xref: /freebsd/sys/geom/raid/md_promise.c (revision 2b15cb3d0922bd70ea592f0da9b4a5b167f4d53f)
1 /*-
2  * Copyright (c) 2011 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 <geom/geom.h>
42 #include "geom/raid/g_raid.h"
43 #include "g_raid_md_if.h"
44 
45 static MALLOC_DEFINE(M_MD_PROMISE, "md_promise_data", "GEOM_RAID Promise metadata");
46 
47 #define	PROMISE_MAX_DISKS	8
48 #define	PROMISE_MAX_SUBDISKS	2
49 #define	PROMISE_META_OFFSET	14
50 
51 struct promise_raid_disk {
52 	uint8_t		flags;			/* Subdisk status. */
53 #define PROMISE_F_VALID		0x01
54 #define PROMISE_F_ONLINE	0x02
55 #define PROMISE_F_ASSIGNED	0x04
56 #define PROMISE_F_SPARE		0x08
57 #define PROMISE_F_DUPLICATE	0x10
58 #define PROMISE_F_REDIR		0x20
59 #define PROMISE_F_DOWN		0x40
60 #define PROMISE_F_READY		0x80
61 
62 	uint8_t		number;			/* Position in a volume. */
63 	uint8_t		channel;		/* ATA channel number. */
64 	uint8_t		device;			/* ATA device number. */
65 	uint64_t	id __packed;		/* Subdisk ID. */
66 } __packed;
67 
68 struct promise_raid_conf {
69 	char		promise_id[24];
70 #define PROMISE_MAGIC		"Promise Technology, Inc."
71 #define FREEBSD_MAGIC		"FreeBSD ATA driver RAID "
72 
73 	uint32_t	dummy_0;
74 	uint64_t	magic_0;
75 #define PROMISE_MAGIC0(x)	(((uint64_t)(x.channel) << 48) | \
76 				((uint64_t)(x.device != 0) << 56))
77 	uint16_t	magic_1;
78 	uint32_t	magic_2;
79 	uint8_t		filler1[470];
80 
81 	uint32_t	integrity;
82 #define PROMISE_I_VALID		0x00000080
83 
84 	struct promise_raid_disk	disk;	/* This subdisk info. */
85 	uint32_t	disk_offset;		/* Subdisk offset. */
86 	uint32_t	disk_sectors;		/* Subdisk size */
87 	uint32_t	disk_rebuild;		/* Rebuild position. */
88 	uint16_t	generation;		/* Generation number. */
89 	uint8_t		status;			/* Volume status. */
90 #define PROMISE_S_VALID		0x01
91 #define PROMISE_S_ONLINE	0x02
92 #define PROMISE_S_INITED	0x04
93 #define PROMISE_S_READY		0x08
94 #define PROMISE_S_DEGRADED	0x10
95 #define PROMISE_S_MARKED	0x20
96 #define PROMISE_S_MIGRATING	0x40
97 #define PROMISE_S_FUNCTIONAL	0x80
98 
99 	uint8_t		type;			/* Voluem type. */
100 #define PROMISE_T_RAID0		0x00
101 #define PROMISE_T_RAID1		0x01
102 #define PROMISE_T_RAID3		0x02
103 #define PROMISE_T_RAID5		0x04
104 #define PROMISE_T_SPAN		0x08
105 #define PROMISE_T_JBOD		0x10
106 
107 	uint8_t		total_disks;		/* Disks in this volume. */
108 	uint8_t		stripe_shift;		/* Strip size. */
109 	uint8_t		array_width;		/* Number of RAID0 stripes. */
110 	uint8_t		array_number;		/* Global volume number. */
111 	uint32_t	total_sectors;		/* Volume size. */
112 	uint16_t	cylinders;		/* Volume geometry: C. */
113 	uint8_t		heads;			/* Volume geometry: H. */
114 	uint8_t		sectors;		/* Volume geometry: S. */
115 	uint64_t	volume_id __packed;	/* Volume ID, */
116 	struct promise_raid_disk	disks[PROMISE_MAX_DISKS];
117 						/* Subdisks in this volume. */
118 	char		name[32];		/* Volume label. */
119 
120 	uint32_t	filler2[8];
121 	uint32_t	magic_3;	/* Something related to rebuild. */
122 	uint64_t	rebuild_lba64;	/* Per-volume rebuild position. */
123 	uint32_t	magic_4;
124 	uint32_t	magic_5;
125 	uint32_t	total_sectors_high;
126 	uint8_t		magic_6;
127 	uint8_t		sector_size;
128 	uint16_t	magic_7;
129 	uint32_t	magic_8[31];
130 	uint32_t	backup_time;
131 	uint16_t	magic_9;
132 	uint32_t	disk_offset_high;
133 	uint32_t	disk_sectors_high;
134 	uint32_t	disk_rebuild_high;
135 	uint16_t	magic_10;
136 	uint32_t	magic_11[3];
137 	uint32_t	filler3[284];
138 	uint32_t	checksum;
139 } __packed;
140 
141 struct g_raid_md_promise_perdisk {
142 	int		 pd_updated;
143 	int		 pd_subdisks;
144 	struct promise_raid_conf	*pd_meta[PROMISE_MAX_SUBDISKS];
145 };
146 
147 struct g_raid_md_promise_pervolume {
148 	struct promise_raid_conf	*pv_meta;
149 	uint64_t			 pv_id;
150 	uint16_t			 pv_generation;
151 	int				 pv_disks_present;
152 	int				 pv_started;
153 	struct callout			 pv_start_co;	/* STARTING state timer. */
154 };
155 
156 static g_raid_md_create_t g_raid_md_create_promise;
157 static g_raid_md_taste_t g_raid_md_taste_promise;
158 static g_raid_md_event_t g_raid_md_event_promise;
159 static g_raid_md_volume_event_t g_raid_md_volume_event_promise;
160 static g_raid_md_ctl_t g_raid_md_ctl_promise;
161 static g_raid_md_write_t g_raid_md_write_promise;
162 static g_raid_md_fail_disk_t g_raid_md_fail_disk_promise;
163 static g_raid_md_free_disk_t g_raid_md_free_disk_promise;
164 static g_raid_md_free_volume_t g_raid_md_free_volume_promise;
165 static g_raid_md_free_t g_raid_md_free_promise;
166 
167 static kobj_method_t g_raid_md_promise_methods[] = {
168 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_promise),
169 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_promise),
170 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_promise),
171 	KOBJMETHOD(g_raid_md_volume_event,	g_raid_md_volume_event_promise),
172 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_promise),
173 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_promise),
174 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_promise),
175 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_promise),
176 	KOBJMETHOD(g_raid_md_free_volume,	g_raid_md_free_volume_promise),
177 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_promise),
178 	{ 0, 0 }
179 };
180 
181 static struct g_raid_md_class g_raid_md_promise_class = {
182 	"Promise",
183 	g_raid_md_promise_methods,
184 	sizeof(struct g_raid_md_object),
185 	.mdc_enable = 1,
186 	.mdc_priority = 100
187 };
188 
189 
190 static void
191 g_raid_md_promise_print(struct promise_raid_conf *meta)
192 {
193 	int i;
194 
195 	if (g_raid_debug < 1)
196 		return;
197 
198 	printf("********* ATA Promise Metadata *********\n");
199 	printf("promise_id          <%.24s>\n", meta->promise_id);
200 	printf("disk                %02x %02x %02x %02x %016jx\n",
201 	    meta->disk.flags, meta->disk.number, meta->disk.channel,
202 	    meta->disk.device, meta->disk.id);
203 	printf("disk_offset         %u\n", meta->disk_offset);
204 	printf("disk_sectors        %u\n", meta->disk_sectors);
205 	printf("disk_rebuild        %u\n", meta->disk_rebuild);
206 	printf("generation          %u\n", meta->generation);
207 	printf("status              0x%02x\n", meta->status);
208 	printf("type                %u\n", meta->type);
209 	printf("total_disks         %u\n", meta->total_disks);
210 	printf("stripe_shift        %u\n", meta->stripe_shift);
211 	printf("array_width         %u\n", meta->array_width);
212 	printf("array_number        %u\n", meta->array_number);
213 	printf("total_sectors       %u\n", meta->total_sectors);
214 	printf("cylinders           %u\n", meta->cylinders);
215 	printf("heads               %u\n", meta->heads);
216 	printf("sectors             %u\n", meta->sectors);
217 	printf("volume_id           0x%016jx\n", meta->volume_id);
218 	printf("disks:\n");
219 	for (i = 0; i < PROMISE_MAX_DISKS; i++ ) {
220 		printf("                    %02x %02x %02x %02x %016jx\n",
221 		    meta->disks[i].flags, meta->disks[i].number,
222 		    meta->disks[i].channel, meta->disks[i].device,
223 		    meta->disks[i].id);
224 	}
225 	printf("name                <%.32s>\n", meta->name);
226 	printf("magic_3             0x%08x\n", meta->magic_3);
227 	printf("rebuild_lba64       %ju\n", meta->rebuild_lba64);
228 	printf("magic_4             0x%08x\n", meta->magic_4);
229 	printf("magic_5             0x%08x\n", meta->magic_5);
230 	printf("total_sectors_high  0x%08x\n", meta->total_sectors_high);
231 	printf("sector_size         %u\n", meta->sector_size);
232 	printf("backup_time         %d\n", meta->backup_time);
233 	printf("disk_offset_high    0x%08x\n", meta->disk_offset_high);
234 	printf("disk_sectors_high   0x%08x\n", meta->disk_sectors_high);
235 	printf("disk_rebuild_high   0x%08x\n", meta->disk_rebuild_high);
236 	printf("=================================================\n");
237 }
238 
239 static struct promise_raid_conf *
240 promise_meta_copy(struct promise_raid_conf *meta)
241 {
242 	struct promise_raid_conf *nmeta;
243 
244 	nmeta = malloc(sizeof(*nmeta), M_MD_PROMISE, M_WAITOK);
245 	memcpy(nmeta, meta, sizeof(*nmeta));
246 	return (nmeta);
247 }
248 
249 static int
250 promise_meta_find_disk(struct promise_raid_conf *meta, uint64_t id)
251 {
252 	int pos;
253 
254 	for (pos = 0; pos < meta->total_disks; pos++) {
255 		if (meta->disks[pos].id == id)
256 			return (pos);
257 	}
258 	return (-1);
259 }
260 
261 static int
262 promise_meta_unused_range(struct promise_raid_conf **metaarr, int nsd,
263     off_t sectors, off_t *off, off_t *size)
264 {
265 	off_t coff, csize, tmp;
266 	int i, j;
267 
268 	sectors -= 131072;
269 	*off = 0;
270 	*size = 0;
271 	coff = 0;
272 	csize = sectors;
273 	i = 0;
274 	while (1) {
275 		for (j = 0; j < nsd; j++) {
276 			tmp = ((off_t)metaarr[j]->disk_offset_high << 32) +
277 			    metaarr[j]->disk_offset;
278 			if (tmp >= coff)
279 				csize = MIN(csize, tmp - coff);
280 		}
281 		if (csize > *size) {
282 			*off = coff;
283 			*size = csize;
284 		}
285 		if (i >= nsd)
286 			break;
287 		coff = ((off_t)metaarr[i]->disk_offset_high << 32) +
288 		     metaarr[i]->disk_offset +
289 		    ((off_t)metaarr[i]->disk_sectors_high << 32) +
290 		     metaarr[i]->disk_sectors;
291 		csize = sectors - coff;
292 		i++;
293 	};
294 	return ((*size > 0) ? 1 : 0);
295 }
296 
297 static int
298 promise_meta_translate_disk(struct g_raid_volume *vol, int md_disk_pos)
299 {
300 	int disk_pos, width;
301 
302 	if (md_disk_pos >= 0 && vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
303 		width = vol->v_disks_count / 2;
304 		disk_pos = (md_disk_pos / width) +
305 		    (md_disk_pos % width) * width;
306 	} else
307 		disk_pos = md_disk_pos;
308 	return (disk_pos);
309 }
310 
311 static void
312 promise_meta_get_name(struct promise_raid_conf *meta, char *buf)
313 {
314 	int i;
315 
316 	strncpy(buf, meta->name, 32);
317 	buf[32] = 0;
318 	for (i = 31; i >= 0; i--) {
319 		if (buf[i] > 0x20)
320 			break;
321 		buf[i] = 0;
322 	}
323 }
324 
325 static void
326 promise_meta_put_name(struct promise_raid_conf *meta, char *buf)
327 {
328 
329 	memset(meta->name, 0x20, 32);
330 	memcpy(meta->name, buf, MIN(strlen(buf), 32));
331 }
332 
333 static int
334 promise_meta_read(struct g_consumer *cp, struct promise_raid_conf **metaarr)
335 {
336 	struct g_provider *pp;
337 	struct promise_raid_conf *meta;
338 	char *buf;
339 	int error, i, subdisks;
340 	uint32_t checksum, *ptr;
341 
342 	pp = cp->provider;
343 	subdisks = 0;
344 next:
345 	/* Read metadata block. */
346 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize *
347 	    (63 - subdisks * PROMISE_META_OFFSET),
348 	    pp->sectorsize * 4, &error);
349 	if (buf == NULL) {
350 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
351 		    pp->name, error);
352 		return (subdisks);
353 	}
354 	meta = (struct promise_raid_conf *)buf;
355 
356 	/* Check if this is an Promise RAID struct */
357 	if (strncmp(meta->promise_id, PROMISE_MAGIC, strlen(PROMISE_MAGIC)) &&
358 	    strncmp(meta->promise_id, FREEBSD_MAGIC, strlen(FREEBSD_MAGIC))) {
359 		if (subdisks == 0)
360 			G_RAID_DEBUG(1,
361 			    "Promise signature check failed on %s", pp->name);
362 		g_free(buf);
363 		return (subdisks);
364 	}
365 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK);
366 	memcpy(meta, buf, MIN(sizeof(*meta), pp->sectorsize * 4));
367 	g_free(buf);
368 
369 	/* Check metadata checksum. */
370 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
371 		checksum += *ptr++;
372 	if (checksum != meta->checksum) {
373 		G_RAID_DEBUG(1, "Promise checksum check failed on %s", pp->name);
374 		free(meta, M_MD_PROMISE);
375 		return (subdisks);
376 	}
377 
378 	if ((meta->integrity & PROMISE_I_VALID) == 0) {
379 		G_RAID_DEBUG(1, "Promise metadata is invalid on %s", pp->name);
380 		free(meta, M_MD_PROMISE);
381 		return (subdisks);
382 	}
383 
384 	if (meta->total_disks > PROMISE_MAX_DISKS) {
385 		G_RAID_DEBUG(1, "Wrong number of disks on %s (%d)",
386 		    pp->name, meta->total_disks);
387 		free(meta, M_MD_PROMISE);
388 		return (subdisks);
389 	}
390 
391 	/* Remove filler garbage from fields used in newer metadata. */
392 	if (meta->disk_offset_high == 0x8b8c8d8e &&
393 	    meta->disk_sectors_high == 0x8788898a &&
394 	    meta->disk_rebuild_high == 0x83848586) {
395 		meta->disk_offset_high = 0;
396 		meta->disk_sectors_high = 0;
397 		if (meta->disk_rebuild == UINT32_MAX)
398 			meta->disk_rebuild_high = UINT32_MAX;
399 		else
400 			meta->disk_rebuild_high = 0;
401 		if (meta->total_sectors_high == 0x15161718) {
402 			meta->total_sectors_high = 0;
403 			meta->backup_time = 0;
404 			if (meta->rebuild_lba64 == 0x2122232425262728)
405 				meta->rebuild_lba64 = UINT64_MAX;
406 		}
407 	}
408 	if (meta->sector_size < 1 || meta->sector_size > 8)
409 		meta->sector_size = 1;
410 
411 	/* Save this part and look for next. */
412 	*metaarr = meta;
413 	metaarr++;
414 	subdisks++;
415 	if (subdisks < PROMISE_MAX_SUBDISKS)
416 		goto next;
417 
418 	return (subdisks);
419 }
420 
421 static int
422 promise_meta_write(struct g_consumer *cp,
423     struct promise_raid_conf **metaarr, int nsd)
424 {
425 	struct g_provider *pp;
426 	struct promise_raid_conf *meta;
427 	char *buf;
428 	off_t off, size;
429 	int error, i, subdisk, fake;
430 	uint32_t checksum, *ptr;
431 
432 	pp = cp->provider;
433 	subdisk = 0;
434 	fake = 0;
435 next:
436 	buf = malloc(pp->sectorsize * 4, M_MD_PROMISE, M_WAITOK | M_ZERO);
437 	meta = NULL;
438 	if (subdisk < nsd) {
439 		meta = metaarr[subdisk];
440 	} else if (!fake && promise_meta_unused_range(metaarr, nsd,
441 	    cp->provider->mediasize / cp->provider->sectorsize,
442 	    &off, &size)) {
443 		/* Optionally add record for unused space. */
444 		meta = (struct promise_raid_conf *)buf;
445 		memcpy(&meta->promise_id[0], PROMISE_MAGIC,
446 		    sizeof(PROMISE_MAGIC) - 1);
447 		meta->dummy_0 = 0x00020000;
448 		meta->integrity = PROMISE_I_VALID;
449 		meta->disk.flags = PROMISE_F_ONLINE | PROMISE_F_VALID;
450 		meta->disk.number = 0xff;
451 		arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
452 		meta->disk_offset_high = off >> 32;
453 		meta->disk_offset = (uint32_t)off;
454 		meta->disk_sectors_high = size >> 32;
455 		meta->disk_sectors = (uint32_t)size;
456 		meta->disk_rebuild_high = UINT32_MAX;
457 		meta->disk_rebuild = UINT32_MAX;
458 		fake = 1;
459 	}
460 	if (meta != NULL) {
461 		/* Recalculate checksum for case if metadata were changed. */
462 		meta->checksum = 0;
463 		for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
464 			checksum += *ptr++;
465 		meta->checksum = checksum;
466 		memcpy(buf, meta, MIN(pp->sectorsize * 4, sizeof(*meta)));
467 	}
468 	error = g_write_data(cp, pp->mediasize - pp->sectorsize *
469 	    (63 - subdisk * PROMISE_META_OFFSET),
470 	    buf, pp->sectorsize * 4);
471 	if (error != 0) {
472 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
473 		    pp->name, error);
474 	}
475 	free(buf, M_MD_PROMISE);
476 
477 	subdisk++;
478 	if (subdisk < PROMISE_MAX_SUBDISKS)
479 		goto next;
480 
481 	return (error);
482 }
483 
484 static int
485 promise_meta_erase(struct g_consumer *cp)
486 {
487 	struct g_provider *pp;
488 	char *buf;
489 	int error, subdisk;
490 
491 	pp = cp->provider;
492 	buf = malloc(4 * pp->sectorsize, M_MD_PROMISE, M_WAITOK | M_ZERO);
493 	for (subdisk = 0; subdisk < PROMISE_MAX_SUBDISKS; subdisk++) {
494 		error = g_write_data(cp, pp->mediasize - pp->sectorsize *
495 		    (63 - subdisk * PROMISE_META_OFFSET),
496 		    buf, 4 * pp->sectorsize);
497 		if (error != 0) {
498 			G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
499 			    pp->name, error);
500 		}
501 	}
502 	free(buf, M_MD_PROMISE);
503 	return (error);
504 }
505 
506 static int
507 promise_meta_write_spare(struct g_consumer *cp)
508 {
509 	struct promise_raid_conf *meta;
510 	off_t tmp;
511 	int error;
512 
513 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
514 	memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC) - 1);
515 	meta->dummy_0 = 0x00020000;
516 	meta->integrity = PROMISE_I_VALID;
517 	meta->disk.flags = PROMISE_F_SPARE | PROMISE_F_ONLINE | PROMISE_F_VALID;
518 	meta->disk.number = 0xff;
519 	arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
520 	tmp = cp->provider->mediasize / cp->provider->sectorsize - 131072;
521 	meta->disk_sectors_high = tmp >> 32;
522 	meta->disk_sectors = (uint32_t)tmp;
523 	meta->disk_rebuild_high = UINT32_MAX;
524 	meta->disk_rebuild = UINT32_MAX;
525 	error = promise_meta_write(cp, &meta, 1);
526 	free(meta, M_MD_PROMISE);
527 	return (error);
528 }
529 
530 static struct g_raid_volume *
531 g_raid_md_promise_get_volume(struct g_raid_softc *sc, uint64_t id)
532 {
533 	struct g_raid_volume	*vol;
534 	struct g_raid_md_promise_pervolume *pv;
535 
536 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
537 		pv = vol->v_md_data;
538 		if (pv->pv_id == id)
539 			break;
540 	}
541 	return (vol);
542 }
543 
544 static int
545 g_raid_md_promise_purge_volumes(struct g_raid_softc *sc)
546 {
547 	struct g_raid_volume	*vol, *tvol;
548 	struct g_raid_md_promise_pervolume *pv;
549 	int i, res;
550 
551 	res = 0;
552 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tvol) {
553 		pv = vol->v_md_data;
554 		if (!pv->pv_started || vol->v_stopping)
555 			continue;
556 		for (i = 0; i < vol->v_disks_count; i++) {
557 			if (vol->v_subdisks[i].sd_state != G_RAID_SUBDISK_S_NONE)
558 				break;
559 		}
560 		if (i >= vol->v_disks_count) {
561 			g_raid_destroy_volume(vol);
562 			res = 1;
563 		}
564 	}
565 	return (res);
566 }
567 
568 static int
569 g_raid_md_promise_purge_disks(struct g_raid_softc *sc)
570 {
571 	struct g_raid_disk	*disk, *tdisk;
572 	struct g_raid_volume	*vol;
573 	struct g_raid_md_promise_perdisk *pd;
574 	int i, j, res;
575 
576 	res = 0;
577 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) {
578 		if (disk->d_state == G_RAID_DISK_S_SPARE)
579 			continue;
580 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
581 
582 		/* Scan for deleted volumes. */
583 		for (i = 0; i < pd->pd_subdisks; ) {
584 			vol = g_raid_md_promise_get_volume(sc,
585 			    pd->pd_meta[i]->volume_id);
586 			if (vol != NULL && !vol->v_stopping) {
587 				i++;
588 				continue;
589 			}
590 			free(pd->pd_meta[i], M_MD_PROMISE);
591 			for (j = i; j < pd->pd_subdisks - 1; j++)
592 				pd->pd_meta[j] = pd->pd_meta[j + 1];
593 			pd->pd_meta[pd->pd_subdisks - 1] = NULL;
594 			pd->pd_subdisks--;
595 			pd->pd_updated = 1;
596 		}
597 
598 		/* If there is no metadata left - erase and delete disk. */
599 		if (pd->pd_subdisks == 0) {
600 			promise_meta_erase(disk->d_consumer);
601 			g_raid_destroy_disk(disk);
602 			res = 1;
603 		}
604 	}
605 	return (res);
606 }
607 
608 static int
609 g_raid_md_promise_supported(int level, int qual, int disks, int force)
610 {
611 
612 	if (disks > PROMISE_MAX_DISKS)
613 		return (0);
614 	switch (level) {
615 	case G_RAID_VOLUME_RL_RAID0:
616 		if (disks < 1)
617 			return (0);
618 		if (!force && disks < 2)
619 			return (0);
620 		break;
621 	case G_RAID_VOLUME_RL_RAID1:
622 		if (disks < 1)
623 			return (0);
624 		if (!force && (disks != 2))
625 			return (0);
626 		break;
627 	case G_RAID_VOLUME_RL_RAID1E:
628 		if (disks < 2)
629 			return (0);
630 		if (disks % 2 != 0)
631 			return (0);
632 		if (!force && (disks != 4))
633 			return (0);
634 		break;
635 	case G_RAID_VOLUME_RL_SINGLE:
636 		if (disks != 1)
637 			return (0);
638 		break;
639 	case G_RAID_VOLUME_RL_CONCAT:
640 		if (disks < 2)
641 			return (0);
642 		break;
643 	case G_RAID_VOLUME_RL_RAID5:
644 		if (disks < 3)
645 			return (0);
646 		if (qual != G_RAID_VOLUME_RLQ_R5LA)
647 			return (0);
648 		break;
649 	default:
650 		return (0);
651 	}
652 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
653 		return (0);
654 	return (1);
655 }
656 
657 static int
658 g_raid_md_promise_start_disk(struct g_raid_disk *disk, int sdn,
659     struct g_raid_volume *vol)
660 {
661 	struct g_raid_softc *sc;
662 	struct g_raid_subdisk *sd;
663 	struct g_raid_md_promise_perdisk *pd;
664 	struct g_raid_md_promise_pervolume *pv;
665 	struct promise_raid_conf *meta;
666 	off_t eoff, esize, size;
667 	int disk_pos, md_disk_pos, i, resurrection = 0;
668 
669 	sc = disk->d_softc;
670 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
671 
672 	pv = vol->v_md_data;
673 	meta = pv->pv_meta;
674 
675 	if (sdn >= 0) {
676 		/* Find disk position in metadata by it's serial. */
677 		md_disk_pos = promise_meta_find_disk(meta, pd->pd_meta[sdn]->disk.id);
678 		/* For RAID0+1 we need to translate order. */
679 		disk_pos = promise_meta_translate_disk(vol, md_disk_pos);
680 	} else {
681 		md_disk_pos = -1;
682 		disk_pos = -1;
683 	}
684 	if (disk_pos < 0) {
685 		G_RAID_DEBUG1(1, sc, "Disk %s is not part of the volume %s",
686 		    g_raid_get_diskname(disk), vol->v_name);
687 		/* Failed stale disk is useless for us. */
688 		if (sdn >= 0 &&
689 		    pd->pd_meta[sdn]->disk.flags & PROMISE_F_DOWN) {
690 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
691 			return (0);
692 		}
693 		/* If we were given specific metadata subdisk - erase it. */
694 		if (sdn >= 0) {
695 			free(pd->pd_meta[sdn], M_MD_PROMISE);
696 			for (i = sdn; i < pd->pd_subdisks - 1; i++)
697 				pd->pd_meta[i] = pd->pd_meta[i + 1];
698 			pd->pd_meta[pd->pd_subdisks - 1] = NULL;
699 			pd->pd_subdisks--;
700 		}
701 		/* If we are in the start process, that's all for now. */
702 		if (!pv->pv_started)
703 			goto nofit;
704 		/*
705 		 * If we have already started - try to get use of the disk.
706 		 * Try to replace OFFLINE disks first, then FAILED.
707 		 */
708 		promise_meta_unused_range(pd->pd_meta, pd->pd_subdisks,
709 		    disk->d_consumer->provider->mediasize /
710 		    disk->d_consumer->provider->sectorsize,
711 		    &eoff, &esize);
712 		if (esize == 0) {
713 			G_RAID_DEBUG1(1, sc, "No free space on disk %s",
714 			    g_raid_get_diskname(disk));
715 			goto nofit;
716 		}
717 		size = INT64_MAX;
718 		for (i = 0; i < vol->v_disks_count; i++) {
719 			sd = &vol->v_subdisks[i];
720 			if (sd->sd_state != G_RAID_SUBDISK_S_NONE)
721 				size = sd->sd_size;
722 			if (sd->sd_state <= G_RAID_SUBDISK_S_FAILED &&
723 			    (disk_pos < 0 ||
724 			     vol->v_subdisks[i].sd_state < sd->sd_state))
725 				disk_pos = i;
726 		}
727 		if (disk_pos >= 0 &&
728 		    vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT &&
729 		    (off_t)esize * 512 < size) {
730 			G_RAID_DEBUG1(1, sc, "Disk %s free space "
731 			    "is too small (%ju < %ju)",
732 			    g_raid_get_diskname(disk),
733 			    (off_t)esize * 512, size);
734 			disk_pos = -1;
735 		}
736 		if (disk_pos >= 0) {
737 			if (vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT)
738 				esize = size / 512;
739 			/* For RAID0+1 we need to translate order. */
740 			md_disk_pos = promise_meta_translate_disk(vol, disk_pos);
741 		} else {
742 nofit:
743 			if (pd->pd_subdisks == 0) {
744 				g_raid_change_disk_state(disk,
745 				    G_RAID_DISK_S_SPARE);
746 			}
747 			return (0);
748 		}
749 		G_RAID_DEBUG1(1, sc, "Disk %s takes pos %d in the volume %s",
750 		    g_raid_get_diskname(disk), disk_pos, vol->v_name);
751 		resurrection = 1;
752 	}
753 
754 	sd = &vol->v_subdisks[disk_pos];
755 
756 	if (resurrection && sd->sd_disk != NULL) {
757 		g_raid_change_disk_state(sd->sd_disk,
758 		    G_RAID_DISK_S_STALE_FAILED);
759 		TAILQ_REMOVE(&sd->sd_disk->d_subdisks,
760 		    sd, sd_next);
761 	}
762 	vol->v_subdisks[disk_pos].sd_disk = disk;
763 	TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
764 
765 	/* Welcome the new disk. */
766 	if (resurrection)
767 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
768 	else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN)
769 		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
770 	else
771 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
772 
773 	if (resurrection) {
774 		sd->sd_offset = (off_t)eoff * 512;
775 		sd->sd_size = (off_t)esize * 512;
776 	} else {
777 		sd->sd_offset = (((off_t)pd->pd_meta[sdn]->disk_offset_high
778 		    << 32) + pd->pd_meta[sdn]->disk_offset) * 512;
779 		sd->sd_size = (((off_t)pd->pd_meta[sdn]->disk_sectors_high
780 		    << 32) + pd->pd_meta[sdn]->disk_sectors) * 512;
781 	}
782 
783 	if (resurrection) {
784 		/* Stale disk, almost same as new. */
785 		g_raid_change_subdisk_state(sd,
786 		    G_RAID_SUBDISK_S_NEW);
787 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) {
788 		/* Failed disk. */
789 		g_raid_change_subdisk_state(sd,
790 		    G_RAID_SUBDISK_S_FAILED);
791 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_REDIR) {
792 		/* Rebuilding disk. */
793 		g_raid_change_subdisk_state(sd,
794 		    G_RAID_SUBDISK_S_REBUILD);
795 		if (pd->pd_meta[sdn]->generation != meta->generation)
796 			sd->sd_rebuild_pos = 0;
797 		else {
798 			sd->sd_rebuild_pos =
799 			    (((off_t)pd->pd_meta[sdn]->disk_rebuild_high << 32) +
800 			     pd->pd_meta[sdn]->disk_rebuild) * 512;
801 		}
802 	} else if (!(meta->disks[md_disk_pos].flags & PROMISE_F_ONLINE)) {
803 		/* Rebuilding disk. */
804 		g_raid_change_subdisk_state(sd,
805 		    G_RAID_SUBDISK_S_NEW);
806 	} else if (pd->pd_meta[sdn]->generation != meta->generation ||
807 	    (meta->status & PROMISE_S_MARKED)) {
808 		/* Stale disk or dirty volume (unclean shutdown). */
809 		g_raid_change_subdisk_state(sd,
810 		    G_RAID_SUBDISK_S_STALE);
811 	} else {
812 		/* Up to date disk. */
813 		g_raid_change_subdisk_state(sd,
814 		    G_RAID_SUBDISK_S_ACTIVE);
815 	}
816 	g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
817 	    G_RAID_EVENT_SUBDISK);
818 
819 	return (resurrection);
820 }
821 
822 static void
823 g_raid_md_promise_refill(struct g_raid_softc *sc)
824 {
825 	struct g_raid_volume *vol;
826 	struct g_raid_subdisk *sd;
827 	struct g_raid_disk *disk;
828 	struct g_raid_md_object *md;
829 	struct g_raid_md_promise_perdisk *pd;
830 	struct g_raid_md_promise_pervolume *pv;
831 	int update, updated, i, bad;
832 
833 	md = sc->sc_md;
834 restart:
835 	updated = 0;
836 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
837 		pv = vol->v_md_data;
838 		if (!pv->pv_started || vol->v_stopping)
839 			continue;
840 
841 		/* Search for subdisk that needs replacement. */
842 		bad = 0;
843 		for (i = 0; i < vol->v_disks_count; i++) {
844 			sd = &vol->v_subdisks[i];
845 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
846 			    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
847 			        bad = 1;
848 		}
849 		if (!bad)
850 			continue;
851 
852 		G_RAID_DEBUG1(1, sc, "Volume %s is not complete, "
853 		    "trying to refill.", vol->v_name);
854 
855 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
856 			/* Skip failed. */
857 			if (disk->d_state < G_RAID_DISK_S_SPARE)
858 				continue;
859 			/* Skip already used by this volume. */
860 			for (i = 0; i < vol->v_disks_count; i++) {
861 				sd = &vol->v_subdisks[i];
862 				if (sd->sd_disk == disk)
863 					break;
864 			}
865 			if (i < vol->v_disks_count)
866 				continue;
867 
868 			/* Try to use disk if it has empty extents. */
869 			pd = disk->d_md_data;
870 			if (pd->pd_subdisks < PROMISE_MAX_SUBDISKS) {
871 				update =
872 				    g_raid_md_promise_start_disk(disk, -1, vol);
873 			} else
874 				update = 0;
875 			if (update) {
876 				updated = 1;
877 				g_raid_md_write_promise(md, vol, NULL, disk);
878 				break;
879 			}
880 		}
881 	}
882 	if (updated)
883 		goto restart;
884 }
885 
886 static void
887 g_raid_md_promise_start(struct g_raid_volume *vol)
888 {
889 	struct g_raid_softc *sc;
890 	struct g_raid_subdisk *sd;
891 	struct g_raid_disk *disk;
892 	struct g_raid_md_object *md;
893 	struct g_raid_md_promise_perdisk *pd;
894 	struct g_raid_md_promise_pervolume *pv;
895 	struct promise_raid_conf *meta;
896 	int i;
897 
898 	sc = vol->v_softc;
899 	md = sc->sc_md;
900 	pv = vol->v_md_data;
901 	meta = pv->pv_meta;
902 
903 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
904 	if (meta->type == PROMISE_T_RAID0)
905 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
906 	else if (meta->type == PROMISE_T_RAID1) {
907 		if (meta->array_width == 1)
908 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
909 		else
910 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
911 	} else if (meta->type == PROMISE_T_RAID3)
912 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID3;
913 	else if (meta->type == PROMISE_T_RAID5) {
914 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
915 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
916 	} else if (meta->type == PROMISE_T_SPAN)
917 		vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
918 	else if (meta->type == PROMISE_T_JBOD)
919 		vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
920 	else
921 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
922 	vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ
923 	vol->v_disks_count = meta->total_disks;
924 	vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ
925 	if (meta->total_sectors_high < 256) /* If value looks sane. */
926 		vol->v_mediasize +=
927 		    ((off_t)meta->total_sectors_high << 32) * 512; //ZZZ
928 	vol->v_sectorsize = 512 * meta->sector_size;
929 	for (i = 0; i < vol->v_disks_count; i++) {
930 		sd = &vol->v_subdisks[i];
931 		sd->sd_offset = (((off_t)meta->disk_offset_high << 32) +
932 		    meta->disk_offset) * 512;
933 		sd->sd_size = (((off_t)meta->disk_sectors_high << 32) +
934 		    meta->disk_sectors) * 512;
935 	}
936 	g_raid_start_volume(vol);
937 
938 	/* Make all disks found till the moment take their places. */
939 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
940 		pd = disk->d_md_data;
941 		for (i = 0; i < pd->pd_subdisks; i++) {
942 			if (pd->pd_meta[i]->volume_id == meta->volume_id)
943 				g_raid_md_promise_start_disk(disk, i, vol);
944 		}
945 	}
946 
947 	pv->pv_started = 1;
948 	callout_stop(&pv->pv_start_co);
949 	G_RAID_DEBUG1(0, sc, "Volume started.");
950 	g_raid_md_write_promise(md, vol, NULL, NULL);
951 
952 	/* Pickup any STALE/SPARE disks to refill array if needed. */
953 	g_raid_md_promise_refill(sc);
954 
955 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
956 }
957 
958 static void
959 g_raid_promise_go(void *arg)
960 {
961 	struct g_raid_volume *vol;
962 	struct g_raid_softc *sc;
963 	struct g_raid_md_promise_pervolume *pv;
964 
965 	vol = arg;
966 	pv = vol->v_md_data;
967 	sc = vol->v_softc;
968 	if (!pv->pv_started) {
969 		G_RAID_DEBUG1(0, sc, "Force volume start due to timeout.");
970 		g_raid_event_send(vol, G_RAID_VOLUME_E_STARTMD,
971 		    G_RAID_EVENT_VOLUME);
972 	}
973 }
974 
975 static void
976 g_raid_md_promise_new_disk(struct g_raid_disk *disk)
977 {
978 	struct g_raid_softc *sc;
979 	struct g_raid_md_object *md;
980 	struct promise_raid_conf *pdmeta;
981 	struct g_raid_md_promise_perdisk *pd;
982 	struct g_raid_md_promise_pervolume *pv;
983 	struct g_raid_volume *vol;
984 	int i;
985 	char buf[33];
986 
987 	sc = disk->d_softc;
988 	md = sc->sc_md;
989 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
990 
991 	if (pd->pd_subdisks == 0) {
992 		g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
993 		g_raid_md_promise_refill(sc);
994 		return;
995 	}
996 
997 	for (i = 0; i < pd->pd_subdisks; i++) {
998 		pdmeta = pd->pd_meta[i];
999 
1000 		/* Look for volume with matching ID. */
1001 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1002 		if (vol == NULL) {
1003 			promise_meta_get_name(pdmeta, buf);
1004 			vol = g_raid_create_volume(sc, buf, pdmeta->array_number);
1005 			pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1006 			pv->pv_id = pdmeta->volume_id;
1007 			vol->v_md_data = pv;
1008 			callout_init(&pv->pv_start_co, 1);
1009 			callout_reset(&pv->pv_start_co,
1010 			    g_raid_start_timeout * hz,
1011 			    g_raid_promise_go, vol);
1012 		} else
1013 			pv = vol->v_md_data;
1014 
1015 		/* If we haven't started yet - check metadata freshness. */
1016 		if (pv->pv_meta == NULL || !pv->pv_started) {
1017 			if (pv->pv_meta == NULL ||
1018 			    ((int16_t)(pdmeta->generation - pv->pv_generation)) > 0) {
1019 				G_RAID_DEBUG1(1, sc, "Newer disk");
1020 				if (pv->pv_meta != NULL)
1021 					free(pv->pv_meta, M_MD_PROMISE);
1022 				pv->pv_meta = promise_meta_copy(pdmeta);
1023 				pv->pv_generation = pv->pv_meta->generation;
1024 				pv->pv_disks_present = 1;
1025 			} else if (pdmeta->generation == pv->pv_generation) {
1026 				pv->pv_disks_present++;
1027 				G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
1028 				    pv->pv_disks_present,
1029 				    pv->pv_meta->total_disks);
1030 			} else {
1031 				G_RAID_DEBUG1(1, sc, "Older disk");
1032 			}
1033 		}
1034 	}
1035 
1036 	for (i = 0; i < pd->pd_subdisks; i++) {
1037 		pdmeta = pd->pd_meta[i];
1038 
1039 		/* Look for volume with matching ID. */
1040 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1041 		if (vol == NULL)
1042 			continue;
1043 		pv = vol->v_md_data;
1044 
1045 		if (pv->pv_started) {
1046 			if (g_raid_md_promise_start_disk(disk, i, vol))
1047 				g_raid_md_write_promise(md, vol, NULL, NULL);
1048 		} else {
1049 			/* If we collected all needed disks - start array. */
1050 			if (pv->pv_disks_present == pv->pv_meta->total_disks)
1051 				g_raid_md_promise_start(vol);
1052 		}
1053 	}
1054 }
1055 
1056 static int
1057 g_raid_md_create_promise(struct g_raid_md_object *md, struct g_class *mp,
1058     struct g_geom **gp)
1059 {
1060 	struct g_geom *geom;
1061 	struct g_raid_softc *sc;
1062 
1063 	/* Search for existing node. */
1064 	LIST_FOREACH(geom, &mp->geom, geom) {
1065 		sc = geom->softc;
1066 		if (sc == NULL)
1067 			continue;
1068 		if (sc->sc_stopping != 0)
1069 			continue;
1070 		if (sc->sc_md->mdo_class != md->mdo_class)
1071 			continue;
1072 		break;
1073 	}
1074 	if (geom != NULL) {
1075 		*gp = geom;
1076 		return (G_RAID_MD_TASTE_EXISTING);
1077 	}
1078 
1079 	/* Create new one if not found. */
1080 	sc = g_raid_create_node(mp, "Promise", md);
1081 	if (sc == NULL)
1082 		return (G_RAID_MD_TASTE_FAIL);
1083 	md->mdo_softc = sc;
1084 	*gp = sc->sc_geom;
1085 	return (G_RAID_MD_TASTE_NEW);
1086 }
1087 
1088 static int
1089 g_raid_md_taste_promise(struct g_raid_md_object *md, struct g_class *mp,
1090                               struct g_consumer *cp, struct g_geom **gp)
1091 {
1092 	struct g_consumer *rcp;
1093 	struct g_provider *pp;
1094 	struct g_raid_softc *sc;
1095 	struct g_raid_disk *disk;
1096 	struct promise_raid_conf *meta, *metaarr[4];
1097 	struct g_raid_md_promise_perdisk *pd;
1098 	struct g_geom *geom;
1099 	int i, j, result, len, subdisks;
1100 	char name[16];
1101 	uint16_t vendor;
1102 
1103 	G_RAID_DEBUG(1, "Tasting Promise on %s", cp->provider->name);
1104 	pp = cp->provider;
1105 
1106 	/* Read metadata from device. */
1107 	meta = NULL;
1108 	g_topology_unlock();
1109 	vendor = 0xffff;
1110 	len = sizeof(vendor);
1111 	if (pp->geom->rank == 1)
1112 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1113 	subdisks = promise_meta_read(cp, metaarr);
1114 	g_topology_lock();
1115 	if (subdisks == 0) {
1116 		if (g_raid_aggressive_spare) {
1117 			if (vendor == 0x105a || vendor == 0x1002) {
1118 				G_RAID_DEBUG(1,
1119 				    "No Promise metadata, forcing spare.");
1120 				goto search;
1121 			} else {
1122 				G_RAID_DEBUG(1,
1123 				    "Promise/ATI vendor mismatch "
1124 				    "0x%04x != 0x105a/0x1002",
1125 				    vendor);
1126 			}
1127 		}
1128 		return (G_RAID_MD_TASTE_FAIL);
1129 	}
1130 
1131 	/* Metadata valid. Print it. */
1132 	for (i = 0; i < subdisks; i++)
1133 		g_raid_md_promise_print(metaarr[i]);
1134 
1135 	/* Purge meaningless (empty/spare) records. */
1136 	for (i = 0; i < subdisks; ) {
1137 		if (metaarr[i]->disk.flags & PROMISE_F_ASSIGNED) {
1138 			i++;
1139 			continue;
1140 		}
1141 		free(metaarr[i], M_MD_PROMISE);
1142 		for (j = i; j < subdisks - 1; j++)
1143 			metaarr[i] = metaarr[j + 1];
1144 		metaarr[subdisks - 1] = NULL;
1145 		subdisks--;
1146 	}
1147 
1148 search:
1149 	/* Search for matching node. */
1150 	sc = NULL;
1151 	LIST_FOREACH(geom, &mp->geom, geom) {
1152 		sc = geom->softc;
1153 		if (sc == NULL)
1154 			continue;
1155 		if (sc->sc_stopping != 0)
1156 			continue;
1157 		if (sc->sc_md->mdo_class != md->mdo_class)
1158 			continue;
1159 		break;
1160 	}
1161 
1162 	/* Found matching node. */
1163 	if (geom != NULL) {
1164 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1165 		result = G_RAID_MD_TASTE_EXISTING;
1166 
1167 	} else { /* Not found matching node -- create one. */
1168 		result = G_RAID_MD_TASTE_NEW;
1169 		snprintf(name, sizeof(name), "Promise");
1170 		sc = g_raid_create_node(mp, name, md);
1171 		md->mdo_softc = sc;
1172 		geom = sc->sc_geom;
1173 	}
1174 
1175 	/* There is no return after this point, so we close passed consumer. */
1176 	g_access(cp, -1, 0, 0);
1177 
1178 	rcp = g_new_consumer(geom);
1179 	rcp->flags |= G_CF_DIRECT_RECEIVE;
1180 	g_attach(rcp, pp);
1181 	if (g_access(rcp, 1, 1, 1) != 0)
1182 		; //goto fail1;
1183 
1184 	g_topology_unlock();
1185 	sx_xlock(&sc->sc_lock);
1186 
1187 	pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1188 	pd->pd_subdisks = subdisks;
1189 	for (i = 0; i < subdisks; i++)
1190 		pd->pd_meta[i] = metaarr[i];
1191 	disk = g_raid_create_disk(sc);
1192 	disk->d_md_data = (void *)pd;
1193 	disk->d_consumer = rcp;
1194 	rcp->private = disk;
1195 
1196 	g_raid_get_disk_info(disk);
1197 
1198 	g_raid_md_promise_new_disk(disk);
1199 
1200 	sx_xunlock(&sc->sc_lock);
1201 	g_topology_lock();
1202 	*gp = geom;
1203 	return (result);
1204 }
1205 
1206 static int
1207 g_raid_md_event_promise(struct g_raid_md_object *md,
1208     struct g_raid_disk *disk, u_int event)
1209 {
1210 	struct g_raid_softc *sc;
1211 
1212 	sc = md->mdo_softc;
1213 	if (disk == NULL)
1214 		return (-1);
1215 	switch (event) {
1216 	case G_RAID_DISK_E_DISCONNECTED:
1217 		/* Delete disk. */
1218 		g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1219 		g_raid_destroy_disk(disk);
1220 		g_raid_md_promise_purge_volumes(sc);
1221 
1222 		/* Write updated metadata to all disks. */
1223 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1224 
1225 		/* Check if anything left. */
1226 		if (g_raid_ndisks(sc, -1) == 0)
1227 			g_raid_destroy_node(sc, 0);
1228 		else
1229 			g_raid_md_promise_refill(sc);
1230 		return (0);
1231 	}
1232 	return (-2);
1233 }
1234 
1235 static int
1236 g_raid_md_volume_event_promise(struct g_raid_md_object *md,
1237     struct g_raid_volume *vol, u_int event)
1238 {
1239 	struct g_raid_md_promise_pervolume *pv;
1240 
1241 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1242 	switch (event) {
1243 	case G_RAID_VOLUME_E_STARTMD:
1244 		if (!pv->pv_started)
1245 			g_raid_md_promise_start(vol);
1246 		return (0);
1247 	}
1248 	return (-2);
1249 }
1250 
1251 static int
1252 g_raid_md_ctl_promise(struct g_raid_md_object *md,
1253     struct gctl_req *req)
1254 {
1255 	struct g_raid_softc *sc;
1256 	struct g_raid_volume *vol, *vol1;
1257 	struct g_raid_subdisk *sd;
1258 	struct g_raid_disk *disk, *disks[PROMISE_MAX_DISKS];
1259 	struct g_raid_md_promise_perdisk *pd;
1260 	struct g_raid_md_promise_pervolume *pv;
1261 	struct g_consumer *cp;
1262 	struct g_provider *pp;
1263 	char arg[16];
1264 	const char *nodename, *verb, *volname, *levelname, *diskname;
1265 	char *tmp;
1266 	int *nargs, *force;
1267 	off_t esize, offs[PROMISE_MAX_DISKS], size, sectorsize, strip;
1268 	intmax_t *sizearg, *striparg;
1269 	int numdisks, i, len, level, qual;
1270 	int error;
1271 
1272 	sc = md->mdo_softc;
1273 	verb = gctl_get_param(req, "verb", NULL);
1274 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1275 	error = 0;
1276 	if (strcmp(verb, "label") == 0) {
1277 
1278 		if (*nargs < 4) {
1279 			gctl_error(req, "Invalid number of arguments.");
1280 			return (-1);
1281 		}
1282 		volname = gctl_get_asciiparam(req, "arg1");
1283 		if (volname == NULL) {
1284 			gctl_error(req, "No volume name.");
1285 			return (-2);
1286 		}
1287 		levelname = gctl_get_asciiparam(req, "arg2");
1288 		if (levelname == NULL) {
1289 			gctl_error(req, "No RAID level.");
1290 			return (-3);
1291 		}
1292 		if (strcasecmp(levelname, "RAID5") == 0)
1293 			levelname = "RAID5-LA";
1294 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1295 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1296 			return (-4);
1297 		}
1298 		numdisks = *nargs - 3;
1299 		force = gctl_get_paraml(req, "force", sizeof(*force));
1300 		if (!g_raid_md_promise_supported(level, qual, numdisks,
1301 		    force ? *force : 0)) {
1302 			gctl_error(req, "Unsupported RAID level "
1303 			    "(0x%02x/0x%02x), or number of disks (%d).",
1304 			    level, qual, numdisks);
1305 			return (-5);
1306 		}
1307 
1308 		/* Search for disks, connect them and probe. */
1309 		size = INT64_MAX;
1310 		sectorsize = 0;
1311 		bzero(disks, sizeof(disks));
1312 		bzero(offs, sizeof(offs));
1313 		for (i = 0; i < numdisks; i++) {
1314 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1315 			diskname = gctl_get_asciiparam(req, arg);
1316 			if (diskname == NULL) {
1317 				gctl_error(req, "No disk name (%s).", arg);
1318 				error = -6;
1319 				break;
1320 			}
1321 			if (strcmp(diskname, "NONE") == 0)
1322 				continue;
1323 
1324 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1325 				if (disk->d_consumer != NULL &&
1326 				    disk->d_consumer->provider != NULL &&
1327 				    strcmp(disk->d_consumer->provider->name,
1328 				     diskname) == 0)
1329 					break;
1330 			}
1331 			if (disk != NULL) {
1332 				if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
1333 					gctl_error(req, "Disk '%s' is in a "
1334 					    "wrong state (%s).", diskname,
1335 					    g_raid_disk_state2str(disk->d_state));
1336 					error = -7;
1337 					break;
1338 				}
1339 				pd = disk->d_md_data;
1340 				if (pd->pd_subdisks >= PROMISE_MAX_SUBDISKS) {
1341 					gctl_error(req, "Disk '%s' already "
1342 					    "used by %d volumes.",
1343 					    diskname, pd->pd_subdisks);
1344 					error = -7;
1345 					break;
1346 				}
1347 				pp = disk->d_consumer->provider;
1348 				disks[i] = disk;
1349 				promise_meta_unused_range(pd->pd_meta,
1350 				    pd->pd_subdisks,
1351 				    pp->mediasize / pp->sectorsize,
1352 				    &offs[i], &esize);
1353 				size = MIN(size, (off_t)esize * pp->sectorsize);
1354 				sectorsize = MAX(sectorsize, pp->sectorsize);
1355 				continue;
1356 			}
1357 
1358 			g_topology_lock();
1359 			cp = g_raid_open_consumer(sc, diskname);
1360 			if (cp == NULL) {
1361 				gctl_error(req, "Can't open disk '%s'.",
1362 				    diskname);
1363 				g_topology_unlock();
1364 				error = -8;
1365 				break;
1366 			}
1367 			pp = cp->provider;
1368 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1369 			disk = g_raid_create_disk(sc);
1370 			disk->d_md_data = (void *)pd;
1371 			disk->d_consumer = cp;
1372 			disks[i] = disk;
1373 			cp->private = disk;
1374 			g_topology_unlock();
1375 
1376 			g_raid_get_disk_info(disk);
1377 
1378 			/* Reserve some space for metadata. */
1379 			size = MIN(size, pp->mediasize - 131072llu * pp->sectorsize);
1380 			sectorsize = MAX(sectorsize, pp->sectorsize);
1381 		}
1382 		if (error != 0) {
1383 			for (i = 0; i < numdisks; i++) {
1384 				if (disks[i] != NULL &&
1385 				    disks[i]->d_state == G_RAID_DISK_S_NONE)
1386 					g_raid_destroy_disk(disks[i]);
1387 			}
1388 			return (error);
1389 		}
1390 
1391 		if (sectorsize <= 0) {
1392 			gctl_error(req, "Can't get sector size.");
1393 			return (-8);
1394 		}
1395 
1396 		/* Handle size argument. */
1397 		len = sizeof(*sizearg);
1398 		sizearg = gctl_get_param(req, "size", &len);
1399 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1400 		    *sizearg > 0) {
1401 			if (*sizearg > size) {
1402 				gctl_error(req, "Size too big %lld > %lld.",
1403 				    (long long)*sizearg, (long long)size);
1404 				return (-9);
1405 			}
1406 			size = *sizearg;
1407 		}
1408 
1409 		/* Handle strip argument. */
1410 		strip = 131072;
1411 		len = sizeof(*striparg);
1412 		striparg = gctl_get_param(req, "strip", &len);
1413 		if (striparg != NULL && len == sizeof(*striparg) &&
1414 		    *striparg > 0) {
1415 			if (*striparg < sectorsize) {
1416 				gctl_error(req, "Strip size too small.");
1417 				return (-10);
1418 			}
1419 			if (*striparg % sectorsize != 0) {
1420 				gctl_error(req, "Incorrect strip size.");
1421 				return (-11);
1422 			}
1423 			strip = *striparg;
1424 		}
1425 
1426 		/* Round size down to strip or sector. */
1427 		if (level == G_RAID_VOLUME_RL_RAID1 ||
1428 		    level == G_RAID_VOLUME_RL_SINGLE ||
1429 		    level == G_RAID_VOLUME_RL_CONCAT)
1430 			size -= (size % sectorsize);
1431 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1432 		    (numdisks & 1) != 0)
1433 			size -= (size % (2 * strip));
1434 		else
1435 			size -= (size % strip);
1436 		if (size <= 0) {
1437 			gctl_error(req, "Size too small.");
1438 			return (-13);
1439 		}
1440 
1441 		/* We have all we need, create things: volume, ... */
1442 		pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1443 		arc4rand(&pv->pv_id, sizeof(pv->pv_id), 0);
1444 		pv->pv_generation = 0;
1445 		pv->pv_started = 1;
1446 		vol = g_raid_create_volume(sc, volname, -1);
1447 		vol->v_md_data = pv;
1448 		vol->v_raid_level = level;
1449 		vol->v_raid_level_qualifier = qual;
1450 		vol->v_strip_size = strip;
1451 		vol->v_disks_count = numdisks;
1452 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1453 		    level == G_RAID_VOLUME_RL_CONCAT ||
1454 		    level == G_RAID_VOLUME_RL_SINGLE)
1455 			vol->v_mediasize = size * numdisks;
1456 		else if (level == G_RAID_VOLUME_RL_RAID1)
1457 			vol->v_mediasize = size;
1458 		else if (level == G_RAID_VOLUME_RL_RAID3 ||
1459 		    level == G_RAID_VOLUME_RL_RAID5)
1460 			vol->v_mediasize = size * (numdisks - 1);
1461 		else { /* RAID1E */
1462 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1463 			    strip;
1464 		}
1465 		vol->v_sectorsize = sectorsize;
1466 		g_raid_start_volume(vol);
1467 
1468 		/* , and subdisks. */
1469 		for (i = 0; i < numdisks; i++) {
1470 			disk = disks[i];
1471 			sd = &vol->v_subdisks[i];
1472 			sd->sd_disk = disk;
1473 			sd->sd_offset = (off_t)offs[i] * 512;
1474 			sd->sd_size = size;
1475 			if (disk == NULL)
1476 				continue;
1477 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1478 			g_raid_change_disk_state(disk,
1479 			    G_RAID_DISK_S_ACTIVE);
1480 			g_raid_change_subdisk_state(sd,
1481 			    G_RAID_SUBDISK_S_ACTIVE);
1482 			g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1483 			    G_RAID_EVENT_SUBDISK);
1484 		}
1485 
1486 		/* Write metadata based on created entities. */
1487 		G_RAID_DEBUG1(0, sc, "Array started.");
1488 		g_raid_md_write_promise(md, vol, NULL, NULL);
1489 
1490 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1491 		g_raid_md_promise_refill(sc);
1492 
1493 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1494 		    G_RAID_EVENT_VOLUME);
1495 		return (0);
1496 	}
1497 	if (strcmp(verb, "add") == 0) {
1498 
1499 		gctl_error(req, "`add` command is not applicable, "
1500 		    "use `label` instead.");
1501 		return (-99);
1502 	}
1503 	if (strcmp(verb, "delete") == 0) {
1504 
1505 		nodename = gctl_get_asciiparam(req, "arg0");
1506 		if (nodename != NULL && strcasecmp(sc->sc_name, nodename) != 0)
1507 			nodename = NULL;
1508 
1509 		/* Full node destruction. */
1510 		if (*nargs == 1 && nodename != NULL) {
1511 			/* Check if some volume is still open. */
1512 			force = gctl_get_paraml(req, "force", sizeof(*force));
1513 			if (force != NULL && *force == 0 &&
1514 			    g_raid_nopens(sc) != 0) {
1515 				gctl_error(req, "Some volume is still open.");
1516 				return (-4);
1517 			}
1518 
1519 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1520 				if (disk->d_consumer)
1521 					promise_meta_erase(disk->d_consumer);
1522 			}
1523 			g_raid_destroy_node(sc, 0);
1524 			return (0);
1525 		}
1526 
1527 		/* Destroy specified volume. If it was last - all node. */
1528 		if (*nargs > 2) {
1529 			gctl_error(req, "Invalid number of arguments.");
1530 			return (-1);
1531 		}
1532 		volname = gctl_get_asciiparam(req,
1533 		    nodename != NULL ? "arg1" : "arg0");
1534 		if (volname == NULL) {
1535 			gctl_error(req, "No volume name.");
1536 			return (-2);
1537 		}
1538 
1539 		/* Search for volume. */
1540 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1541 			if (strcmp(vol->v_name, volname) == 0)
1542 				break;
1543 			pp = vol->v_provider;
1544 			if (pp == NULL)
1545 				continue;
1546 			if (strcmp(pp->name, volname) == 0)
1547 				break;
1548 			if (strncmp(pp->name, "raid/", 5) == 0 &&
1549 			    strcmp(pp->name + 5, volname) == 0)
1550 				break;
1551 		}
1552 		if (vol == NULL) {
1553 			i = strtol(volname, &tmp, 10);
1554 			if (verb != volname && tmp[0] == 0) {
1555 				TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1556 					if (vol->v_global_id == i)
1557 						break;
1558 				}
1559 			}
1560 		}
1561 		if (vol == NULL) {
1562 			gctl_error(req, "Volume '%s' not found.", volname);
1563 			return (-3);
1564 		}
1565 
1566 		/* Check if volume is still open. */
1567 		force = gctl_get_paraml(req, "force", sizeof(*force));
1568 		if (force != NULL && *force == 0 &&
1569 		    vol->v_provider_open != 0) {
1570 			gctl_error(req, "Volume is still open.");
1571 			return (-4);
1572 		}
1573 
1574 		/* Destroy volume and potentially node. */
1575 		i = 0;
1576 		TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
1577 			i++;
1578 		if (i >= 2) {
1579 			g_raid_destroy_volume(vol);
1580 			g_raid_md_promise_purge_disks(sc);
1581 			g_raid_md_write_promise(md, NULL, NULL, NULL);
1582 		} else {
1583 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1584 				if (disk->d_consumer)
1585 					promise_meta_erase(disk->d_consumer);
1586 			}
1587 			g_raid_destroy_node(sc, 0);
1588 		}
1589 		return (0);
1590 	}
1591 	if (strcmp(verb, "remove") == 0 ||
1592 	    strcmp(verb, "fail") == 0) {
1593 		if (*nargs < 2) {
1594 			gctl_error(req, "Invalid number of arguments.");
1595 			return (-1);
1596 		}
1597 		for (i = 1; i < *nargs; i++) {
1598 			snprintf(arg, sizeof(arg), "arg%d", i);
1599 			diskname = gctl_get_asciiparam(req, arg);
1600 			if (diskname == NULL) {
1601 				gctl_error(req, "No disk name (%s).", arg);
1602 				error = -2;
1603 				break;
1604 			}
1605 			if (strncmp(diskname, "/dev/", 5) == 0)
1606 				diskname += 5;
1607 
1608 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1609 				if (disk->d_consumer != NULL &&
1610 				    disk->d_consumer->provider != NULL &&
1611 				    strcmp(disk->d_consumer->provider->name,
1612 				     diskname) == 0)
1613 					break;
1614 			}
1615 			if (disk == NULL) {
1616 				gctl_error(req, "Disk '%s' not found.",
1617 				    diskname);
1618 				error = -3;
1619 				break;
1620 			}
1621 
1622 			if (strcmp(verb, "fail") == 0) {
1623 				g_raid_md_fail_disk_promise(md, NULL, disk);
1624 				continue;
1625 			}
1626 
1627 			/* Erase metadata on deleting disk and destroy it. */
1628 			promise_meta_erase(disk->d_consumer);
1629 			g_raid_destroy_disk(disk);
1630 		}
1631 		g_raid_md_promise_purge_volumes(sc);
1632 
1633 		/* Write updated metadata to remaining disks. */
1634 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1635 
1636 		/* Check if anything left. */
1637 		if (g_raid_ndisks(sc, -1) == 0)
1638 			g_raid_destroy_node(sc, 0);
1639 		else
1640 			g_raid_md_promise_refill(sc);
1641 		return (error);
1642 	}
1643 	if (strcmp(verb, "insert") == 0) {
1644 		if (*nargs < 2) {
1645 			gctl_error(req, "Invalid number of arguments.");
1646 			return (-1);
1647 		}
1648 		for (i = 1; i < *nargs; i++) {
1649 			/* Get disk name. */
1650 			snprintf(arg, sizeof(arg), "arg%d", i);
1651 			diskname = gctl_get_asciiparam(req, arg);
1652 			if (diskname == NULL) {
1653 				gctl_error(req, "No disk name (%s).", arg);
1654 				error = -3;
1655 				break;
1656 			}
1657 
1658 			/* Try to find provider with specified name. */
1659 			g_topology_lock();
1660 			cp = g_raid_open_consumer(sc, diskname);
1661 			if (cp == NULL) {
1662 				gctl_error(req, "Can't open disk '%s'.",
1663 				    diskname);
1664 				g_topology_unlock();
1665 				error = -4;
1666 				break;
1667 			}
1668 			pp = cp->provider;
1669 			g_topology_unlock();
1670 
1671 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1672 
1673 			disk = g_raid_create_disk(sc);
1674 			disk->d_consumer = cp;
1675 			disk->d_md_data = (void *)pd;
1676 			cp->private = disk;
1677 
1678 			g_raid_get_disk_info(disk);
1679 
1680 			/* Welcome the "new" disk. */
1681 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1682 			promise_meta_write_spare(cp);
1683 			g_raid_md_promise_refill(sc);
1684 		}
1685 		return (error);
1686 	}
1687 	return (-100);
1688 }
1689 
1690 static int
1691 g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1692     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1693 {
1694 	struct g_raid_softc *sc;
1695 	struct g_raid_volume *vol;
1696 	struct g_raid_subdisk *sd;
1697 	struct g_raid_disk *disk;
1698 	struct g_raid_md_promise_perdisk *pd;
1699 	struct g_raid_md_promise_pervolume *pv;
1700 	struct promise_raid_conf *meta;
1701 	off_t rebuild_lba64;
1702 	int i, j, pos, rebuild;
1703 
1704 	sc = md->mdo_softc;
1705 
1706 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1707 		return (0);
1708 
1709 	/* Generate new per-volume metadata for affected volumes. */
1710 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1711 		if (vol->v_stopping)
1712 			continue;
1713 
1714 		/* Skip volumes not related to specified targets. */
1715 		if (tvol != NULL && vol != tvol)
1716 			continue;
1717 		if (tsd != NULL && vol != tsd->sd_volume)
1718 			continue;
1719 		if (tdisk != NULL) {
1720 			for (i = 0; i < vol->v_disks_count; i++) {
1721 				if (vol->v_subdisks[i].sd_disk == tdisk)
1722 					break;
1723 			}
1724 			if (i >= vol->v_disks_count)
1725 				continue;
1726 		}
1727 
1728 		pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1729 		pv->pv_generation++;
1730 
1731 		meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
1732 		if (pv->pv_meta != NULL)
1733 			memcpy(meta, pv->pv_meta, sizeof(*meta));
1734 		memcpy(meta->promise_id, PROMISE_MAGIC,
1735 		    sizeof(PROMISE_MAGIC) - 1);
1736 		meta->dummy_0 = 0x00020000;
1737 		meta->integrity = PROMISE_I_VALID;
1738 
1739 		meta->generation = pv->pv_generation;
1740 		meta->status = PROMISE_S_VALID | PROMISE_S_ONLINE |
1741 		    PROMISE_S_INITED | PROMISE_S_READY;
1742 		if (vol->v_state <= G_RAID_VOLUME_S_DEGRADED)
1743 			meta->status |= PROMISE_S_DEGRADED;
1744 		if (vol->v_dirty)
1745 			meta->status |= PROMISE_S_MARKED; /* XXX: INVENTED! */
1746 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0 ||
1747 		    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1748 			meta->type = PROMISE_T_RAID0;
1749 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1750 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1751 			meta->type = PROMISE_T_RAID1;
1752 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3)
1753 			meta->type = PROMISE_T_RAID3;
1754 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1755 			meta->type = PROMISE_T_RAID5;
1756 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT)
1757 			meta->type = PROMISE_T_SPAN;
1758 		else
1759 			meta->type = PROMISE_T_JBOD;
1760 		meta->total_disks = vol->v_disks_count;
1761 		meta->stripe_shift = ffs(vol->v_strip_size / 1024);
1762 		meta->array_width = vol->v_disks_count;
1763 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1764 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1765 			meta->array_width /= 2;
1766 		meta->array_number = vol->v_global_id;
1767 		meta->total_sectors = vol->v_mediasize / 512;
1768 		meta->total_sectors_high = (vol->v_mediasize / 512) >> 32;
1769 		meta->sector_size = vol->v_sectorsize / 512;
1770 		meta->cylinders = meta->total_sectors / (255 * 63) - 1;
1771 		meta->heads = 254;
1772 		meta->sectors = 63;
1773 		meta->volume_id = pv->pv_id;
1774 		rebuild_lba64 = UINT64_MAX;
1775 		rebuild = 0;
1776 		for (i = 0; i < vol->v_disks_count; i++) {
1777 			sd = &vol->v_subdisks[i];
1778 			/* For RAID0+1 we need to translate order. */
1779 			pos = promise_meta_translate_disk(vol, i);
1780 			meta->disks[pos].flags = PROMISE_F_VALID |
1781 			    PROMISE_F_ASSIGNED;
1782 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE) {
1783 				meta->disks[pos].flags |= 0;
1784 			} else if (sd->sd_state == G_RAID_SUBDISK_S_FAILED) {
1785 				meta->disks[pos].flags |=
1786 				    PROMISE_F_DOWN | PROMISE_F_REDIR;
1787 			} else if (sd->sd_state <= G_RAID_SUBDISK_S_REBUILD) {
1788 				meta->disks[pos].flags |=
1789 				    PROMISE_F_ONLINE | PROMISE_F_REDIR;
1790 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1791 					rebuild_lba64 = MIN(rebuild_lba64,
1792 					    sd->sd_rebuild_pos / 512);
1793 				} else
1794 					rebuild_lba64 = 0;
1795 				rebuild = 1;
1796 			} else {
1797 				meta->disks[pos].flags |= PROMISE_F_ONLINE;
1798 				if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) {
1799 					meta->status |= PROMISE_S_MARKED;
1800 					if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
1801 						rebuild_lba64 = MIN(rebuild_lba64,
1802 						    sd->sd_rebuild_pos / 512);
1803 					} else
1804 						rebuild_lba64 = 0;
1805 				}
1806 			}
1807 			if (pv->pv_meta != NULL) {
1808 				meta->disks[pos].id = pv->pv_meta->disks[pos].id;
1809 			} else {
1810 				meta->disks[pos].number = i * 2;
1811 				arc4rand(&meta->disks[pos].id,
1812 				    sizeof(meta->disks[pos].id), 0);
1813 			}
1814 		}
1815 		promise_meta_put_name(meta, vol->v_name);
1816 
1817 		/* Try to mimic AMD BIOS rebuild/resync behavior. */
1818 		if (rebuild_lba64 != UINT64_MAX) {
1819 			if (rebuild)
1820 				meta->magic_3 = 0x03040010UL; /* Rebuild? */
1821 			else
1822 				meta->magic_3 = 0x03040008UL; /* Resync? */
1823 			/* Translate from per-disk to per-volume LBA. */
1824 			if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1825 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1826 				rebuild_lba64 *= meta->array_width;
1827 			} else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1828 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) {
1829 				rebuild_lba64 *= meta->array_width - 1;
1830 			} else
1831 				rebuild_lba64 = 0;
1832 		} else
1833 			meta->magic_3 = 0x03000000UL;
1834 		meta->rebuild_lba64 = rebuild_lba64;
1835 		meta->magic_4 = 0x04010101UL;
1836 
1837 		/* Replace per-volume metadata with new. */
1838 		if (pv->pv_meta != NULL)
1839 			free(pv->pv_meta, M_MD_PROMISE);
1840 		pv->pv_meta = meta;
1841 
1842 		/* Copy new metadata to the disks, adding or replacing old. */
1843 		for (i = 0; i < vol->v_disks_count; i++) {
1844 			sd = &vol->v_subdisks[i];
1845 			disk = sd->sd_disk;
1846 			if (disk == NULL)
1847 				continue;
1848 			/* For RAID0+1 we need to translate order. */
1849 			pos = promise_meta_translate_disk(vol, i);
1850 			pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1851 			for (j = 0; j < pd->pd_subdisks; j++) {
1852 				if (pd->pd_meta[j]->volume_id == meta->volume_id)
1853 					break;
1854 			}
1855 			if (j == pd->pd_subdisks)
1856 				pd->pd_subdisks++;
1857 			if (pd->pd_meta[j] != NULL)
1858 				free(pd->pd_meta[j], M_MD_PROMISE);
1859 			pd->pd_meta[j] = promise_meta_copy(meta);
1860 			pd->pd_meta[j]->disk = meta->disks[pos];
1861 			pd->pd_meta[j]->disk.number = pos;
1862 			pd->pd_meta[j]->disk_offset_high =
1863 			    (sd->sd_offset / 512) >> 32;
1864 			pd->pd_meta[j]->disk_offset = sd->sd_offset / 512;
1865 			pd->pd_meta[j]->disk_sectors_high =
1866 			    (sd->sd_size / 512) >> 32;
1867 			pd->pd_meta[j]->disk_sectors = sd->sd_size / 512;
1868 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1869 				pd->pd_meta[j]->disk_rebuild_high =
1870 				    (sd->sd_rebuild_pos / 512) >> 32;
1871 				pd->pd_meta[j]->disk_rebuild =
1872 				    sd->sd_rebuild_pos / 512;
1873 			} else if (sd->sd_state < G_RAID_SUBDISK_S_REBUILD) {
1874 				pd->pd_meta[j]->disk_rebuild_high = 0;
1875 				pd->pd_meta[j]->disk_rebuild = 0;
1876 			} else {
1877 				pd->pd_meta[j]->disk_rebuild_high = UINT32_MAX;
1878 				pd->pd_meta[j]->disk_rebuild = UINT32_MAX;
1879 			}
1880 			pd->pd_updated = 1;
1881 		}
1882 	}
1883 
1884 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1885 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1886 		if (disk->d_state != G_RAID_DISK_S_ACTIVE)
1887 			continue;
1888 		if (!pd->pd_updated)
1889 			continue;
1890 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1891 		    g_raid_get_diskname(disk));
1892 		for (i = 0; i < pd->pd_subdisks; i++)
1893 			g_raid_md_promise_print(pd->pd_meta[i]);
1894 		promise_meta_write(disk->d_consumer,
1895 		    pd->pd_meta, pd->pd_subdisks);
1896 		pd->pd_updated = 0;
1897 	}
1898 
1899 	return (0);
1900 }
1901 
1902 static int
1903 g_raid_md_fail_disk_promise(struct g_raid_md_object *md,
1904     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1905 {
1906 	struct g_raid_softc *sc;
1907 	struct g_raid_md_promise_perdisk *pd;
1908 	struct g_raid_subdisk *sd;
1909 	int i, pos;
1910 
1911 	sc = md->mdo_softc;
1912 	pd = (struct g_raid_md_promise_perdisk *)tdisk->d_md_data;
1913 
1914 	/* We can't fail disk that is not a part of array now. */
1915 	if (tdisk->d_state != G_RAID_DISK_S_ACTIVE)
1916 		return (-1);
1917 
1918 	/*
1919 	 * Mark disk as failed in metadata and try to write that metadata
1920 	 * to the disk itself to prevent it's later resurrection as STALE.
1921 	 */
1922 	if (pd->pd_subdisks > 0 && tdisk->d_consumer != NULL)
1923 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1924 		    g_raid_get_diskname(tdisk));
1925 	for (i = 0; i < pd->pd_subdisks; i++) {
1926 		pd->pd_meta[i]->disk.flags |=
1927 		    PROMISE_F_DOWN | PROMISE_F_REDIR;
1928 		pos = pd->pd_meta[i]->disk.number;
1929 		if (pos >= 0 && pos < PROMISE_MAX_DISKS) {
1930 			pd->pd_meta[i]->disks[pos].flags |=
1931 			    PROMISE_F_DOWN | PROMISE_F_REDIR;
1932 		}
1933 		g_raid_md_promise_print(pd->pd_meta[i]);
1934 	}
1935 	if (tdisk->d_consumer != NULL)
1936 		promise_meta_write(tdisk->d_consumer,
1937 		    pd->pd_meta, pd->pd_subdisks);
1938 
1939 	/* Change states. */
1940 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1941 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1942 		g_raid_change_subdisk_state(sd,
1943 		    G_RAID_SUBDISK_S_FAILED);
1944 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1945 		    G_RAID_EVENT_SUBDISK);
1946 	}
1947 
1948 	/* Write updated metadata to remaining disks. */
1949 	g_raid_md_write_promise(md, NULL, NULL, tdisk);
1950 
1951 	g_raid_md_promise_refill(sc);
1952 	return (0);
1953 }
1954 
1955 static int
1956 g_raid_md_free_disk_promise(struct g_raid_md_object *md,
1957     struct g_raid_disk *disk)
1958 {
1959 	struct g_raid_md_promise_perdisk *pd;
1960 	int i;
1961 
1962 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1963 	for (i = 0; i < pd->pd_subdisks; i++) {
1964 		if (pd->pd_meta[i] != NULL) {
1965 			free(pd->pd_meta[i], M_MD_PROMISE);
1966 			pd->pd_meta[i] = NULL;
1967 		}
1968 	}
1969 	free(pd, M_MD_PROMISE);
1970 	disk->d_md_data = NULL;
1971 	return (0);
1972 }
1973 
1974 static int
1975 g_raid_md_free_volume_promise(struct g_raid_md_object *md,
1976     struct g_raid_volume *vol)
1977 {
1978 	struct g_raid_md_promise_pervolume *pv;
1979 
1980 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1981 	if (pv && pv->pv_meta != NULL) {
1982 		free(pv->pv_meta, M_MD_PROMISE);
1983 		pv->pv_meta = NULL;
1984 	}
1985 	if (pv && !pv->pv_started) {
1986 		pv->pv_started = 1;
1987 		callout_stop(&pv->pv_start_co);
1988 	}
1989 	free(pv, M_MD_PROMISE);
1990 	vol->v_md_data = NULL;
1991 	return (0);
1992 }
1993 
1994 static int
1995 g_raid_md_free_promise(struct g_raid_md_object *md)
1996 {
1997 
1998 	return (0);
1999 }
2000 
2001 G_RAID_MD_DECLARE(promise, "Promise");
2002