)
510 				break;
511 		}
512 		if (i >= vol->v_disks_count) {
513 			g_raid_destroy_volume(vol);
514 			res = 1;
515 		}
516 	}
517 	return (res);
518 }
519 
520 static int
521 g_raid_md_promise_purge_disks(struct g_raid_softc *sc)
522 {
523 	struct g_raid_disk	*disk, *tdisk;
524 	struct g_raid_volume	*vol;
525 	struct g_raid_md_promise_perdisk *pd;
526 	int i, j, res;
527 
528 	res = 0;
529 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) {
530 		if (disk->d_state == G_RAID_DISK_S_SPARE)
531 			continue;
532 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
533 
534 		/* Scan for deleted volumes. */
535 		for (i = 0; i < pd->pd_subdisks; ) {
536 			vol = g_raid_md_promise_get_volume(sc,
537 			    pd->pd_meta[i]->volume_id);
538 			if (vol != NULL && !vol->v_stopping) {
539 				i++;
540 				continue;
541 			}
542 			free(pd->pd_meta[i], M_MD_PROMISE);
543 			for (j = i; j < pd->pd_subdisks - 1; j++)
544 				pd->pd_meta[j] = pd->pd_meta[j + 1];
545 			pd->pd_meta[PROMISE_MAX_SUBDISKS - 1] = NULL;
546 			pd->pd_subdisks--;
547 			pd->pd_updated = 1;
548 		}
549 
550 		/* If there is no metadata left - erase and delete disk. */
551 		if (pd->pd_subdisks == 0) {
552 			promise_meta_erase(disk->d_consumer);
553 			g_raid_destroy_disk(disk);
554 			res = 1;
555 		}
556 	}
557 	return (res);
558 }
559 
560 static int
561 g_raid_md_promise_supported(int level, int qual, int disks, int force)
562 {
563 
564 	if (disks > PROMISE_MAX_DISKS)
565 		return (0);
566 	switch (level) {
567 	case G_RAID_VOLUME_RL_RAID0:
568 		if (disks < 1)
569 			return (0);
570 		if (!force && disks < 2)
571 			return (0);
572 		break;
573 	case G_RAID_VOLUME_RL_RAID1:
574 		if (disks < 1)
575 			return (0);
576 		if (!force && (disks != 2))
577 			return (0);
578 		break;
579 	case G_RAID_VOLUME_RL_RAID1E:
580 		if (disks < 2)
581 			return (0);
582 		if (disks % 2 != 0)
583 			return (0);
584 		if (!force && (disks != 4))
585 			return (0);
586 		break;
587 	case G_RAID_VOLUME_RL_SINGLE:
588 		if (disks != 1)
589 			return (0);
590 		break;
591 	case G_RAID_VOLUME_RL_CONCAT:
592 		if (disks < 2)
593 			return (0);
594 		break;
595 	case G_RAID_VOLUME_RL_RAID5:
596 		if (disks < 3)
597 			return (0);
598 		break;
599 	default:
600 		return (0);
601 	}
602 	if (qual != G_RAID_VOLUME_RLQ_NONE)
603 		return (0);
604 	return (1);
605 }
606 
607 static int
608 g_raid_md_promise_start_disk(struct g_raid_disk *disk, int sdn,
609     struct g_raid_volume *vol)
610 {
611 	struct g_raid_softc *sc;
612 	struct g_raid_subdisk *sd;
613 	struct g_raid_md_promise_perdisk *pd;
614 	struct g_raid_md_promise_pervolume *pv;
615 	struct promise_raid_conf *meta;
616 	off_t size;
617 	int disk_pos, md_disk_pos, i, resurrection = 0;
618 	uint32_t eoff, esize;
619 
620 	sc = disk->d_softc;
621 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
622 
623 	pv = vol->v_md_data;
624 	meta = pv->pv_meta;
625 
626 	if (sdn >= 0) {
627 		/* Find disk position in metadata by it's serial. */
628 		md_disk_pos = promise_meta_find_disk(meta, pd->pd_meta[sdn]->disk.id);
629 		/* For RAID0+1 we need to translate order. */
630 		disk_pos = promise_meta_translate_disk(vol, md_disk_pos);
631 	} else {
632 		md_disk_pos = -1;
633 		disk_pos = -1;
634 	}
635 	if (disk_pos < 0) {
636 		G_RAID_DEBUG1(1, sc, "Disk %s is not part of the volume %s",
637 		    g_raid_get_diskname(disk), vol->v_name);
638 		/* Failed stale disk is useless for us. */
639 		if (sdn >= 0 &&
640 		    pd->pd_meta[sdn]->disk.flags & PROMISE_F_DOWN) {
641 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
642 			return (0);
643 		}
644 		/* If we were given specific metadata subdisk - erase it. */
645 		if (sdn >= 0) {
646 			free(pd->pd_meta[sdn], M_MD_PROMISE);
647 			for (i = sdn; i < pd->pd_subdisks - 1; i++)
648 				pd->pd_meta[i] = pd->pd_meta[i + 1];
649 			pd->pd_meta[PROMISE_MAX_SUBDISKS - 1] = NULL;
650 			pd->pd_subdisks--;
651 		}
652 		/* If we are in the start process, that's all for now. */
653 		if (!pv->pv_started)
654 			goto nofit;
655 		/*
656 		 * If we have already started - try to get use of the disk.
657 		 * Try to replace OFFLINE disks first, then FAILED.
658 		 */
659 		promise_meta_unused_range(pd->pd_meta, pd->pd_subdisks,
660 		    disk->d_consumer->provider->mediasize /
661 		    disk->d_consumer->provider->sectorsize,
662 		    &eoff, &esize);
663 		if (esize == 0) {
664 			G_RAID_DEBUG1(1, sc, "No free space on disk %s",
665 			    g_raid_get_diskname(disk));
666 			goto nofit;
667 		}
668 		size = INT64_MAX;
669 		for (i = 0; i < vol->v_disks_count; i++) {
670 			sd = &vol->v_subdisks[i];
671 			if (sd->sd_state != G_RAID_SUBDISK_S_NONE)
672 				size = sd->sd_size;
673 			if (sd->sd_state <= G_RAID_SUBDISK_S_FAILED &&
674 			    (disk_pos < 0 ||
675 			     vol->v_subdisks[i].sd_state < sd->sd_state))
676 				disk_pos = i;
677 		}
678 		if (disk_pos >= 0 &&
679 		    vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT &&
680 		    (off_t)esize * 512 < size) {
681 			G_RAID_DEBUG1(1, sc, "Disk %s free space "
682 			    "is too small (%ju < %ju)",
683 			    g_raid_get_diskname(disk),
684 			    (off_t)esize * 512, size);
685 			disk_pos = -1;
686 		}
687 		if (disk_pos >= 0) {
688 			if (vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT)
689 				esize = size / 512;
690 			/* For RAID0+1 we need to translate order. */
691 			md_disk_pos = promise_meta_translate_disk(vol, disk_pos);
692 		} else {
693 nofit:
694 			if (pd->pd_subdisks == 0) {
695 				g_raid_change_disk_state(disk,
696 				    G_RAID_DISK_S_SPARE);
697 			}
698 			return (0);
699 		}
700 		G_RAID_DEBUG1(1, sc, "Disk %s takes pos %d in the volume %s",
701 		    g_raid_get_diskname(disk), disk_pos, vol->v_name);
702 		resurrection = 1;
703 	}
704 
705 	sd = &vol->v_subdisks[disk_pos];
706 
707 	if (resurrection && sd->sd_disk != NULL) {
708 		g_raid_change_disk_state(sd->sd_disk,
709 		    G_RAID_DISK_S_STALE_FAILED);
710 		TAILQ_REMOVE(&sd->sd_disk->d_subdisks,
711 		    sd, sd_next);
712 	}
713 	vol->v_subdisks[disk_pos].sd_disk = disk;
714 	TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
715 
716 	/* Welcome the new disk. */
717 	if (resurrection)
718 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
719 	else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN)
720 		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
721 	else
722 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
723 
724 	if (resurrection) {
725 		sd->sd_offset = (off_t)eoff * 512;
726 		sd->sd_size = (off_t)esize * 512;
727 	} else {
728 		sd->sd_offset = (off_t)pd->pd_meta[sdn]->disk_offset * 512;
729 		sd->sd_size = (off_t)pd->pd_meta[sdn]->disk_sectors * 512;
730 	}
731 
732 	if (resurrection) {
733 		/* Stale disk, almost same as new. */
734 		g_raid_change_subdisk_state(sd,
735 		    G_RAID_SUBDISK_S_NEW);
736 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) {
737 		/* Failed disk. */
738 		g_raid_change_subdisk_state(sd,
739 		    G_RAID_SUBDISK_S_FAILED);
740 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_REDIR) {
741 		/* Rebuilding disk. */
742 		g_raid_change_subdisk_state(sd,
743 		    G_RAID_SUBDISK_S_REBUILD);
744 		if (pd->pd_meta[sdn]->generation != meta->generation)
745 			sd->sd_rebuild_pos = 0;
746 		else {
747 			sd->sd_rebuild_pos =
748 			    (off_t)pd->pd_meta[sdn]->rebuild_lba * 512;
749 		}
750 	} else if (!(meta->disks[md_disk_pos].flags & PROMISE_F_ONLINE)) {
751 		/* Rebuilding disk. */
752 		g_raid_change_subdisk_state(sd,
753 		    G_RAID_SUBDISK_S_NEW);
754 	} else if (pd->pd_meta[sdn]->generation != meta->generation ||
755 	    (meta->status & PROMISE_S_MARKED)) {
756 		/* Stale disk or dirty volume (unclean shutdown). */
757 		g_raid_change_subdisk_state(sd,
758 		    G_RAID_SUBDISK_S_STALE);
759 	} else {
760 		/* Up to date disk. */
761 		g_raid_change_subdisk_state(sd,
762 		    G_RAID_SUBDISK_S_ACTIVE);
763 	}
764 	g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
765 	    G_RAID_EVENT_SUBDISK);
766 
767 	return (resurrection);
768 }
769 
770 static void
771 g_raid_md_promise_refill(struct g_raid_softc *sc)
772 {
773 	struct g_raid_volume *vol;
774 	struct g_raid_subdisk *sd;
775 	struct g_raid_disk *disk;
776 	struct g_raid_md_object *md;
777 	struct g_raid_md_promise_perdisk *pd;
778 	struct g_raid_md_promise_pervolume *pv;
779 	int update, updated, i, bad;
780 
781 	md = sc->sc_md;
782 restart:
783 	updated = 0;
784 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
785 		pv = vol->v_md_data;
786 		if (!pv->pv_started || vol->v_stopping)
787 			continue;
788 
789 		/* Search for subdisk that needs replacement. */
790 		bad = 0;
791 		for (i = 0; i < vol->v_disks_count; i++) {
792 			sd = &vol->v_subdisks[i];
793 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
794 			    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
795 			        bad = 1;
796 		}
797 		if (!bad)
798 			continue;
799 
800 		G_RAID_DEBUG1(1, sc, "Volume %s is not complete, "
801 		    "trying to refill.", vol->v_name);
802 
803 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
804 			/* Skip failed. */
805 			if (disk->d_state < G_RAID_DISK_S_SPARE)
806 				continue;
807 			/* Skip already used by this volume. */
808 			for (i = 0; i < vol->v_disks_count; i++) {
809 				sd = &vol->v_subdisks[i];
810 				if (sd->sd_disk == disk)
811 					break;
812 			}
813 			if (i < vol->v_disks_count)
814 				continue;
815 
816 			/* Try to use disk if it has empty extents. */
817 			pd = disk->d_md_data;
818 			if (pd->pd_subdisks < PROMISE_MAX_SUBDISKS) {
819 				update =
820 				    g_raid_md_promise_start_disk(disk, -1, vol);
821 			} else
822 				update = 0;
823 			if (update) {
824 				updated = 1;
825 				g_raid_md_write_promise(md, vol, NULL, disk);
826 				break;
827 			}
828 		}
829 	}
830 	if (updated)
831 		goto restart;
832 }
833 
834 static void
835 g_raid_md_promise_start(struct g_raid_volume *vol)
836 {
837 	struct g_raid_softc *sc;
838 	struct g_raid_subdisk *sd;
839 	struct g_raid_disk *disk;
840 	struct g_raid_md_object *md;
841 	struct g_raid_md_promise_perdisk *pd;
842 	struct g_raid_md_promise_pervolume *pv;
843 	struct promise_raid_conf *meta;
844 	int i;
845 
846 	sc = vol->v_softc;
847 	md = sc->sc_md;
848 	pv = vol->v_md_data;
849 	meta = pv->pv_meta;
850 
851 	if (meta->type == PROMISE_T_RAID0)
852 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
853 	else if (meta->type == PROMISE_T_RAID1) {
854 		if (meta->array_width == 1)
855 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
856 		else
857 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
858 	} else if (meta->type == PROMISE_T_RAID3)
859 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID3;
860 	else if (meta->type == PROMISE_T_RAID5)
861 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
862 	else if (meta->type == PROMISE_T_SPAN)
863 		vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
864 	else if (meta->type == PROMISE_T_JBOD)
865 		vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
866 	else
867 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
868 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
869 	vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ
870 	vol->v_disks_count = meta->total_disks;
871 	vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ
872 	if (meta->total_sectors_high < 256) /* If value looks sane. */
873 		vol->v_mediasize |=
874 		    ((off_t)meta->total_sectors_high << 32) * 512; //ZZZ
875 	vol->v_sectorsize = 512; //ZZZ
876 	for (i = 0; i < vol->v_disks_count; i++) {
877 		sd = &vol->v_subdisks[i];
878 		sd->sd_offset = (off_t)meta->disk_offset * 512; //ZZZ
879 		sd->sd_size = (off_t)meta->disk_sectors * 512; //ZZZ
880 	}
881 	g_raid_start_volume(vol);
882 
883 	/* Make all disks found till the moment take their places. */
884 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
885 		pd = disk->d_md_data;
886 		for (i = 0; i < pd->pd_subdisks; i++) {
887 			if (pd->pd_meta[i]->volume_id == meta->volume_id)
888 				g_raid_md_promise_start_disk(disk, i, vol);
889 		}
890 	}
891 
892 	pv->pv_started = 1;
893 	callout_stop(&pv->pv_start_co);
894 	G_RAID_DEBUG1(0, sc, "Volume started.");
895 	g_raid_md_write_promise(md, vol, NULL, NULL);
896 
897 	/* Pickup any STALE/SPARE disks to refill array if needed. */
898 	g_raid_md_promise_refill(sc);
899 
900 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
901 }
902 
903 static void
904 g_raid_promise_go(void *arg)
905 {
906 	struct g_raid_volume *vol;
907 	struct g_raid_softc *sc;
908 	struct g_raid_md_promise_pervolume *pv;
909 
910 	vol = arg;
911 	pv = vol->v_md_data;
912 	sc = vol->v_softc;
913 	if (!pv->pv_started) {
914 		G_RAID_DEBUG1(0, sc, "Force volume start due to timeout.");
915 		g_raid_event_send(vol, G_RAID_VOLUME_E_STARTMD,
916 		    G_RAID_EVENT_VOLUME);
917 	}
918 }
919 
920 static void
921 g_raid_md_promise_new_disk(struct g_raid_disk *disk)
922 {
923 	struct g_raid_softc *sc;
924 	struct g_raid_md_object *md;
925 	struct promise_raid_conf *pdmeta;
926 	struct g_raid_md_promise_perdisk *pd;
927 	struct g_raid_md_promise_pervolume *pv;
928 	struct g_raid_volume *vol;
929 	int i;
930 	char buf[33];
931 
932 	sc = disk->d_softc;
933 	md = sc->sc_md;
934 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
935 
936 	if (pd->pd_subdisks == 0) {
937 		g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
938 		g_raid_md_promise_refill(sc);
939 		return;
940 	}
941 
942 	for (i = 0; i < pd->pd_subdisks; i++) {
943 		pdmeta = pd->pd_meta[i];
944 
945 		/* Look for volume with matching ID. */
946 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
947 		if (vol == NULL) {
948 			promise_meta_get_name(pdmeta, buf);
949 			vol = g_raid_create_volume(sc, buf, pdmeta->array_number);
950 			pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
951 			pv->pv_id = pdmeta->volume_id;
952 			vol->v_md_data = pv;
953 			callout_init(&pv->pv_start_co, 1);
954 			callout_reset(&pv->pv_start_co,
955 			    g_raid_start_timeout * hz,
956 			    g_raid_promise_go, vol);
957 		} else
958 			pv = vol->v_md_data;
959 
960 		/* If we haven't started yet - check metadata freshness. */
961 		if (pv->pv_meta == NULL || !pv->pv_started) {
962 			if (pv->pv_meta == NULL ||
963 			    ((int16_t)(pdmeta->generation - pv->pv_generation)) > 0) {
964 				G_RAID_DEBUG1(1, sc, "Newer disk");
965 				if (pv->pv_meta != NULL)
966 					free(pv->pv_meta, M_MD_PROMISE);
967 				pv->pv_meta = promise_meta_copy(pdmeta);
968 				pv->pv_generation = pv->pv_meta->generation;
969 				pv->pv_disks_present = 1;
970 			} else if (pdmeta->generation == pv->pv_generation) {
971 				pv->pv_disks_present++;
972 				G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
973 				    pv->pv_disks_present,
974 				    pv->pv_meta->total_disks);
975 			} else {
976 				G_RAID_DEBUG1(1, sc, "Older disk");
977 			}
978 		}
979 	}
980 
981 	for (i = 0; i < pd->pd_subdisks; i++) {
982 		pdmeta = pd->pd_meta[i];
983 
984 		/* Look for volume with matching ID. */
985 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
986 		if (vol == NULL)
987 			continue;
988 		pv = vol->v_md_data;
989 
990 		if (pv->pv_started) {
991 			if (g_raid_md_promise_start_disk(disk, i, vol))
992 				g_raid_md_write_promise(md, vol, NULL, NULL);
993 		} else {
994 			/* If we collected all needed disks - start array. */
995 			if (pv->pv_disks_present == pv->pv_meta->total_disks)
996 				g_raid_md_promise_start(vol);
997 		}
998 	}
999 }
1000 
1001 static int
1002 g_raid_md_create_promise(struct g_raid_md_object *md, struct g_class *mp,
1003     struct g_geom **gp)
1004 {
1005 	struct g_geom *geom;
1006 	struct g_raid_softc *sc;
1007 
1008 	/* Search for existing node. */
1009 	LIST_FOREACH(geom, &mp->geom, geom) {
1010 		sc = geom->softc;
1011 		if (sc == NULL)
1012 			continue;
1013 		if (sc->sc_stopping != 0)
1014 			continue;
1015 		if (sc->sc_md->mdo_class != md->mdo_class)
1016 			continue;
1017 		break;
1018 	}
1019 	if (geom != NULL) {
1020 		*gp = geom;
1021 		return (G_RAID_MD_TASTE_EXISTING);
1022 	}
1023 
1024 	/* Create new one if not found. */
1025 	sc = g_raid_create_node(mp, "Promise", md);
1026 	if (sc == NULL)
1027 		return (G_RAID_MD_TASTE_FAIL);
1028 	md->mdo_softc = sc;
1029 	*gp = sc->sc_geom;
1030 	return (G_RAID_MD_TASTE_NEW);
1031 }
1032 
1033 static int
1034 g_raid_md_taste_promise(struct g_raid_md_object *md, struct g_class *mp,
1035                               struct g_consumer *cp, struct g_geom **gp)
1036 {
1037 	struct g_consumer *rcp;
1038 	struct g_provider *pp;
1039 	struct g_raid_softc *sc;
1040 	struct g_raid_disk *disk;
1041 	struct promise_raid_conf *meta, *metaarr[4];
1042 	struct g_raid_md_promise_perdisk *pd;
1043 	struct g_geom *geom;
1044 	int error, i, j, result, len, subdisks;
1045 	char name[16];
1046 	uint16_t vendor;
1047 
1048 	G_RAID_DEBUG(1, "Tasting Promise on %s", cp->provider->name);
1049 	pp = cp->provider;
1050 
1051 	/* Read metadata from device. */
1052 	meta = NULL;
1053 	vendor = 0xffff;
1054 	if (g_access(cp, 1, 0, 0) != 0)
1055 		return (G_RAID_MD_TASTE_FAIL);
1056 	g_topology_unlock();
1057 	len = 2;
1058 	if (pp->geom->rank == 1)
1059 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1060 	subdisks = promise_meta_read(cp, metaarr);
1061 	g_topology_lock();
1062 	g_access(cp, -1, 0, 0);
1063 	if (subdisks == 0) {
1064 		if (g_raid_aggressive_spare) {
1065 			if (vendor == 0x105a || vendor == 0x1002) {
1066 				G_RAID_DEBUG(1,
1067 				    "No Promise metadata, forcing spare.");
1068 				goto search;
1069 			} else {
1070 				G_RAID_DEBUG(1,
1071 				    "Promise/ATI vendor mismatch "
1072 				    "0x%04x != 0x105a/0x1002",
1073 				    vendor);
1074 			}
1075 		}
1076 		return (G_RAID_MD_TASTE_FAIL);
1077 	}
1078 
1079 	/* Metadata valid. Print it. */
1080 	for (i = 0; i < subdisks; i++)
1081 		g_raid_md_promise_print(metaarr[i]);
1082 
1083 	/* Purge meaningless (empty/spare) records. */
1084 	for (i = 0; i < subdisks; ) {
1085 		if (metaarr[i]->disk.flags & PROMISE_F_ASSIGNED) {
1086 			i++;
1087 			continue;
1088 		}
1089 		free(metaarr[i], M_MD_PROMISE);
1090 		for (j = i; j < subdisks - 1; j++)
1091 			metaarr[i] = metaarr[j + 1];
1092 		metaarr[PROMISE_MAX_SUBDISKS - 1] = NULL;
1093 		subdisks--;
1094 	}
1095 
1096 search:
1097 	/* Search for matching node. */
1098 	sc = NULL;
1099 	LIST_FOREACH(geom, &mp->geom, geom) {
1100 		sc = geom->softc;
1101 		if (sc == NULL)
1102 			continue;
1103 		if (sc->sc_stopping != 0)
1104 			continue;
1105 		if (sc->sc_md->mdo_class != md->mdo_class)
1106 			continue;
1107 		break;
1108 	}
1109 
1110 	/* Found matching node. */
1111 	if (geom != NULL) {
1112 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1113 		result = G_RAID_MD_TASTE_EXISTING;
1114 
1115 	} else { /* Not found matching node -- create one. */
1116 		result = G_RAID_MD_TASTE_NEW;
1117 		snprintf(name, sizeof(name), "Promise");
1118 		sc = g_raid_create_node(mp, name, md);
1119 		md->mdo_softc = sc;
1120 		geom = sc->sc_geom;
1121 	}
1122 
1123 	rcp = g_new_consumer(geom);
1124 	g_attach(rcp, pp);
1125 	if (g_access(rcp, 1, 1, 1) != 0)
1126 		; //goto fail1;
1127 
1128 	g_topology_unlock();
1129 	sx_xlock(&sc->sc_lock);
1130 
1131 	pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1132 	pd->pd_subdisks = subdisks;
1133 	for (i = 0; i < subdisks; i++)
1134 		pd->pd_meta[i] = metaarr[i];
1135 	disk = g_raid_create_disk(sc);
1136 	disk->d_md_data = (void *)pd;
1137 	disk->d_consumer = rcp;
1138 	rcp->private = disk;
1139 
1140 	/* Read kernel dumping information. */
1141 	disk->d_kd.offset = 0;
1142 	disk->d_kd.length = OFF_MAX;
1143 	len = sizeof(disk->d_kd);
1144 	error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd);
1145 	if (disk->d_kd.di.dumper == NULL)
1146 		G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.",
1147 		    rcp->provider->name, error);
1148 
1149 	g_raid_md_promise_new_disk(disk);
1150 
1151 	sx_xunlock(&sc->sc_lock);
1152 	g_topology_lock();
1153 	*gp = geom;
1154 	return (result);
1155 }
1156 
1157 static int
1158 g_raid_md_event_promise(struct g_raid_md_object *md,
1159     struct g_raid_disk *disk, u_int event)
1160 {
1161 	struct g_raid_softc *sc;
1162 
1163 	sc = md->mdo_softc;
1164 	if (disk == NULL)
1165 		return (-1);
1166 	switch (event) {
1167 	case G_RAID_DISK_E_DISCONNECTED:
1168 		/* Delete disk. */
1169 		g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1170 		g_raid_destroy_disk(disk);
1171 		g_raid_md_promise_purge_volumes(sc);
1172 
1173 		/* Write updated metadata to all disks. */
1174 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1175 
1176 		/* Check if anything left. */
1177 		if (g_raid_ndisks(sc, -1) == 0)
1178 			g_raid_destroy_node(sc, 0);
1179 		else
1180 			g_raid_md_promise_refill(sc);
1181 		return (0);
1182 	}
1183 	return (-2);
1184 }
1185 
1186 static int
1187 g_raid_md_volume_event_promise(struct g_raid_md_object *md,
1188     struct g_raid_volume *vol, u_int event)
1189 {
1190 	struct g_raid_md_promise_pervolume *pv;
1191 
1192 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1193 	switch (event) {
1194 	case G_RAID_VOLUME_E_STARTMD:
1195 		if (!pv->pv_started)
1196 			g_raid_md_promise_start(vol);
1197 		return (0);
1198 	}
1199 	return (-2);
1200 }
1201 
1202 static int
1203 g_raid_md_ctl_promise(struct g_raid_md_object *md,
1204     struct gctl_req *req)
1205 {
1206 	struct g_raid_softc *sc;
1207 	struct g_raid_volume *vol, *vol1;
1208 	struct g_raid_subdisk *sd;
1209 	struct g_raid_disk *disk, *disks[PROMISE_MAX_DISKS];
1210 	struct g_raid_md_promise_perdisk *pd;
1211 	struct g_raid_md_promise_pervolume *pv;
1212 	struct g_consumer *cp;
1213 	struct g_provider *pp;
1214 	char arg[16];
1215 	const char *verb, *volname, *levelname, *diskname;
1216 	char *tmp;
1217 	int *nargs, *force;
1218 	off_t size, sectorsize, strip;
1219 	intmax_t *sizearg, *striparg;
1220 	uint32_t offs[PROMISE_MAX_DISKS], esize;
1221 	int numdisks, i, len, level, qual;
1222 	int error;
1223 
1224 	sc = md->mdo_softc;
1225 	verb = gctl_get_param(req, "verb", NULL);
1226 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1227 	error = 0;
1228 	if (strcmp(verb, "label") == 0) {
1229 
1230 		if (*nargs < 4) {
1231 			gctl_error(req, "Invalid number of arguments.");
1232 			return (-1);
1233 		}
1234 		volname = gctl_get_asciiparam(req, "arg1");
1235 		if (volname == NULL) {
1236 			gctl_error(req, "No volume name.");
1237 			return (-2);
1238 		}
1239 		levelname = gctl_get_asciiparam(req, "arg2");
1240 		if (levelname == NULL) {
1241 			gctl_error(req, "No RAID level.");
1242 			return (-3);
1243 		}
1244 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1245 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1246 			return (-4);
1247 		}
1248 		numdisks = *nargs - 3;
1249 		force = gctl_get_paraml(req, "force", sizeof(*force));
1250 		if (!g_raid_md_promise_supported(level, qual, numdisks,
1251 		    force ? *force : 0)) {
1252 			gctl_error(req, "Unsupported RAID level "
1253 			    "(0x%02x/0x%02x), or number of disks (%d).",
1254 			    level, qual, numdisks);
1255 			return (-5);
1256 		}
1257 
1258 		/* Search for disks, connect them and probe. */
1259 		size = INT64_MAX;
1260 		sectorsize = 0;
1261 		bzero(disks, sizeof(disks));
1262 		bzero(offs, sizeof(offs));
1263 		for (i = 0; i < numdisks; i++) {
1264 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1265 			diskname = gctl_get_asciiparam(req, arg);
1266 			if (diskname == NULL) {
1267 				gctl_error(req, "No disk name (%s).", arg);
1268 				error = -6;
1269 				break;
1270 			}
1271 			if (strcmp(diskname, "NONE") == 0)
1272 				continue;
1273 
1274 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1275 				if (disk->d_consumer != NULL &&
1276 				    disk->d_consumer->provider != NULL &&
1277 				    strcmp(disk->d_consumer->provider->name,
1278 				     diskname) == 0)
1279 					break;
1280 			}
1281 			if (disk != NULL) {
1282 				if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
1283 					gctl_error(req, "Disk '%s' is in a "
1284 					    "wrong state (%s).", diskname,
1285 					    g_raid_disk_state2str(disk->d_state));
1286 					error = -7;
1287 					break;
1288 				}
1289 				pd = disk->d_md_data;
1290 				if (pd->pd_subdisks >= PROMISE_MAX_SUBDISKS) {
1291 					gctl_error(req, "Disk '%s' already "
1292 					    "used by %d volumes.",
1293 					    diskname, pd->pd_subdisks);
1294 					error = -7;
1295 					break;
1296 				}
1297 				pp = disk->d_consumer->provider;
1298 				disks[i] = disk;
1299 				promise_meta_unused_range(pd->pd_meta,
1300 				    pd->pd_subdisks,
1301 				    pp->mediasize / pp->sectorsize,
1302 				    &offs[i], &esize);
1303 				size = MIN(size, (off_t)esize * pp->sectorsize);
1304 				sectorsize = MAX(sectorsize, pp->sectorsize);
1305 				continue;
1306 			}
1307 
1308 			g_topology_lock();
1309 			cp = g_raid_open_consumer(sc, diskname);
1310 			if (cp == NULL) {
1311 				gctl_error(req, "Can't open disk '%s'.",
1312 				    diskname);
1313 				g_topology_unlock();
1314 				error = -8;
1315 				break;
1316 			}
1317 			pp = cp->provider;
1318 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1319 			disk = g_raid_create_disk(sc);
1320 			disk->d_md_data = (void *)pd;
1321 			disk->d_consumer = cp;
1322 			disks[i] = disk;
1323 			cp->private = disk;
1324 			g_topology_unlock();
1325 
1326 			if (pp->mediasize / pp->sectorsize > UINT32_MAX) {
1327 				gctl_error(req,
1328 				    "Disk '%s' is too big.", diskname);
1329 				error = -8;
1330 				break;
1331 			}
1332 
1333 			/* Read kernel dumping information. */
1334 			disk->d_kd.offset = 0;
1335 			disk->d_kd.length = OFF_MAX;
1336 			len = sizeof(disk->d_kd);
1337 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1338 			if (disk->d_kd.di.dumper == NULL)
1339 				G_RAID_DEBUG1(2, sc,
1340 				    "Dumping not supported by %s.",
1341 				    cp->provider->name);
1342 
1343 			/* Reserve some space for metadata. */
1344 			size = MIN(size, pp->mediasize - 131072llu * pp->sectorsize);
1345 			sectorsize = MAX(sectorsize, pp->sectorsize);
1346 		}
1347 		if (error != 0) {
1348 			for (i = 0; i < numdisks; i++) {
1349 				if (disks[i] != NULL &&
1350 				    disks[i]->d_state == G_RAID_DISK_S_NONE)
1351 					g_raid_destroy_disk(disks[i]);
1352 			}
1353 			return (error);
1354 		}
1355 
1356 		if (sectorsize <= 0) {
1357 			gctl_error(req, "Can't get sector size.");
1358 			return (-8);
1359 		}
1360 
1361 		/* Handle size argument. */
1362 		len = sizeof(*sizearg);
1363 		sizearg = gctl_get_param(req, "size", &len);
1364 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1365 		    *sizearg > 0) {
1366 			if (*sizearg > size) {
1367 				gctl_error(req, "Size too big %lld > %lld.",
1368 				    (long long)*sizearg, (long long)size);
1369 				return (-9);
1370 			}
1371 			size = *sizearg;
1372 		}
1373 
1374 		/* Handle strip argument. */
1375 		strip = 131072;
1376 		len = sizeof(*striparg);
1377 		striparg = gctl_get_param(req, "strip", &len);
1378 		if (striparg != NULL && len == sizeof(*striparg) &&
1379 		    *striparg > 0) {
1380 			if (*striparg < sectorsize) {
1381 				gctl_error(req, "Strip size too small.");
1382 				return (-10);
1383 			}
1384 			if (*striparg % sectorsize != 0) {
1385 				gctl_error(req, "Incorrect strip size.");
1386 				return (-11);
1387 			}
1388 			strip = *striparg;
1389 		}
1390 
1391 		/* Round size down to strip or sector. */
1392 		if (level == G_RAID_VOLUME_RL_RAID1 ||
1393 		    level == G_RAID_VOLUME_RL_SINGLE ||
1394 		    level == G_RAID_VOLUME_RL_CONCAT)
1395 			size -= (size % sectorsize);
1396 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1397 		    (numdisks & 1) != 0)
1398 			size -= (size % (2 * strip));
1399 		else
1400 			size -= (size % strip);
1401 		if (size <= 0) {
1402 			gctl_error(req, "Size too small.");
1403 			return (-13);
1404 		}
1405 		if (size > 0xffffffffllu * sectorsize) {
1406 			gctl_error(req, "Size too big.");
1407 			return (-14);
1408 		}
1409 
1410 		/* We have all we need, create things: volume, ... */
1411 		pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1412 		arc4rand(&pv->pv_id, sizeof(pv->pv_id), 0);
1413 		pv->pv_generation = 0;
1414 		pv->pv_started = 1;
1415 		vol = g_raid_create_volume(sc, volname, -1);
1416 		vol->v_md_data = pv;
1417 		vol->v_raid_level = level;
1418 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
1419 		vol->v_strip_size = strip;
1420 		vol->v_disks_count = numdisks;
1421 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1422 		    level == G_RAID_VOLUME_RL_CONCAT ||
1423 		    level == G_RAID_VOLUME_RL_SINGLE)
1424 			vol->v_mediasize = size * numdisks;
1425 		else if (level == G_RAID_VOLUME_RL_RAID1)
1426 			vol->v_mediasize = size;
1427 		else if (level == G_RAID_VOLUME_RL_RAID3 ||
1428 		    level == G_RAID_VOLUME_RL_RAID5)
1429 			vol->v_mediasize = size * (numdisks - 1);
1430 		else { /* RAID1E */
1431 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1432 			    strip;
1433 		}
1434 		vol->v_sectorsize = sectorsize;
1435 		g_raid_start_volume(vol);
1436 
1437 		/* , and subdisks. */
1438 		for (i = 0; i < numdisks; i++) {
1439 			disk = disks[i];
1440 			sd = &vol->v_subdisks[i];
1441 			sd->sd_disk = disk;
1442 			sd->sd_offset = (off_t)offs[i] * 512;
1443 			sd->sd_size = size;
1444 			if (disk == NULL)
1445 				continue;
1446 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1447 			g_raid_change_disk_state(disk,
1448 			    G_RAID_DISK_S_ACTIVE);
1449 			g_raid_change_subdisk_state(sd,
1450 			    G_RAID_SUBDISK_S_ACTIVE);
1451 			g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1452 			    G_RAID_EVENT_SUBDISK);
1453 		}
1454 
1455 		/* Write metadata based on created entities. */
1456 		G_RAID_DEBUG1(0, sc, "Array started.");
1457 		g_raid_md_write_promise(md, vol, NULL, NULL);
1458 
1459 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1460 		g_raid_md_promise_refill(sc);
1461 
1462 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1463 		    G_RAID_EVENT_VOLUME);
1464 		return (0);
1465 	}
1466 	if (strcmp(verb, "add") == 0) {
1467 
1468 		gctl_error(req, "`add` command is not applicable, "
1469 		    "use `label` instead.");
1470 		return (-99);
1471 	}
1472 	if (strcmp(verb, "delete") == 0) {
1473 
1474 		/* Full node destruction. */
1475 		if (*nargs == 1) {
1476 			/* Check if some volume is still open. */
1477 			force = gctl_get_paraml(req, "force", sizeof(*force));
1478 			if (force != NULL && *force == 0 &&
1479 			    g_raid_nopens(sc) != 0) {
1480 				gctl_error(req, "Some volume is still open.");
1481 				return (-4);
1482 			}
1483 
1484 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1485 				if (disk->d_consumer)
1486 					promise_meta_erase(disk->d_consumer);
1487 			}
1488 			g_raid_destroy_node(sc, 0);
1489 			return (0);
1490 		}
1491 
1492 		/* Destroy specified volume. If it was last - all node. */
1493 		if (*nargs != 2) {
1494 			gctl_error(req, "Invalid number of arguments.");
1495 			return (-1);
1496 		}
1497 		volname = gctl_get_asciiparam(req, "arg1");
1498 		if (volname == NULL) {
1499 			gctl_error(req, "No volume name.");
1500 			return (-2);
1501 		}
1502 
1503 		/* Search for volume. */
1504 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1505 			if (strcmp(vol->v_name, volname) == 0)
1506 				break;
1507 		}
1508 		if (vol == NULL) {
1509 			i = strtol(volname, &tmp, 10);
1510 			if (verb != volname && tmp[0] == 0) {
1511 				TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1512 					if (vol->v_global_id == i)
1513 						break;
1514 				}
1515 			}
1516 		}
1517 		if (vol == NULL) {
1518 			gctl_error(req, "Volume '%s' not found.", volname);
1519 			return (-3);
1520 		}
1521 
1522 		/* Check if volume is still open. */
1523 		force = gctl_get_paraml(req, "force", sizeof(*force));
1524 		if (force != NULL && *force == 0 &&
1525 		    vol->v_provider_open != 0) {
1526 			gctl_error(req, "Volume is still open.");
1527 			return (-4);
1528 		}
1529 
1530 		/* Destroy volume and potentially node. */
1531 		i = 0;
1532 		TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
1533 			i++;
1534 		if (i >= 2) {
1535 			g_raid_destroy_volume(vol);
1536 			g_raid_md_promise_purge_disks(sc);
1537 			g_raid_md_write_promise(md, NULL, NULL, NULL);
1538 		} else {
1539 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1540 				if (disk->d_consumer)
1541 					promise_meta_erase(disk->d_consumer);
1542 			}
1543 			g_raid_destroy_node(sc, 0);
1544 		}
1545 		return (0);
1546 	}
1547 	if (strcmp(verb, "remove") == 0 ||
1548 	    strcmp(verb, "fail") == 0) {
1549 		if (*nargs < 2) {
1550 			gctl_error(req, "Invalid number of arguments.");
1551 			return (-1);
1552 		}
1553 		for (i = 1; i < *nargs; i++) {
1554 			snprintf(arg, sizeof(arg), "arg%d", i);
1555 			diskname = gctl_get_asciiparam(req, arg);
1556 			if (diskname == NULL) {
1557 				gctl_error(req, "No disk name (%s).", arg);
1558 				error = -2;
1559 				break;
1560 			}
1561 			if (strncmp(diskname, "/dev/", 5) == 0)
1562 				diskname += 5;
1563 
1564 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1565 				if (disk->d_consumer != NULL &&
1566 				    disk->d_consumer->provider != NULL &&
1567 				    strcmp(disk->d_consumer->provider->name,
1568 				     diskname) == 0)
1569 					break;
1570 			}
1571 			if (disk == NULL) {
1572 				gctl_error(req, "Disk '%s' not found.",
1573 				    diskname);
1574 				error = -3;
1575 				break;
1576 			}
1577 
1578 			if (strcmp(verb, "fail") == 0) {
1579 				g_raid_md_fail_disk_promise(md, NULL, disk);
1580 				continue;
1581 			}
1582 
1583 			/* Erase metadata on deleting disk and destroy it. */
1584 			promise_meta_erase(disk->d_consumer);
1585 			g_raid_destroy_disk(disk);
1586 		}
1587 		g_raid_md_promise_purge_volumes(sc);
1588 
1589 		/* Write updated metadata to remaining disks. */
1590 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1591 
1592 		/* Check if anything left. */
1593 		if (g_raid_ndisks(sc, -1) == 0)
1594 			g_raid_destroy_node(sc, 0);
1595 		else
1596 			g_raid_md_promise_refill(sc);
1597 		return (error);
1598 	}
1599 	if (strcmp(verb, "insert") == 0) {
1600 		if (*nargs < 2) {
1601 			gctl_error(req, "Invalid number of arguments.");
1602 			return (-1);
1603 		}
1604 		for (i = 1; i < *nargs; i++) {
1605 			/* Get disk name. */
1606 			snprintf(arg, sizeof(arg), "arg%d", i);
1607 			diskname = gctl_get_asciiparam(req, arg);
1608 			if (diskname == NULL) {
1609 				gctl_error(req, "No disk name (%s).", arg);
1610 				error = -3;
1611 				break;
1612 			}
1613 
1614 			/* Try to find provider with specified name. */
1615 			g_topology_lock();
1616 			cp = g_raid_open_consumer(sc, diskname);
1617 			if (cp == NULL) {
1618 				gctl_error(req, "Can't open disk '%s'.",
1619 				    diskname);
1620 				g_topology_unlock();
1621 				error = -4;
1622 				break;
1623 			}
1624 			pp = cp->provider;
1625 			g_topology_unlock();
1626 
1627 			if (pp->mediasize / pp->sectorsize > UINT32_MAX) {
1628 				gctl_error(req,
1629 				    "Disk '%s' is too big.", diskname);
1630 				g_raid_kill_consumer(sc, cp);
1631 				error = -8;
1632 				break;
1633 			}
1634 
1635 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1636 
1637 			disk = g_raid_create_disk(sc);
1638 			disk->d_consumer = cp;
1639 			disk->d_md_data = (void *)pd;
1640 			cp->private = disk;
1641 
1642 			/* Read kernel dumping information. */
1643 			disk->d_kd.offset = 0;
1644 			disk->d_kd.length = OFF_MAX;
1645 			len = sizeof(disk->d_kd);
1646 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1647 			if (disk->d_kd.di.dumper == NULL)
1648 				G_RAID_DEBUG1(2, sc,
1649 				    "Dumping not supported by %s.",
1650 				    cp->provider->name);
1651 
1652 			/* Welcome the "new" disk. */
1653 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1654 			promise_meta_write_spare(cp);
1655 			g_raid_md_promise_refill(sc);
1656 		}
1657 		return (error);
1658 	}
1659 	return (-100);
1660 }
1661 
1662 static int
1663 g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1664     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1665 {
1666 	struct g_raid_softc *sc;
1667 	struct g_raid_volume *vol;
1668 	struct g_raid_subdisk *sd;
1669 	struct g_raid_disk *disk;
1670 	struct g_raid_md_promise_perdisk *pd;
1671 	struct g_raid_md_promise_pervolume *pv;
1672 	struct promise_raid_conf *meta;
1673 	off_t rebuild_lba64;
1674 	int i, j, pos, rebuild;
1675 
1676 	sc = md->mdo_softc;
1677 
1678 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1679 		return (0);
1680 
1681 	/* Generate new per-volume metadata for affected volumes. */
1682 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1683 		if (vol->v_stopping)
1684 			continue;
1685 
1686 		/* Skip volumes not related to specified targets. */
1687 		if (tvol != NULL && vol != tvol)
1688 			continue;
1689 		if (tsd != NULL && vol != tsd->sd_volume)
1690 			continue;
1691 		if (tdisk != NULL) {
1692 			for (i = 0; i < vol->v_disks_count; i++) {
1693 				if (vol->v_subdisks[i].sd_disk == tdisk)
1694 					break;
1695 			}
1696 			if (i >= vol->v_disks_count)
1697 				continue;
1698 		}
1699 
1700 		pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1701 		pv->pv_generation++;
1702 
1703 		meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
1704 		if (pv->pv_meta != NULL)
1705 			memcpy(meta, pv->pv_meta, sizeof(*meta));
1706 		memcpy(meta->promise_id, PROMISE_MAGIC,
1707 		    sizeof(PROMISE_MAGIC) - 1);
1708 		meta->dummy_0 = 0x00020000;
1709 		meta->integrity = PROMISE_I_VALID;
1710 
1711 		meta->generation = pv->pv_generation;
1712 		meta->status = PROMISE_S_VALID | PROMISE_S_ONLINE |
1713 		    PROMISE_S_INITED | PROMISE_S_READY;
1714 		if (vol->v_state <= G_RAID_VOLUME_S_DEGRADED)
1715 			meta->status |= PROMISE_S_DEGRADED;
1716 		if (vol->v_dirty)
1717 			meta->status |= PROMISE_S_MARKED; /* XXX: INVENTED! */
1718 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0 ||
1719 		    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1720 			meta->type = PROMISE_T_RAID0;
1721 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1722 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1723 			meta->type = PROMISE_T_RAID1;
1724 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3)
1725 			meta->type = PROMISE_T_RAID3;
1726 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1727 			meta->type = PROMISE_T_RAID5;
1728 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT)
1729 			meta->type = PROMISE_T_SPAN;
1730 		else
1731 			meta->type = PROMISE_T_JBOD;
1732 		meta->total_disks = vol->v_disks_count;
1733 		meta->stripe_shift = ffs(vol->v_strip_size / 1024);
1734 		meta->array_width = vol->v_disks_count;
1735 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1736 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1737 			meta->array_width /= 2;
1738 		meta->array_number = vol->v_global_id;
1739 		meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1740 		meta->total_sectors_high =
1741 		    (vol->v_mediasize / vol->v_sectorsize) >> 32;
1742 		meta->cylinders = meta->total_sectors / (255 * 63) - 1;
1743 		meta->heads = 254;
1744 		meta->sectors = 63;
1745 		meta->volume_id = pv->pv_id;
1746 		rebuild_lba64 = UINT64_MAX;
1747 		rebuild = 0;
1748 		for (i = 0; i < vol->v_disks_count; i++) {
1749 			sd = &vol->v_subdisks[i];
1750 			/* For RAID0+1 we need to translate order. */
1751 			pos = promise_meta_translate_disk(vol, i);
1752 			meta->disks[pos].flags = PROMISE_F_VALID |
1753 			    PROMISE_F_ASSIGNED;
1754 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE) {
1755 				meta->disks[pos].flags |= 0;
1756 			} else if (sd->sd_state == G_RAID_SUBDISK_S_FAILED) {
1757 				meta->disks[pos].flags |=
1758 				    PROMISE_F_DOWN | PROMISE_F_REDIR;
1759 			} else if (sd->sd_state <= G_RAID_SUBDISK_S_REBUILD) {
1760 				meta->disks[pos].flags |=
1761 				    PROMISE_F_ONLINE | PROMISE_F_REDIR;
1762 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1763 					rebuild_lba64 = MIN(rebuild_lba64,
1764 					    sd->sd_rebuild_pos / 512);
1765 				} else
1766 					rebuild_lba64 = 0;
1767 				rebuild = 1;
1768 			} else {
1769 				meta->disks[pos].flags |= PROMISE_F_ONLINE;
1770 				if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) {
1771 					meta->status |= PROMISE_S_MARKED;
1772 					if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
1773 						rebuild_lba64 = MIN(rebuild_lba64,
1774 						    sd->sd_rebuild_pos / 512);
1775 					} else
1776 						rebuild_lba64 = 0;
1777 				}
1778 			}
1779 			if (pv->pv_meta != NULL) {
1780 				meta->disks[pos].id = pv->pv_meta->disks[pos].id;
1781 			} else {
1782 				meta->disks[pos].number = i * 2;
1783 				arc4rand(&meta->disks[pos].id,
1784 				    sizeof(meta->disks[pos].id), 0);
1785 			}
1786 		}
1787 		promise_meta_put_name(meta, vol->v_name);
1788 
1789 		/* Try to mimic AMD BIOS rebuild/resync behavior. */
1790 		if (rebuild_lba64 != UINT64_MAX) {
1791 			if (rebuild)
1792 				meta->magic_3 = 0x03040010UL; /* Rebuild? */
1793 			else
1794 				meta->magic_3 = 0x03040008UL; /* Resync? */
1795 			/* Translate from per-disk to per-volume LBA. */
1796 			if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1797 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1798 				rebuild_lba64 *= meta->array_width;
1799 			} else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1800 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) {
1801 				rebuild_lba64 *= meta->array_width - 1;
1802 			} else
1803 				rebuild_lba64 = 0;
1804 		} else
1805 			meta->magic_3 = 0x03000000UL;
1806 		meta->rebuild_lba64 = rebuild_lba64;
1807 		meta->magic_4 = 0x04010101UL;
1808 
1809 		/* Replace per-volume metadata with new. */
1810 		if (pv->pv_meta != NULL)
1811 			free(pv->pv_meta, M_MD_PROMISE);
1812 		pv->pv_meta = meta;
1813 
1814 		/* Copy new metadata to the disks, adding or replacing old. */
1815 		for (i = 0; i < vol->v_disks_count; i++) {
1816 			sd = &vol->v_subdisks[i];
1817 			disk = sd->sd_disk;
1818 			if (disk == NULL)
1819 				continue;
1820 			/* For RAID0+1 we need to translate order. */
1821 			pos = promise_meta_translate_disk(vol, i);
1822 			pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1823 			for (j = 0; j < pd->pd_subdisks; j++) {
1824 				if (pd->pd_meta[j]->volume_id == meta->volume_id)
1825 					break;
1826 			}
1827 			if (j == pd->pd_subdisks)
1828 				pd->pd_subdisks++;
1829 			if (pd->pd_meta[j] != NULL)
1830 				free(pd->pd_meta[j], M_MD_PROMISE);
1831 			pd->pd_meta[j] = promise_meta_copy(meta);
1832 			pd->pd_meta[j]->disk = meta->disks[pos];
1833 			pd->pd_meta[j]->disk.number = pos;
1834 			pd->pd_meta[j]->disk_offset = sd->sd_offset / 512;
1835 			pd->pd_meta[j]->disk_sectors = sd->sd_size / 512;
1836 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1837 				pd->pd_meta[j]->rebuild_lba =
1838 				    sd->sd_rebuild_pos / 512;
1839 			} else if (sd->sd_state < G_RAID_SUBDISK_S_REBUILD)
1840 				pd->pd_meta[j]->rebuild_lba = 0;
1841 			else
1842 				pd->pd_meta[j]->rebuild_lba = UINT32_MAX;
1843 			pd->pd_updated = 1;
1844 		}
1845 	}
1846 
1847 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1848 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1849 		if (disk->d_state != G_RAID_DISK_S_ACTIVE)
1850 			continue;
1851 		if (!pd->pd_updated)
1852 			continue;
1853 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1854 		    g_raid_get_diskname(disk));
1855 		for (i = 0; i < pd->pd_subdisks; i++)
1856 			g_raid_md_promise_print(pd->pd_meta[i]);
1857 		promise_meta_write(disk->d_consumer,
1858 		    pd->pd_meta, pd->pd_subdisks);
1859 		pd->pd_updated = 0;
1860 	}
1861 
1862 	return (0);
1863 }
1864 
1865 static int
1866 g_raid_md_fail_disk_promise(struct g_raid_md_object *md,
1867     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1868 {
1869 	struct g_raid_softc *sc;
1870 	struct g_raid_md_promise_perdisk *pd;
1871 	struct g_raid_subdisk *sd;
1872 	int i, pos;
1873 
1874 	sc = md->mdo_softc;
1875 	pd = (struct g_raid_md_promise_perdisk *)tdisk->d_md_data;
1876 
1877 	/* We can't fail disk that is not a part of array now. */
1878 	if (tdisk->d_state != G_RAID_DISK_S_ACTIVE)
1879 		return (-1);
1880 
1881 	/*
1882 	 * Mark disk as failed in metadata and try to write that metadata
1883 	 * to the disk itself to prevent it's later resurrection as STALE.
1884 	 */
1885 	if (pd->pd_subdisks > 0 && tdisk->d_consumer != NULL)
1886 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1887 		    g_raid_get_diskname(tdisk));
1888 	for (i = 0; i < pd->pd_subdisks; i++) {
1889 		pd->pd_meta[i]->disk.flags |=
1890 		    PROMISE_F_DOWN | PROMISE_F_REDIR;
1891 		pos = pd->pd_meta[i]->disk.number;
1892 		if (pos >= 0 && pos < PROMISE_MAX_DISKS) {
1893 			pd->pd_meta[i]->disks[pos].flags |=
1894 			    PROMISE_F_DOWN | PROMISE_F_REDIR;
1895 		}
1896 		g_raid_md_promise_print(pd->pd_meta[i]);
1897 	}
1898 	if (tdisk->d_consumer != NULL)
1899 		promise_meta_write(tdisk->d_consumer,
1900 		    pd->pd_meta, pd->pd_subdisks);
1901 
1902 	/* Change states. */
1903 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1904 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1905 		g_raid_change_subdisk_state(sd,
1906 		    G_RAID_SUBDISK_S_FAILED);
1907 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1908 		    G_RAID_EVENT_SUBDISK);
1909 	}
1910 
1911 	/* Write updated metadata to remaining disks. */
1912 	g_raid_md_write_promise(md, NULL, NULL, tdisk);
1913 
1914 	g_raid_md_promise_refill(sc);
1915 	return (0);
1916 }
1917 
1918 static int
1919 g_raid_md_free_disk_promise(struct g_raid_md_object *md,
1920     struct g_raid_disk *disk)
1921 {
1922 	struct g_raid_md_promise_perdisk *pd;
1923 	int i;
1924 
1925 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1926 	for (i = 0; i < pd->pd_subdisks; i++) {
1927 		if (pd->pd_meta[i] != NULL) {
1928 			free(pd->pd_meta[i], M_MD_PROMISE);
1929 			pd->pd_meta[i] = NULL;
1930 		}
1931 	}
1932 	free(pd, M_MD_PROMISE);
1933 	disk->d_md_data = NULL;
1934 	return (0);
1935 }
1936 
1937 static int
1938 g_raid_md_free_volume_promise(struct g_raid_md_object *md,
1939     struct g_raid_volume *vol)
1940 {
1941 	struct g_raid_md_promise_pervolume *pv;
1942 
1943 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1944 	if (pv && pv->pv_meta != NULL) {
1945 		free(pv->pv_meta, M_MD_PROMISE);
1946 		pv->pv_meta = NULL;
1947 	}
1948 	if (pv && !pv->pv_started) {
1949 		pv->pv_started = 1;
1950 		callout_stop(&pv->pv_start_co);
1951 	}
1952 	return (0);
1953 }
1954 
1955 static int
1956 g_raid_md_free_promise(struct g_raid_md_object *md)
1957 {
1958 
1959 	return (0);
1960 }
1961 
1962 G_RAID_MD_DECLARE(g_raid_md_promise);
1963