xref: /freebsd/sys/geom/raid/md_promise.c (revision 2e3f49888ec8851bafb22011533217487764fdb0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
5  * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
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/geom_dbg.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 static void
190 g_raid_md_promise_print(struct promise_raid_conf *meta)
191 {
192 	int i;
193 
194 	if (g_raid_debug < 1)
195 		return;
196 
197 	printf("********* ATA Promise Metadata *********\n");
198 	printf("promise_id          <%.24s>\n", meta->promise_id);
199 	printf("disk                %02x %02x %02x %02x %016jx\n",
200 	    meta->disk.flags, meta->disk.number, meta->disk.channel,
201 	    meta->disk.device, meta->disk.id);
202 	printf("disk_offset         %u\n", meta->disk_offset);
203 	printf("disk_sectors        %u\n", meta->disk_sectors);
204 	printf("disk_rebuild        %u\n", meta->disk_rebuild);
205 	printf("generation          %u\n", meta->generation);
206 	printf("status              0x%02x\n", meta->status);
207 	printf("type                %u\n", meta->type);
208 	printf("total_disks         %u\n", meta->total_disks);
209 	printf("stripe_shift        %u\n", meta->stripe_shift);
210 	printf("array_width         %u\n", meta->array_width);
211 	printf("array_number        %u\n", meta->array_number);
212 	printf("total_sectors       %u\n", meta->total_sectors);
213 	printf("cylinders           %u\n", meta->cylinders);
214 	printf("heads               %u\n", meta->heads);
215 	printf("sectors             %u\n", meta->sectors);
216 	printf("volume_id           0x%016jx\n", meta->volume_id);
217 	printf("disks:\n");
218 	for (i = 0; i < PROMISE_MAX_DISKS; i++ ) {
219 		printf("                    %02x %02x %02x %02x %016jx\n",
220 		    meta->disks[i].flags, meta->disks[i].number,
221 		    meta->disks[i].channel, meta->disks[i].device,
222 		    meta->disks[i].id);
223 	}
224 	printf("name                <%.32s>\n", meta->name);
225 	printf("magic_3             0x%08x\n", meta->magic_3);
226 	printf("rebuild_lba64       %ju\n", meta->rebuild_lba64);
227 	printf("magic_4             0x%08x\n", meta->magic_4);
228 	printf("magic_5             0x%08x\n", meta->magic_5);
229 	printf("total_sectors_high  0x%08x\n", meta->total_sectors_high);
230 	printf("sector_size         %u\n", meta->sector_size);
231 	printf("backup_time         %d\n", meta->backup_time);
232 	printf("disk_offset_high    0x%08x\n", meta->disk_offset_high);
233 	printf("disk_sectors_high   0x%08x\n", meta->disk_sectors_high);
234 	printf("disk_rebuild_high   0x%08x\n", meta->disk_rebuild_high);
235 	printf("=================================================\n");
236 }
237 
238 static struct promise_raid_conf *
239 promise_meta_copy(struct promise_raid_conf *meta)
240 {
241 	struct promise_raid_conf *nmeta;
242 
243 	nmeta = malloc(sizeof(*nmeta), M_MD_PROMISE, M_WAITOK);
244 	memcpy(nmeta, meta, sizeof(*nmeta));
245 	return (nmeta);
246 }
247 
248 static int
249 promise_meta_find_disk(struct promise_raid_conf *meta, uint64_t id)
250 {
251 	int pos;
252 
253 	for (pos = 0; pos < meta->total_disks; pos++) {
254 		if (meta->disks[pos].id == id)
255 			return (pos);
256 	}
257 	return (-1);
258 }
259 
260 static int
261 promise_meta_unused_range(struct promise_raid_conf **metaarr, int nsd,
262     off_t sectors, off_t *off, off_t *size)
263 {
264 	off_t coff, csize, tmp;
265 	int i, j;
266 
267 	sectors -= 131072;
268 	*off = 0;
269 	*size = 0;
270 	coff = 0;
271 	csize = sectors;
272 	i = 0;
273 	while (1) {
274 		for (j = 0; j < nsd; j++) {
275 			tmp = ((off_t)metaarr[j]->disk_offset_high << 32) +
276 			    metaarr[j]->disk_offset;
277 			if (tmp >= coff)
278 				csize = MIN(csize, tmp - coff);
279 		}
280 		if (csize > *size) {
281 			*off = coff;
282 			*size = csize;
283 		}
284 		if (i >= nsd)
285 			break;
286 		coff = ((off_t)metaarr[i]->disk_offset_high << 32) +
287 		     metaarr[i]->disk_offset +
288 		    ((off_t)metaarr[i]->disk_sectors_high << 32) +
289 		     metaarr[i]->disk_sectors;
290 		csize = sectors - coff;
291 		i++;
292 	}
293 	return ((*size > 0) ? 1 : 0);
294 }
295 
296 static int
297 promise_meta_translate_disk(struct g_raid_volume *vol, int md_disk_pos)
298 {
299 	int disk_pos, width;
300 
301 	if (md_disk_pos >= 0 && vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
302 		width = vol->v_disks_count / 2;
303 		disk_pos = (md_disk_pos / width) +
304 		    (md_disk_pos % width) * width;
305 	} else
306 		disk_pos = md_disk_pos;
307 	return (disk_pos);
308 }
309 
310 static void
311 promise_meta_get_name(struct promise_raid_conf *meta, char *buf)
312 {
313 	int i;
314 
315 	strncpy(buf, meta->name, 32);
316 	buf[32] = 0;
317 	for (i = 31; i >= 0; i--) {
318 		if (buf[i] > 0x20)
319 			break;
320 		buf[i] = 0;
321 	}
322 }
323 
324 static void
325 promise_meta_put_name(struct promise_raid_conf *meta, char *buf)
326 {
327 
328 	memset(meta->name, 0x20, 32);
329 	memcpy(meta->name, buf, MIN(strlen(buf), 32));
330 }
331 
332 static int
333 promise_meta_read(struct g_consumer *cp, struct promise_raid_conf **metaarr)
334 {
335 	struct g_provider *pp;
336 	struct promise_raid_conf *meta;
337 	char *buf;
338 	int error, i, subdisks;
339 	uint32_t checksum, *ptr;
340 
341 	pp = cp->provider;
342 	subdisks = 0;
343 
344 	if (pp->sectorsize * 4 < sizeof(*meta))
345 		return (subdisks);
346 	if (pp->sectorsize * 4 > maxphys) {
347 		G_RAID_DEBUG(1, "%s: Blocksize is too big.", pp->name);
348 		return (subdisks);
349 	}
350 next:
351 	/* Read metadata block. */
352 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize *
353 	    (63 - subdisks * PROMISE_META_OFFSET),
354 	    pp->sectorsize * 4, &error);
355 	if (buf == NULL) {
356 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
357 		    pp->name, error);
358 		return (subdisks);
359 	}
360 	meta = (struct promise_raid_conf *)buf;
361 
362 	/* Check if this is an Promise RAID struct */
363 	if (strncmp(meta->promise_id, PROMISE_MAGIC, strlen(PROMISE_MAGIC)) &&
364 	    strncmp(meta->promise_id, FREEBSD_MAGIC, strlen(FREEBSD_MAGIC))) {
365 		if (subdisks == 0)
366 			G_RAID_DEBUG(1,
367 			    "Promise signature check failed on %s", pp->name);
368 		g_free(buf);
369 		return (subdisks);
370 	}
371 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK);
372 	memcpy(meta, buf, MIN(sizeof(*meta), pp->sectorsize * 4));
373 	g_free(buf);
374 
375 	/* Check metadata checksum. */
376 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
377 		checksum += *ptr++;
378 	if (checksum != meta->checksum) {
379 		G_RAID_DEBUG(1, "Promise checksum check failed on %s", pp->name);
380 		free(meta, M_MD_PROMISE);
381 		return (subdisks);
382 	}
383 
384 	if ((meta->integrity & PROMISE_I_VALID) == 0) {
385 		G_RAID_DEBUG(1, "Promise metadata is invalid on %s", pp->name);
386 		free(meta, M_MD_PROMISE);
387 		return (subdisks);
388 	}
389 
390 	if (meta->total_disks > PROMISE_MAX_DISKS) {
391 		G_RAID_DEBUG(1, "Wrong number of disks on %s (%d)",
392 		    pp->name, meta->total_disks);
393 		free(meta, M_MD_PROMISE);
394 		return (subdisks);
395 	}
396 
397 	/* Remove filler garbage from fields used in newer metadata. */
398 	if (meta->disk_offset_high == 0x8b8c8d8e &&
399 	    meta->disk_sectors_high == 0x8788898a &&
400 	    meta->disk_rebuild_high == 0x83848586) {
401 		meta->disk_offset_high = 0;
402 		meta->disk_sectors_high = 0;
403 		if (meta->disk_rebuild == UINT32_MAX)
404 			meta->disk_rebuild_high = UINT32_MAX;
405 		else
406 			meta->disk_rebuild_high = 0;
407 		if (meta->total_sectors_high == 0x15161718) {
408 			meta->total_sectors_high = 0;
409 			meta->backup_time = 0;
410 			if (meta->rebuild_lba64 == 0x2122232425262728)
411 				meta->rebuild_lba64 = UINT64_MAX;
412 		}
413 	}
414 	if (meta->sector_size < 1 || meta->sector_size > 8)
415 		meta->sector_size = 1;
416 
417 	/* Save this part and look for next. */
418 	*metaarr = meta;
419 	metaarr++;
420 	subdisks++;
421 	if (subdisks < PROMISE_MAX_SUBDISKS)
422 		goto next;
423 
424 	return (subdisks);
425 }
426 
427 static int
428 promise_meta_write(struct g_consumer *cp,
429     struct promise_raid_conf **metaarr, int nsd)
430 {
431 	struct g_provider *pp;
432 	struct promise_raid_conf *meta;
433 	char *buf;
434 	off_t off, size;
435 	int error, i, subdisk, fake;
436 	uint32_t checksum, *ptr;
437 
438 	pp = cp->provider;
439 	subdisk = 0;
440 	fake = 0;
441 next:
442 	buf = malloc(pp->sectorsize * 4, M_MD_PROMISE, M_WAITOK | M_ZERO);
443 	meta = NULL;
444 	if (subdisk < nsd) {
445 		meta = metaarr[subdisk];
446 	} else if (!fake && promise_meta_unused_range(metaarr, nsd,
447 	    cp->provider->mediasize / cp->provider->sectorsize,
448 	    &off, &size)) {
449 		/* Optionally add record for unused space. */
450 		meta = (struct promise_raid_conf *)buf;
451 		memcpy(&meta->promise_id[0], PROMISE_MAGIC,
452 		    sizeof(PROMISE_MAGIC) - 1);
453 		meta->dummy_0 = 0x00020000;
454 		meta->integrity = PROMISE_I_VALID;
455 		meta->disk.flags = PROMISE_F_ONLINE | PROMISE_F_VALID;
456 		meta->disk.number = 0xff;
457 		arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
458 		meta->disk_offset_high = off >> 32;
459 		meta->disk_offset = (uint32_t)off;
460 		meta->disk_sectors_high = size >> 32;
461 		meta->disk_sectors = (uint32_t)size;
462 		meta->disk_rebuild_high = UINT32_MAX;
463 		meta->disk_rebuild = UINT32_MAX;
464 		fake = 1;
465 	}
466 	if (meta != NULL) {
467 		/* Recalculate checksum for case if metadata were changed. */
468 		meta->checksum = 0;
469 		for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
470 			checksum += *ptr++;
471 		meta->checksum = checksum;
472 		memcpy(buf, meta, MIN(pp->sectorsize * 4, sizeof(*meta)));
473 	}
474 	error = g_write_data(cp, pp->mediasize - pp->sectorsize *
475 	    (63 - subdisk * PROMISE_META_OFFSET),
476 	    buf, pp->sectorsize * 4);
477 	if (error != 0) {
478 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
479 		    pp->name, error);
480 	}
481 	free(buf, M_MD_PROMISE);
482 
483 	subdisk++;
484 	if (subdisk < PROMISE_MAX_SUBDISKS)
485 		goto next;
486 
487 	return (error);
488 }
489 
490 static int
491 promise_meta_erase(struct g_consumer *cp)
492 {
493 	struct g_provider *pp;
494 	char *buf;
495 	int error, subdisk;
496 
497 	pp = cp->provider;
498 	buf = malloc(4 * pp->sectorsize, M_MD_PROMISE, M_WAITOK | M_ZERO);
499 	for (subdisk = 0; subdisk < PROMISE_MAX_SUBDISKS; subdisk++) {
500 		error = g_write_data(cp, pp->mediasize - pp->sectorsize *
501 		    (63 - subdisk * PROMISE_META_OFFSET),
502 		    buf, 4 * pp->sectorsize);
503 		if (error != 0) {
504 			G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
505 			    pp->name, error);
506 		}
507 	}
508 	free(buf, M_MD_PROMISE);
509 	return (error);
510 }
511 
512 static int
513 promise_meta_write_spare(struct g_consumer *cp)
514 {
515 	struct promise_raid_conf *meta;
516 	off_t tmp;
517 	int error;
518 
519 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
520 	memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC) - 1);
521 	meta->dummy_0 = 0x00020000;
522 	meta->integrity = PROMISE_I_VALID;
523 	meta->disk.flags = PROMISE_F_SPARE | PROMISE_F_ONLINE | PROMISE_F_VALID;
524 	meta->disk.number = 0xff;
525 	arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
526 	tmp = cp->provider->mediasize / cp->provider->sectorsize - 131072;
527 	meta->disk_sectors_high = tmp >> 32;
528 	meta->disk_sectors = (uint32_t)tmp;
529 	meta->disk_rebuild_high = UINT32_MAX;
530 	meta->disk_rebuild = UINT32_MAX;
531 	error = promise_meta_write(cp, &meta, 1);
532 	free(meta, M_MD_PROMISE);
533 	return (error);
534 }
535 
536 static struct g_raid_volume *
537 g_raid_md_promise_get_volume(struct g_raid_softc *sc, uint64_t id)
538 {
539 	struct g_raid_volume	*vol;
540 	struct g_raid_md_promise_pervolume *pv;
541 
542 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
543 		pv = vol->v_md_data;
544 		if (pv->pv_id == id)
545 			break;
546 	}
547 	return (vol);
548 }
549 
550 static int
551 g_raid_md_promise_purge_volumes(struct g_raid_softc *sc)
552 {
553 	struct g_raid_volume	*vol, *tvol;
554 	struct g_raid_md_promise_pervolume *pv;
555 	int i, res;
556 
557 	res = 0;
558 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tvol) {
559 		pv = vol->v_md_data;
560 		if (!pv->pv_started || vol->v_stopping)
561 			continue;
562 		for (i = 0; i < vol->v_disks_count; i++) {
563 			if (vol->v_subdisks[i].sd_state != G_RAID_SUBDISK_S_NONE)
564 				break;
565 		}
566 		if (i >= vol->v_disks_count) {
567 			g_raid_destroy_volume(vol);
568 			res = 1;
569 		}
570 	}
571 	return (res);
572 }
573 
574 static int
575 g_raid_md_promise_purge_disks(struct g_raid_softc *sc)
576 {
577 	struct g_raid_disk	*disk, *tdisk;
578 	struct g_raid_volume	*vol;
579 	struct g_raid_md_promise_perdisk *pd;
580 	int i, j, res;
581 
582 	res = 0;
583 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) {
584 		if (disk->d_state == G_RAID_DISK_S_SPARE)
585 			continue;
586 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
587 
588 		/* Scan for deleted volumes. */
589 		for (i = 0; i < pd->pd_subdisks; ) {
590 			vol = g_raid_md_promise_get_volume(sc,
591 			    pd->pd_meta[i]->volume_id);
592 			if (vol != NULL && !vol->v_stopping) {
593 				i++;
594 				continue;
595 			}
596 			free(pd->pd_meta[i], M_MD_PROMISE);
597 			for (j = i; j < pd->pd_subdisks - 1; j++)
598 				pd->pd_meta[j] = pd->pd_meta[j + 1];
599 			pd->pd_meta[pd->pd_subdisks - 1] = NULL;
600 			pd->pd_subdisks--;
601 			pd->pd_updated = 1;
602 		}
603 
604 		/* If there is no metadata left - erase and delete disk. */
605 		if (pd->pd_subdisks == 0) {
606 			promise_meta_erase(disk->d_consumer);
607 			g_raid_destroy_disk(disk);
608 			res = 1;
609 		}
610 	}
611 	return (res);
612 }
613 
614 static int
615 g_raid_md_promise_supported(int level, int qual, int disks, int force)
616 {
617 
618 	if (disks > PROMISE_MAX_DISKS)
619 		return (0);
620 	switch (level) {
621 	case G_RAID_VOLUME_RL_RAID0:
622 		if (disks < 1)
623 			return (0);
624 		if (!force && disks < 2)
625 			return (0);
626 		break;
627 	case G_RAID_VOLUME_RL_RAID1:
628 		if (disks < 1)
629 			return (0);
630 		if (!force && (disks != 2))
631 			return (0);
632 		break;
633 	case G_RAID_VOLUME_RL_RAID1E:
634 		if (disks < 2)
635 			return (0);
636 		if (disks % 2 != 0)
637 			return (0);
638 		if (!force && (disks != 4))
639 			return (0);
640 		break;
641 	case G_RAID_VOLUME_RL_SINGLE:
642 		if (disks != 1)
643 			return (0);
644 		break;
645 	case G_RAID_VOLUME_RL_CONCAT:
646 		if (disks < 2)
647 			return (0);
648 		break;
649 	case G_RAID_VOLUME_RL_RAID5:
650 		if (disks < 3)
651 			return (0);
652 		if (qual != G_RAID_VOLUME_RLQ_R5LA)
653 			return (0);
654 		break;
655 	default:
656 		return (0);
657 	}
658 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
659 		return (0);
660 	return (1);
661 }
662 
663 static int
664 g_raid_md_promise_start_disk(struct g_raid_disk *disk, int sdn,
665     struct g_raid_volume *vol)
666 {
667 	struct g_raid_softc *sc;
668 	struct g_raid_subdisk *sd;
669 	struct g_raid_md_promise_perdisk *pd;
670 	struct g_raid_md_promise_pervolume *pv;
671 	struct promise_raid_conf *meta;
672 	off_t eoff, esize, size;
673 	int disk_pos, md_disk_pos, i, resurrection = 0;
674 
675 	sc = disk->d_softc;
676 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
677 
678 	pv = vol->v_md_data;
679 	meta = pv->pv_meta;
680 
681 	if (sdn >= 0) {
682 		/* Find disk position in metadata by its serial. */
683 		md_disk_pos = promise_meta_find_disk(meta, pd->pd_meta[sdn]->disk.id);
684 		/* For RAID0+1 we need to translate order. */
685 		disk_pos = promise_meta_translate_disk(vol, md_disk_pos);
686 	} else {
687 		md_disk_pos = -1;
688 		disk_pos = -1;
689 	}
690 	if (disk_pos < 0) {
691 		G_RAID_DEBUG1(1, sc, "Disk %s is not part of the volume %s",
692 		    g_raid_get_diskname(disk), vol->v_name);
693 		/* Failed stale disk is useless for us. */
694 		if (sdn >= 0 &&
695 		    pd->pd_meta[sdn]->disk.flags & PROMISE_F_DOWN) {
696 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
697 			return (0);
698 		}
699 		/* If we were given specific metadata subdisk - erase it. */
700 		if (sdn >= 0) {
701 			free(pd->pd_meta[sdn], M_MD_PROMISE);
702 			for (i = sdn; i < pd->pd_subdisks - 1; i++)
703 				pd->pd_meta[i] = pd->pd_meta[i + 1];
704 			pd->pd_meta[pd->pd_subdisks - 1] = NULL;
705 			pd->pd_subdisks--;
706 		}
707 		/* If we are in the start process, that's all for now. */
708 		if (!pv->pv_started)
709 			goto nofit;
710 		/*
711 		 * If we have already started - try to get use of the disk.
712 		 * Try to replace OFFLINE disks first, then FAILED.
713 		 */
714 		promise_meta_unused_range(pd->pd_meta, pd->pd_subdisks,
715 		    disk->d_consumer->provider->mediasize /
716 		    disk->d_consumer->provider->sectorsize,
717 		    &eoff, &esize);
718 		if (esize == 0) {
719 			G_RAID_DEBUG1(1, sc, "No free space on disk %s",
720 			    g_raid_get_diskname(disk));
721 			goto nofit;
722 		}
723 		size = INT64_MAX;
724 		for (i = 0; i < vol->v_disks_count; i++) {
725 			sd = &vol->v_subdisks[i];
726 			if (sd->sd_state != G_RAID_SUBDISK_S_NONE)
727 				size = sd->sd_size;
728 			if (sd->sd_state <= G_RAID_SUBDISK_S_FAILED &&
729 			    (disk_pos < 0 ||
730 			     vol->v_subdisks[i].sd_state < sd->sd_state))
731 				disk_pos = i;
732 		}
733 		if (disk_pos >= 0 &&
734 		    vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT &&
735 		    (off_t)esize * 512 < size) {
736 			G_RAID_DEBUG1(1, sc, "Disk %s free space "
737 			    "is too small (%ju < %ju)",
738 			    g_raid_get_diskname(disk),
739 			    (off_t)esize * 512, size);
740 			disk_pos = -1;
741 		}
742 		if (disk_pos >= 0) {
743 			if (vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT)
744 				esize = size / 512;
745 			/* For RAID0+1 we need to translate order. */
746 			md_disk_pos = promise_meta_translate_disk(vol, disk_pos);
747 		} else {
748 nofit:
749 			if (pd->pd_subdisks == 0) {
750 				g_raid_change_disk_state(disk,
751 				    G_RAID_DISK_S_SPARE);
752 			}
753 			return (0);
754 		}
755 		G_RAID_DEBUG1(1, sc, "Disk %s takes pos %d in the volume %s",
756 		    g_raid_get_diskname(disk), disk_pos, vol->v_name);
757 		resurrection = 1;
758 	}
759 
760 	sd = &vol->v_subdisks[disk_pos];
761 
762 	if (resurrection && sd->sd_disk != NULL) {
763 		g_raid_change_disk_state(sd->sd_disk,
764 		    G_RAID_DISK_S_STALE_FAILED);
765 		TAILQ_REMOVE(&sd->sd_disk->d_subdisks,
766 		    sd, sd_next);
767 	}
768 	vol->v_subdisks[disk_pos].sd_disk = disk;
769 	TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
770 
771 	/* Welcome the new disk. */
772 	if (resurrection)
773 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
774 	else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN)
775 		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
776 	else
777 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
778 
779 	if (resurrection) {
780 		sd->sd_offset = (off_t)eoff * 512;
781 		sd->sd_size = (off_t)esize * 512;
782 	} else {
783 		sd->sd_offset = (((off_t)pd->pd_meta[sdn]->disk_offset_high
784 		    << 32) + pd->pd_meta[sdn]->disk_offset) * 512;
785 		sd->sd_size = (((off_t)pd->pd_meta[sdn]->disk_sectors_high
786 		    << 32) + pd->pd_meta[sdn]->disk_sectors) * 512;
787 	}
788 
789 	if (resurrection) {
790 		/* Stale disk, almost same as new. */
791 		g_raid_change_subdisk_state(sd,
792 		    G_RAID_SUBDISK_S_NEW);
793 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) {
794 		/* Failed disk. */
795 		g_raid_change_subdisk_state(sd,
796 		    G_RAID_SUBDISK_S_FAILED);
797 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_REDIR) {
798 		/* Rebuilding disk. */
799 		g_raid_change_subdisk_state(sd,
800 		    G_RAID_SUBDISK_S_REBUILD);
801 		if (pd->pd_meta[sdn]->generation != meta->generation)
802 			sd->sd_rebuild_pos = 0;
803 		else {
804 			sd->sd_rebuild_pos =
805 			    (((off_t)pd->pd_meta[sdn]->disk_rebuild_high << 32) +
806 			     pd->pd_meta[sdn]->disk_rebuild) * 512;
807 		}
808 	} else if (!(meta->disks[md_disk_pos].flags & PROMISE_F_ONLINE)) {
809 		/* Rebuilding disk. */
810 		g_raid_change_subdisk_state(sd,
811 		    G_RAID_SUBDISK_S_NEW);
812 	} else if (pd->pd_meta[sdn]->generation != meta->generation ||
813 	    (meta->status & PROMISE_S_MARKED)) {
814 		/* Stale disk or dirty volume (unclean shutdown). */
815 		g_raid_change_subdisk_state(sd,
816 		    G_RAID_SUBDISK_S_STALE);
817 	} else {
818 		/* Up to date disk. */
819 		g_raid_change_subdisk_state(sd,
820 		    G_RAID_SUBDISK_S_ACTIVE);
821 	}
822 	g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
823 	    G_RAID_EVENT_SUBDISK);
824 
825 	return (resurrection);
826 }
827 
828 static void
829 g_raid_md_promise_refill(struct g_raid_softc *sc)
830 {
831 	struct g_raid_volume *vol;
832 	struct g_raid_subdisk *sd;
833 	struct g_raid_disk *disk;
834 	struct g_raid_md_object *md;
835 	struct g_raid_md_promise_perdisk *pd;
836 	struct g_raid_md_promise_pervolume *pv;
837 	int update, updated, i, bad;
838 
839 	md = sc->sc_md;
840 restart:
841 	updated = 0;
842 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
843 		pv = vol->v_md_data;
844 		if (!pv->pv_started || vol->v_stopping)
845 			continue;
846 
847 		/* Search for subdisk that needs replacement. */
848 		bad = 0;
849 		for (i = 0; i < vol->v_disks_count; i++) {
850 			sd = &vol->v_subdisks[i];
851 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
852 			    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
853 			        bad = 1;
854 		}
855 		if (!bad)
856 			continue;
857 
858 		G_RAID_DEBUG1(1, sc, "Volume %s is not complete, "
859 		    "trying to refill.", vol->v_name);
860 
861 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
862 			/* Skip failed. */
863 			if (disk->d_state < G_RAID_DISK_S_SPARE)
864 				continue;
865 			/* Skip already used by this volume. */
866 			for (i = 0; i < vol->v_disks_count; i++) {
867 				sd = &vol->v_subdisks[i];
868 				if (sd->sd_disk == disk)
869 					break;
870 			}
871 			if (i < vol->v_disks_count)
872 				continue;
873 
874 			/* Try to use disk if it has empty extents. */
875 			pd = disk->d_md_data;
876 			if (pd->pd_subdisks < PROMISE_MAX_SUBDISKS) {
877 				update =
878 				    g_raid_md_promise_start_disk(disk, -1, vol);
879 			} else
880 				update = 0;
881 			if (update) {
882 				updated = 1;
883 				g_raid_md_write_promise(md, vol, NULL, disk);
884 				break;
885 			}
886 		}
887 	}
888 	if (updated)
889 		goto restart;
890 }
891 
892 static void
893 g_raid_md_promise_start(struct g_raid_volume *vol)
894 {
895 	struct g_raid_softc *sc;
896 	struct g_raid_subdisk *sd;
897 	struct g_raid_disk *disk;
898 	struct g_raid_md_object *md;
899 	struct g_raid_md_promise_perdisk *pd;
900 	struct g_raid_md_promise_pervolume *pv;
901 	struct promise_raid_conf *meta;
902 	u_int i;
903 
904 	sc = vol->v_softc;
905 	md = sc->sc_md;
906 	pv = vol->v_md_data;
907 	meta = pv->pv_meta;
908 
909 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
910 	if (meta->type == PROMISE_T_RAID0)
911 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
912 	else if (meta->type == PROMISE_T_RAID1) {
913 		if (meta->array_width == 1)
914 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
915 		else
916 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
917 	} else if (meta->type == PROMISE_T_RAID3)
918 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID3;
919 	else if (meta->type == PROMISE_T_RAID5) {
920 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
921 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
922 	} else if (meta->type == PROMISE_T_SPAN)
923 		vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
924 	else if (meta->type == PROMISE_T_JBOD)
925 		vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
926 	else
927 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
928 	vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ
929 	vol->v_disks_count = meta->total_disks;
930 	vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ
931 	if (meta->total_sectors_high < 256) /* If value looks sane. */
932 		vol->v_mediasize +=
933 		    ((off_t)meta->total_sectors_high << 32) * 512; //ZZZ
934 	vol->v_sectorsize = 512 * meta->sector_size;
935 	for (i = 0; i < vol->v_disks_count; i++) {
936 		sd = &vol->v_subdisks[i];
937 		sd->sd_offset = (((off_t)meta->disk_offset_high << 32) +
938 		    meta->disk_offset) * 512;
939 		sd->sd_size = (((off_t)meta->disk_sectors_high << 32) +
940 		    meta->disk_sectors) * 512;
941 	}
942 	g_raid_start_volume(vol);
943 
944 	/* Make all disks found till the moment take their places. */
945 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
946 		pd = disk->d_md_data;
947 		for (i = 0; i < pd->pd_subdisks; i++) {
948 			if (pd->pd_meta[i]->volume_id == meta->volume_id)
949 				g_raid_md_promise_start_disk(disk, i, vol);
950 		}
951 	}
952 
953 	pv->pv_started = 1;
954 	callout_stop(&pv->pv_start_co);
955 	G_RAID_DEBUG1(0, sc, "Volume started.");
956 	g_raid_md_write_promise(md, vol, NULL, NULL);
957 
958 	/* Pickup any STALE/SPARE disks to refill array if needed. */
959 	g_raid_md_promise_refill(sc);
960 
961 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
962 }
963 
964 static void
965 g_raid_promise_go(void *arg)
966 {
967 	struct g_raid_volume *vol;
968 	struct g_raid_softc *sc;
969 	struct g_raid_md_promise_pervolume *pv;
970 
971 	vol = arg;
972 	pv = vol->v_md_data;
973 	sc = vol->v_softc;
974 	if (!pv->pv_started) {
975 		G_RAID_DEBUG1(0, sc, "Force volume start due to timeout.");
976 		g_raid_event_send(vol, G_RAID_VOLUME_E_STARTMD,
977 		    G_RAID_EVENT_VOLUME);
978 	}
979 }
980 
981 static void
982 g_raid_md_promise_new_disk(struct g_raid_disk *disk)
983 {
984 	struct g_raid_softc *sc;
985 	struct g_raid_md_object *md;
986 	struct promise_raid_conf *pdmeta;
987 	struct g_raid_md_promise_perdisk *pd;
988 	struct g_raid_md_promise_pervolume *pv;
989 	struct g_raid_volume *vol;
990 	int i;
991 	char buf[33];
992 
993 	sc = disk->d_softc;
994 	md = sc->sc_md;
995 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
996 
997 	if (pd->pd_subdisks == 0) {
998 		g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
999 		g_raid_md_promise_refill(sc);
1000 		return;
1001 	}
1002 
1003 	for (i = 0; i < pd->pd_subdisks; i++) {
1004 		pdmeta = pd->pd_meta[i];
1005 
1006 		/* Look for volume with matching ID. */
1007 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1008 		if (vol == NULL) {
1009 			promise_meta_get_name(pdmeta, buf);
1010 			vol = g_raid_create_volume(sc, buf, pdmeta->array_number);
1011 			pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1012 			pv->pv_id = pdmeta->volume_id;
1013 			vol->v_md_data = pv;
1014 			callout_init(&pv->pv_start_co, 1);
1015 			callout_reset(&pv->pv_start_co,
1016 			    g_raid_start_timeout * hz,
1017 			    g_raid_promise_go, vol);
1018 		} else
1019 			pv = vol->v_md_data;
1020 
1021 		/* If we haven't started yet - check metadata freshness. */
1022 		if (pv->pv_meta == NULL || !pv->pv_started) {
1023 			if (pv->pv_meta == NULL ||
1024 			    ((int16_t)(pdmeta->generation - pv->pv_generation)) > 0) {
1025 				G_RAID_DEBUG1(1, sc, "Newer disk");
1026 				if (pv->pv_meta != NULL)
1027 					free(pv->pv_meta, M_MD_PROMISE);
1028 				pv->pv_meta = promise_meta_copy(pdmeta);
1029 				pv->pv_generation = pv->pv_meta->generation;
1030 				pv->pv_disks_present = 1;
1031 			} else if (pdmeta->generation == pv->pv_generation) {
1032 				pv->pv_disks_present++;
1033 				G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
1034 				    pv->pv_disks_present,
1035 				    pv->pv_meta->total_disks);
1036 			} else {
1037 				G_RAID_DEBUG1(1, sc, "Older disk");
1038 			}
1039 		}
1040 	}
1041 
1042 	for (i = 0; i < pd->pd_subdisks; i++) {
1043 		pdmeta = pd->pd_meta[i];
1044 
1045 		/* Look for volume with matching ID. */
1046 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1047 		if (vol == NULL)
1048 			continue;
1049 		pv = vol->v_md_data;
1050 
1051 		if (pv->pv_started) {
1052 			if (g_raid_md_promise_start_disk(disk, i, vol))
1053 				g_raid_md_write_promise(md, vol, NULL, NULL);
1054 		} else {
1055 			/* If we collected all needed disks - start array. */
1056 			if (pv->pv_disks_present == pv->pv_meta->total_disks)
1057 				g_raid_md_promise_start(vol);
1058 		}
1059 	}
1060 }
1061 
1062 static int
1063 g_raid_md_create_promise(struct g_raid_md_object *md, struct g_class *mp,
1064     struct g_geom **gp)
1065 {
1066 	struct g_geom *geom;
1067 	struct g_raid_softc *sc;
1068 
1069 	/* Search for existing node. */
1070 	LIST_FOREACH(geom, &mp->geom, geom) {
1071 		sc = geom->softc;
1072 		if (sc == NULL)
1073 			continue;
1074 		if (sc->sc_stopping != 0)
1075 			continue;
1076 		if (sc->sc_md->mdo_class != md->mdo_class)
1077 			continue;
1078 		break;
1079 	}
1080 	if (geom != NULL) {
1081 		*gp = geom;
1082 		return (G_RAID_MD_TASTE_EXISTING);
1083 	}
1084 
1085 	/* Create new one if not found. */
1086 	sc = g_raid_create_node(mp, "Promise", md);
1087 	if (sc == NULL)
1088 		return (G_RAID_MD_TASTE_FAIL);
1089 	md->mdo_softc = sc;
1090 	*gp = sc->sc_geom;
1091 	return (G_RAID_MD_TASTE_NEW);
1092 }
1093 
1094 static int
1095 g_raid_md_taste_promise(struct g_raid_md_object *md, struct g_class *mp,
1096                               struct g_consumer *cp, struct g_geom **gp)
1097 {
1098 	struct g_consumer *rcp;
1099 	struct g_provider *pp;
1100 	struct g_raid_softc *sc;
1101 	struct g_raid_disk *disk;
1102 	struct promise_raid_conf *metaarr[4];
1103 	struct g_raid_md_promise_perdisk *pd;
1104 	struct g_geom *geom;
1105 	int i, j, result, len, subdisks;
1106 	char name[16];
1107 	uint16_t vendor;
1108 
1109 	G_RAID_DEBUG(1, "Tasting Promise on %s", cp->provider->name);
1110 	pp = cp->provider;
1111 
1112 	/* Read metadata from device. */
1113 	g_topology_unlock();
1114 	vendor = 0xffff;
1115 	len = sizeof(vendor);
1116 	if (pp->geom->rank == 1)
1117 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1118 	subdisks = promise_meta_read(cp, metaarr);
1119 	g_topology_lock();
1120 	if (subdisks == 0) {
1121 		if (g_raid_aggressive_spare) {
1122 			if (vendor == 0x105a || vendor == 0x1002) {
1123 				G_RAID_DEBUG(1,
1124 				    "No Promise metadata, forcing spare.");
1125 				goto search;
1126 			} else {
1127 				G_RAID_DEBUG(1,
1128 				    "Promise/ATI vendor mismatch "
1129 				    "0x%04x != 0x105a/0x1002",
1130 				    vendor);
1131 			}
1132 		}
1133 		return (G_RAID_MD_TASTE_FAIL);
1134 	}
1135 
1136 	/* Metadata valid. Print it. */
1137 	for (i = 0; i < subdisks; i++)
1138 		g_raid_md_promise_print(metaarr[i]);
1139 
1140 	/* Purge meaningless (empty/spare) records. */
1141 	for (i = 0; i < subdisks; ) {
1142 		if (metaarr[i]->disk.flags & PROMISE_F_ASSIGNED) {
1143 			i++;
1144 			continue;
1145 		}
1146 		free(metaarr[i], M_MD_PROMISE);
1147 		for (j = i; j < subdisks - 1; j++)
1148 			metaarr[i] = metaarr[j + 1];
1149 		metaarr[subdisks - 1] = NULL;
1150 		subdisks--;
1151 	}
1152 
1153 search:
1154 	/* Search for matching node. */
1155 	sc = NULL;
1156 	LIST_FOREACH(geom, &mp->geom, geom) {
1157 		sc = geom->softc;
1158 		if (sc == NULL)
1159 			continue;
1160 		if (sc->sc_stopping != 0)
1161 			continue;
1162 		if (sc->sc_md->mdo_class != md->mdo_class)
1163 			continue;
1164 		break;
1165 	}
1166 
1167 	/* Found matching node. */
1168 	if (geom != NULL) {
1169 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1170 		result = G_RAID_MD_TASTE_EXISTING;
1171 
1172 	} else { /* Not found matching node -- create one. */
1173 		result = G_RAID_MD_TASTE_NEW;
1174 		snprintf(name, sizeof(name), "Promise");
1175 		sc = g_raid_create_node(mp, name, md);
1176 		md->mdo_softc = sc;
1177 		geom = sc->sc_geom;
1178 	}
1179 
1180 	/* There is no return after this point, so we close passed consumer. */
1181 	g_access(cp, -1, 0, 0);
1182 
1183 	rcp = g_new_consumer(geom);
1184 	rcp->flags |= G_CF_DIRECT_RECEIVE;
1185 	g_attach(rcp, pp);
1186 	if (g_access(rcp, 1, 1, 1) != 0)
1187 		; //goto fail1;
1188 
1189 	g_topology_unlock();
1190 	sx_xlock(&sc->sc_lock);
1191 
1192 	pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1193 	pd->pd_subdisks = subdisks;
1194 	for (i = 0; i < subdisks; i++)
1195 		pd->pd_meta[i] = metaarr[i];
1196 	disk = g_raid_create_disk(sc);
1197 	disk->d_md_data = (void *)pd;
1198 	disk->d_consumer = rcp;
1199 	rcp->private = disk;
1200 
1201 	g_raid_get_disk_info(disk);
1202 
1203 	g_raid_md_promise_new_disk(disk);
1204 
1205 	sx_xunlock(&sc->sc_lock);
1206 	g_topology_lock();
1207 	*gp = geom;
1208 	return (result);
1209 }
1210 
1211 static int
1212 g_raid_md_event_promise(struct g_raid_md_object *md,
1213     struct g_raid_disk *disk, u_int event)
1214 {
1215 	struct g_raid_softc *sc;
1216 
1217 	sc = md->mdo_softc;
1218 	if (disk == NULL)
1219 		return (-1);
1220 	switch (event) {
1221 	case G_RAID_DISK_E_DISCONNECTED:
1222 		/* Delete disk. */
1223 		g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1224 		g_raid_destroy_disk(disk);
1225 		g_raid_md_promise_purge_volumes(sc);
1226 
1227 		/* Write updated metadata to all disks. */
1228 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1229 
1230 		/* Check if anything left. */
1231 		if (g_raid_ndisks(sc, -1) == 0)
1232 			g_raid_destroy_node(sc, 0);
1233 		else
1234 			g_raid_md_promise_refill(sc);
1235 		return (0);
1236 	}
1237 	return (-2);
1238 }
1239 
1240 static int
1241 g_raid_md_volume_event_promise(struct g_raid_md_object *md,
1242     struct g_raid_volume *vol, u_int event)
1243 {
1244 	struct g_raid_md_promise_pervolume *pv;
1245 
1246 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1247 	switch (event) {
1248 	case G_RAID_VOLUME_E_STARTMD:
1249 		if (!pv->pv_started)
1250 			g_raid_md_promise_start(vol);
1251 		return (0);
1252 	}
1253 	return (-2);
1254 }
1255 
1256 static int
1257 g_raid_md_ctl_promise(struct g_raid_md_object *md,
1258     struct gctl_req *req)
1259 {
1260 	struct g_raid_softc *sc;
1261 	struct g_raid_volume *vol, *vol1;
1262 	struct g_raid_subdisk *sd;
1263 	struct g_raid_disk *disk, *disks[PROMISE_MAX_DISKS];
1264 	struct g_raid_md_promise_perdisk *pd;
1265 	struct g_raid_md_promise_pervolume *pv;
1266 	struct g_consumer *cp;
1267 	struct g_provider *pp;
1268 	char arg[16];
1269 	const char *nodename, *verb, *volname, *levelname, *diskname;
1270 	char *tmp;
1271 	int *nargs, *force;
1272 	off_t esize, offs[PROMISE_MAX_DISKS], size, sectorsize, strip;
1273 	intmax_t *sizearg, *striparg;
1274 	int numdisks, i, len, level, qual;
1275 	int error;
1276 
1277 	sc = md->mdo_softc;
1278 	verb = gctl_get_param(req, "verb", NULL);
1279 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1280 	error = 0;
1281 	if (strcmp(verb, "label") == 0) {
1282 		if (*nargs < 4) {
1283 			gctl_error(req, "Invalid number of arguments.");
1284 			return (-1);
1285 		}
1286 		volname = gctl_get_asciiparam(req, "arg1");
1287 		if (volname == NULL) {
1288 			gctl_error(req, "No volume name.");
1289 			return (-2);
1290 		}
1291 		levelname = gctl_get_asciiparam(req, "arg2");
1292 		if (levelname == NULL) {
1293 			gctl_error(req, "No RAID level.");
1294 			return (-3);
1295 		}
1296 		if (strcasecmp(levelname, "RAID5") == 0)
1297 			levelname = "RAID5-LA";
1298 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1299 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1300 			return (-4);
1301 		}
1302 		numdisks = *nargs - 3;
1303 		force = gctl_get_paraml(req, "force", sizeof(*force));
1304 		if (!g_raid_md_promise_supported(level, qual, numdisks,
1305 		    force ? *force : 0)) {
1306 			gctl_error(req, "Unsupported RAID level "
1307 			    "(0x%02x/0x%02x), or number of disks (%d).",
1308 			    level, qual, numdisks);
1309 			return (-5);
1310 		}
1311 
1312 		/* Search for disks, connect them and probe. */
1313 		size = INT64_MAX;
1314 		sectorsize = 0;
1315 		bzero(disks, sizeof(disks));
1316 		bzero(offs, sizeof(offs));
1317 		for (i = 0; i < numdisks; i++) {
1318 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1319 			diskname = gctl_get_asciiparam(req, arg);
1320 			if (diskname == NULL) {
1321 				gctl_error(req, "No disk name (%s).", arg);
1322 				error = -6;
1323 				break;
1324 			}
1325 			if (strcmp(diskname, "NONE") == 0)
1326 				continue;
1327 
1328 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1329 				if (disk->d_consumer != NULL &&
1330 				    disk->d_consumer->provider != NULL &&
1331 				    strcmp(disk->d_consumer->provider->name,
1332 				     diskname) == 0)
1333 					break;
1334 			}
1335 			if (disk != NULL) {
1336 				if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
1337 					gctl_error(req, "Disk '%s' is in a "
1338 					    "wrong state (%s).", diskname,
1339 					    g_raid_disk_state2str(disk->d_state));
1340 					error = -7;
1341 					break;
1342 				}
1343 				pd = disk->d_md_data;
1344 				if (pd->pd_subdisks >= PROMISE_MAX_SUBDISKS) {
1345 					gctl_error(req, "Disk '%s' already "
1346 					    "used by %d volumes.",
1347 					    diskname, pd->pd_subdisks);
1348 					error = -7;
1349 					break;
1350 				}
1351 				pp = disk->d_consumer->provider;
1352 				disks[i] = disk;
1353 				promise_meta_unused_range(pd->pd_meta,
1354 				    pd->pd_subdisks,
1355 				    pp->mediasize / pp->sectorsize,
1356 				    &offs[i], &esize);
1357 				size = MIN(size, (off_t)esize * pp->sectorsize);
1358 				sectorsize = MAX(sectorsize, pp->sectorsize);
1359 				continue;
1360 			}
1361 
1362 			g_topology_lock();
1363 			cp = g_raid_open_consumer(sc, diskname);
1364 			if (cp == NULL) {
1365 				gctl_error(req, "Can't open disk '%s'.",
1366 				    diskname);
1367 				g_topology_unlock();
1368 				error = -8;
1369 				break;
1370 			}
1371 			pp = cp->provider;
1372 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1373 			disk = g_raid_create_disk(sc);
1374 			disk->d_md_data = (void *)pd;
1375 			disk->d_consumer = cp;
1376 			disks[i] = disk;
1377 			cp->private = disk;
1378 			g_topology_unlock();
1379 
1380 			g_raid_get_disk_info(disk);
1381 
1382 			/* Reserve some space for metadata. */
1383 			size = MIN(size, pp->mediasize - 131072llu * pp->sectorsize);
1384 			sectorsize = MAX(sectorsize, pp->sectorsize);
1385 		}
1386 		if (error != 0) {
1387 			for (i = 0; i < numdisks; i++) {
1388 				if (disks[i] != NULL &&
1389 				    disks[i]->d_state == G_RAID_DISK_S_NONE)
1390 					g_raid_destroy_disk(disks[i]);
1391 			}
1392 			return (error);
1393 		}
1394 
1395 		if (sectorsize <= 0) {
1396 			gctl_error(req, "Can't get sector size.");
1397 			return (-8);
1398 		}
1399 
1400 		/* Handle size argument. */
1401 		len = sizeof(*sizearg);
1402 		sizearg = gctl_get_param(req, "size", &len);
1403 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1404 		    *sizearg > 0) {
1405 			if (*sizearg > size) {
1406 				gctl_error(req, "Size too big %lld > %lld.",
1407 				    (long long)*sizearg, (long long)size);
1408 				return (-9);
1409 			}
1410 			size = *sizearg;
1411 		}
1412 
1413 		/* Handle strip argument. */
1414 		strip = 131072;
1415 		len = sizeof(*striparg);
1416 		striparg = gctl_get_param(req, "strip", &len);
1417 		if (striparg != NULL && len == sizeof(*striparg) &&
1418 		    *striparg > 0) {
1419 			if (*striparg < sectorsize) {
1420 				gctl_error(req, "Strip size too small.");
1421 				return (-10);
1422 			}
1423 			if (*striparg % sectorsize != 0) {
1424 				gctl_error(req, "Incorrect strip size.");
1425 				return (-11);
1426 			}
1427 			strip = *striparg;
1428 		}
1429 
1430 		/* Round size down to strip or sector. */
1431 		if (level == G_RAID_VOLUME_RL_RAID1 ||
1432 		    level == G_RAID_VOLUME_RL_SINGLE ||
1433 		    level == G_RAID_VOLUME_RL_CONCAT)
1434 			size -= (size % sectorsize);
1435 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1436 		    (numdisks & 1) != 0)
1437 			size -= (size % (2 * strip));
1438 		else
1439 			size -= (size % strip);
1440 		if (size <= 0) {
1441 			gctl_error(req, "Size too small.");
1442 			return (-13);
1443 		}
1444 
1445 		/* We have all we need, create things: volume, ... */
1446 		pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1447 		arc4rand(&pv->pv_id, sizeof(pv->pv_id), 0);
1448 		pv->pv_generation = 0;
1449 		pv->pv_started = 1;
1450 		vol = g_raid_create_volume(sc, volname, -1);
1451 		vol->v_md_data = pv;
1452 		vol->v_raid_level = level;
1453 		vol->v_raid_level_qualifier = qual;
1454 		vol->v_strip_size = strip;
1455 		vol->v_disks_count = numdisks;
1456 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1457 		    level == G_RAID_VOLUME_RL_CONCAT ||
1458 		    level == G_RAID_VOLUME_RL_SINGLE)
1459 			vol->v_mediasize = size * numdisks;
1460 		else if (level == G_RAID_VOLUME_RL_RAID1)
1461 			vol->v_mediasize = size;
1462 		else if (level == G_RAID_VOLUME_RL_RAID3 ||
1463 		    level == G_RAID_VOLUME_RL_RAID5)
1464 			vol->v_mediasize = size * (numdisks - 1);
1465 		else { /* RAID1E */
1466 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1467 			    strip;
1468 		}
1469 		vol->v_sectorsize = sectorsize;
1470 		g_raid_start_volume(vol);
1471 
1472 		/* , and subdisks. */
1473 		for (i = 0; i < numdisks; i++) {
1474 			disk = disks[i];
1475 			sd = &vol->v_subdisks[i];
1476 			sd->sd_disk = disk;
1477 			sd->sd_offset = (off_t)offs[i] * 512;
1478 			sd->sd_size = size;
1479 			if (disk == NULL)
1480 				continue;
1481 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1482 			g_raid_change_disk_state(disk,
1483 			    G_RAID_DISK_S_ACTIVE);
1484 			g_raid_change_subdisk_state(sd,
1485 			    G_RAID_SUBDISK_S_ACTIVE);
1486 			g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1487 			    G_RAID_EVENT_SUBDISK);
1488 		}
1489 
1490 		/* Write metadata based on created entities. */
1491 		G_RAID_DEBUG1(0, sc, "Array started.");
1492 		g_raid_md_write_promise(md, vol, NULL, NULL);
1493 
1494 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1495 		g_raid_md_promise_refill(sc);
1496 
1497 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1498 		    G_RAID_EVENT_VOLUME);
1499 		return (0);
1500 	}
1501 	if (strcmp(verb, "add") == 0) {
1502 		gctl_error(req, "`add` command is not applicable, "
1503 		    "use `label` instead.");
1504 		return (-99);
1505 	}
1506 	if (strcmp(verb, "delete") == 0) {
1507 		nodename = gctl_get_asciiparam(req, "arg0");
1508 		if (nodename != NULL && strcasecmp(sc->sc_name, nodename) != 0)
1509 			nodename = NULL;
1510 
1511 		/* Full node destruction. */
1512 		if (*nargs == 1 && nodename != NULL) {
1513 			/* Check if some volume is still open. */
1514 			force = gctl_get_paraml(req, "force", sizeof(*force));
1515 			if (force != NULL && *force == 0 &&
1516 			    g_raid_nopens(sc) != 0) {
1517 				gctl_error(req, "Some volume is still open.");
1518 				return (-4);
1519 			}
1520 
1521 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1522 				if (disk->d_consumer)
1523 					promise_meta_erase(disk->d_consumer);
1524 			}
1525 			g_raid_destroy_node(sc, 0);
1526 			return (0);
1527 		}
1528 
1529 		/* Destroy specified volume. If it was last - all node. */
1530 		if (*nargs > 2) {
1531 			gctl_error(req, "Invalid number of arguments.");
1532 			return (-1);
1533 		}
1534 		volname = gctl_get_asciiparam(req,
1535 		    nodename != NULL ? "arg1" : "arg0");
1536 		if (volname == NULL) {
1537 			gctl_error(req, "No volume name.");
1538 			return (-2);
1539 		}
1540 
1541 		/* Search for volume. */
1542 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1543 			if (strcmp(vol->v_name, volname) == 0)
1544 				break;
1545 			pp = vol->v_provider;
1546 			if (pp == NULL)
1547 				continue;
1548 			if (strcmp(pp->name, volname) == 0)
1549 				break;
1550 			if (strncmp(pp->name, "raid/", 5) == 0 &&
1551 			    strcmp(pp->name + 5, volname) == 0)
1552 				break;
1553 		}
1554 		if (vol == NULL) {
1555 			i = strtol(volname, &tmp, 10);
1556 			if (verb != volname && tmp[0] == 0) {
1557 				TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1558 					if (vol->v_global_id == i)
1559 						break;
1560 				}
1561 			}
1562 		}
1563 		if (vol == NULL) {
1564 			gctl_error(req, "Volume '%s' not found.", volname);
1565 			return (-3);
1566 		}
1567 
1568 		/* Check if volume is still open. */
1569 		force = gctl_get_paraml(req, "force", sizeof(*force));
1570 		if (force != NULL && *force == 0 &&
1571 		    vol->v_provider_open != 0) {
1572 			gctl_error(req, "Volume is still open.");
1573 			return (-4);
1574 		}
1575 
1576 		/* Destroy volume and potentially node. */
1577 		i = 0;
1578 		TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
1579 			i++;
1580 		if (i >= 2) {
1581 			g_raid_destroy_volume(vol);
1582 			g_raid_md_promise_purge_disks(sc);
1583 			g_raid_md_write_promise(md, NULL, NULL, NULL);
1584 		} else {
1585 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1586 				if (disk->d_consumer)
1587 					promise_meta_erase(disk->d_consumer);
1588 			}
1589 			g_raid_destroy_node(sc, 0);
1590 		}
1591 		return (0);
1592 	}
1593 	if (strcmp(verb, "remove") == 0 ||
1594 	    strcmp(verb, "fail") == 0) {
1595 		if (*nargs < 2) {
1596 			gctl_error(req, "Invalid number of arguments.");
1597 			return (-1);
1598 		}
1599 		for (i = 1; i < *nargs; i++) {
1600 			snprintf(arg, sizeof(arg), "arg%d", i);
1601 			diskname = gctl_get_asciiparam(req, arg);
1602 			if (diskname == NULL) {
1603 				gctl_error(req, "No disk name (%s).", arg);
1604 				error = -2;
1605 				break;
1606 			}
1607 			if (strncmp(diskname, _PATH_DEV, 5) == 0)
1608 				diskname += 5;
1609 
1610 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1611 				if (disk->d_consumer != NULL &&
1612 				    disk->d_consumer->provider != NULL &&
1613 				    strcmp(disk->d_consumer->provider->name,
1614 				     diskname) == 0)
1615 					break;
1616 			}
1617 			if (disk == NULL) {
1618 				gctl_error(req, "Disk '%s' not found.",
1619 				    diskname);
1620 				error = -3;
1621 				break;
1622 			}
1623 
1624 			if (strcmp(verb, "fail") == 0) {
1625 				g_raid_md_fail_disk_promise(md, NULL, disk);
1626 				continue;
1627 			}
1628 
1629 			/* Erase metadata on deleting disk and destroy it. */
1630 			promise_meta_erase(disk->d_consumer);
1631 			g_raid_destroy_disk(disk);
1632 		}
1633 		g_raid_md_promise_purge_volumes(sc);
1634 
1635 		/* Write updated metadata to remaining disks. */
1636 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1637 
1638 		/* Check if anything left. */
1639 		if (g_raid_ndisks(sc, -1) == 0)
1640 			g_raid_destroy_node(sc, 0);
1641 		else
1642 			g_raid_md_promise_refill(sc);
1643 		return (error);
1644 	}
1645 	if (strcmp(verb, "insert") == 0) {
1646 		if (*nargs < 2) {
1647 			gctl_error(req, "Invalid number of arguments.");
1648 			return (-1);
1649 		}
1650 		for (i = 1; i < *nargs; i++) {
1651 			/* Get disk name. */
1652 			snprintf(arg, sizeof(arg), "arg%d", i);
1653 			diskname = gctl_get_asciiparam(req, arg);
1654 			if (diskname == NULL) {
1655 				gctl_error(req, "No disk name (%s).", arg);
1656 				error = -3;
1657 				break;
1658 			}
1659 
1660 			/* Try to find provider with specified name. */
1661 			g_topology_lock();
1662 			cp = g_raid_open_consumer(sc, diskname);
1663 			if (cp == NULL) {
1664 				gctl_error(req, "Can't open disk '%s'.",
1665 				    diskname);
1666 				g_topology_unlock();
1667 				error = -4;
1668 				break;
1669 			}
1670 			pp = cp->provider;
1671 			g_topology_unlock();
1672 
1673 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1674 
1675 			disk = g_raid_create_disk(sc);
1676 			disk->d_consumer = cp;
1677 			disk->d_md_data = (void *)pd;
1678 			cp->private = disk;
1679 
1680 			g_raid_get_disk_info(disk);
1681 
1682 			/* Welcome the "new" disk. */
1683 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1684 			promise_meta_write_spare(cp);
1685 			g_raid_md_promise_refill(sc);
1686 		}
1687 		return (error);
1688 	}
1689 	return (-100);
1690 }
1691 
1692 static int
1693 g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1694     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1695 {
1696 	struct g_raid_softc *sc;
1697 	struct g_raid_volume *vol;
1698 	struct g_raid_subdisk *sd;
1699 	struct g_raid_disk *disk;
1700 	struct g_raid_md_promise_perdisk *pd;
1701 	struct g_raid_md_promise_pervolume *pv;
1702 	struct promise_raid_conf *meta;
1703 	off_t rebuild_lba64;
1704 	int i, j, pos, rebuild;
1705 
1706 	sc = md->mdo_softc;
1707 
1708 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1709 		return (0);
1710 
1711 	/* Generate new per-volume metadata for affected volumes. */
1712 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1713 		if (vol->v_stopping)
1714 			continue;
1715 
1716 		/* Skip volumes not related to specified targets. */
1717 		if (tvol != NULL && vol != tvol)
1718 			continue;
1719 		if (tsd != NULL && vol != tsd->sd_volume)
1720 			continue;
1721 		if (tdisk != NULL) {
1722 			for (i = 0; i < vol->v_disks_count; i++) {
1723 				if (vol->v_subdisks[i].sd_disk == tdisk)
1724 					break;
1725 			}
1726 			if (i >= vol->v_disks_count)
1727 				continue;
1728 		}
1729 
1730 		pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1731 		pv->pv_generation++;
1732 
1733 		meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
1734 		if (pv->pv_meta != NULL)
1735 			memcpy(meta, pv->pv_meta, sizeof(*meta));
1736 		memcpy(meta->promise_id, PROMISE_MAGIC,
1737 		    sizeof(PROMISE_MAGIC) - 1);
1738 		meta->dummy_0 = 0x00020000;
1739 		meta->integrity = PROMISE_I_VALID;
1740 
1741 		meta->generation = pv->pv_generation;
1742 		meta->status = PROMISE_S_VALID | PROMISE_S_ONLINE |
1743 		    PROMISE_S_INITED | PROMISE_S_READY;
1744 		if (vol->v_state <= G_RAID_VOLUME_S_DEGRADED)
1745 			meta->status |= PROMISE_S_DEGRADED;
1746 		if (vol->v_dirty)
1747 			meta->status |= PROMISE_S_MARKED; /* XXX: INVENTED! */
1748 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0 ||
1749 		    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1750 			meta->type = PROMISE_T_RAID0;
1751 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1752 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1753 			meta->type = PROMISE_T_RAID1;
1754 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3)
1755 			meta->type = PROMISE_T_RAID3;
1756 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1757 			meta->type = PROMISE_T_RAID5;
1758 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT)
1759 			meta->type = PROMISE_T_SPAN;
1760 		else
1761 			meta->type = PROMISE_T_JBOD;
1762 		meta->total_disks = vol->v_disks_count;
1763 		meta->stripe_shift = ffs(vol->v_strip_size / 1024);
1764 		meta->array_width = vol->v_disks_count;
1765 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1766 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1767 			meta->array_width /= 2;
1768 		meta->array_number = vol->v_global_id;
1769 		meta->total_sectors = vol->v_mediasize / 512;
1770 		meta->total_sectors_high = (vol->v_mediasize / 512) >> 32;
1771 		meta->sector_size = vol->v_sectorsize / 512;
1772 		meta->cylinders = meta->total_sectors / (255 * 63) - 1;
1773 		meta->heads = 254;
1774 		meta->sectors = 63;
1775 		meta->volume_id = pv->pv_id;
1776 		rebuild_lba64 = UINT64_MAX;
1777 		rebuild = 0;
1778 		for (i = 0; i < vol->v_disks_count; i++) {
1779 			sd = &vol->v_subdisks[i];
1780 			/* For RAID0+1 we need to translate order. */
1781 			pos = promise_meta_translate_disk(vol, i);
1782 			meta->disks[pos].flags = PROMISE_F_VALID |
1783 			    PROMISE_F_ASSIGNED;
1784 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE) {
1785 				meta->disks[pos].flags |= 0;
1786 			} else if (sd->sd_state == G_RAID_SUBDISK_S_FAILED) {
1787 				meta->disks[pos].flags |=
1788 				    PROMISE_F_DOWN | PROMISE_F_REDIR;
1789 			} else if (sd->sd_state <= G_RAID_SUBDISK_S_REBUILD) {
1790 				meta->disks[pos].flags |=
1791 				    PROMISE_F_ONLINE | PROMISE_F_REDIR;
1792 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1793 					rebuild_lba64 = MIN(rebuild_lba64,
1794 					    sd->sd_rebuild_pos / 512);
1795 				} else
1796 					rebuild_lba64 = 0;
1797 				rebuild = 1;
1798 			} else {
1799 				meta->disks[pos].flags |= PROMISE_F_ONLINE;
1800 				if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) {
1801 					meta->status |= PROMISE_S_MARKED;
1802 					if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
1803 						rebuild_lba64 = MIN(rebuild_lba64,
1804 						    sd->sd_rebuild_pos / 512);
1805 					} else
1806 						rebuild_lba64 = 0;
1807 				}
1808 			}
1809 			if (pv->pv_meta != NULL) {
1810 				meta->disks[pos].id = pv->pv_meta->disks[pos].id;
1811 			} else {
1812 				meta->disks[pos].number = i * 2;
1813 				arc4rand(&meta->disks[pos].id,
1814 				    sizeof(meta->disks[pos].id), 0);
1815 			}
1816 		}
1817 		promise_meta_put_name(meta, vol->v_name);
1818 
1819 		/* Try to mimic AMD BIOS rebuild/resync behavior. */
1820 		if (rebuild_lba64 != UINT64_MAX) {
1821 			if (rebuild)
1822 				meta->magic_3 = 0x03040010UL; /* Rebuild? */
1823 			else
1824 				meta->magic_3 = 0x03040008UL; /* Resync? */
1825 			/* Translate from per-disk to per-volume LBA. */
1826 			if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1827 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1828 				rebuild_lba64 *= meta->array_width;
1829 			} else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1830 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) {
1831 				rebuild_lba64 *= meta->array_width - 1;
1832 			} else
1833 				rebuild_lba64 = 0;
1834 		} else
1835 			meta->magic_3 = 0x03000000UL;
1836 		meta->rebuild_lba64 = rebuild_lba64;
1837 		meta->magic_4 = 0x04010101UL;
1838 
1839 		/* Replace per-volume metadata with new. */
1840 		if (pv->pv_meta != NULL)
1841 			free(pv->pv_meta, M_MD_PROMISE);
1842 		pv->pv_meta = meta;
1843 
1844 		/* Copy new metadata to the disks, adding or replacing old. */
1845 		for (i = 0; i < vol->v_disks_count; i++) {
1846 			sd = &vol->v_subdisks[i];
1847 			disk = sd->sd_disk;
1848 			if (disk == NULL)
1849 				continue;
1850 			/* For RAID0+1 we need to translate order. */
1851 			pos = promise_meta_translate_disk(vol, i);
1852 			pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1853 			for (j = 0; j < pd->pd_subdisks; j++) {
1854 				if (pd->pd_meta[j]->volume_id == meta->volume_id)
1855 					break;
1856 			}
1857 			if (j == pd->pd_subdisks)
1858 				pd->pd_subdisks++;
1859 			if (pd->pd_meta[j] != NULL)
1860 				free(pd->pd_meta[j], M_MD_PROMISE);
1861 			pd->pd_meta[j] = promise_meta_copy(meta);
1862 			pd->pd_meta[j]->disk = meta->disks[pos];
1863 			pd->pd_meta[j]->disk.number = pos;
1864 			pd->pd_meta[j]->disk_offset_high =
1865 			    (sd->sd_offset / 512) >> 32;
1866 			pd->pd_meta[j]->disk_offset = sd->sd_offset / 512;
1867 			pd->pd_meta[j]->disk_sectors_high =
1868 			    (sd->sd_size / 512) >> 32;
1869 			pd->pd_meta[j]->disk_sectors = sd->sd_size / 512;
1870 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1871 				pd->pd_meta[j]->disk_rebuild_high =
1872 				    (sd->sd_rebuild_pos / 512) >> 32;
1873 				pd->pd_meta[j]->disk_rebuild =
1874 				    sd->sd_rebuild_pos / 512;
1875 			} else if (sd->sd_state < G_RAID_SUBDISK_S_REBUILD) {
1876 				pd->pd_meta[j]->disk_rebuild_high = 0;
1877 				pd->pd_meta[j]->disk_rebuild = 0;
1878 			} else {
1879 				pd->pd_meta[j]->disk_rebuild_high = UINT32_MAX;
1880 				pd->pd_meta[j]->disk_rebuild = UINT32_MAX;
1881 			}
1882 			pd->pd_updated = 1;
1883 		}
1884 	}
1885 
1886 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1887 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1888 		if (disk->d_state != G_RAID_DISK_S_ACTIVE)
1889 			continue;
1890 		if (!pd->pd_updated)
1891 			continue;
1892 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1893 		    g_raid_get_diskname(disk));
1894 		for (i = 0; i < pd->pd_subdisks; i++)
1895 			g_raid_md_promise_print(pd->pd_meta[i]);
1896 		promise_meta_write(disk->d_consumer,
1897 		    pd->pd_meta, pd->pd_subdisks);
1898 		pd->pd_updated = 0;
1899 	}
1900 
1901 	return (0);
1902 }
1903 
1904 static int
1905 g_raid_md_fail_disk_promise(struct g_raid_md_object *md,
1906     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1907 {
1908 	struct g_raid_softc *sc;
1909 	struct g_raid_md_promise_perdisk *pd;
1910 	struct g_raid_subdisk *sd;
1911 	int i, pos;
1912 
1913 	sc = md->mdo_softc;
1914 	pd = (struct g_raid_md_promise_perdisk *)tdisk->d_md_data;
1915 
1916 	/* We can't fail disk that is not a part of array now. */
1917 	if (tdisk->d_state != G_RAID_DISK_S_ACTIVE)
1918 		return (-1);
1919 
1920 	/*
1921 	 * Mark disk as failed in metadata and try to write that metadata
1922 	 * to the disk itself to prevent it's later resurrection as STALE.
1923 	 */
1924 	if (pd->pd_subdisks > 0 && tdisk->d_consumer != NULL)
1925 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1926 		    g_raid_get_diskname(tdisk));
1927 	for (i = 0; i < pd->pd_subdisks; i++) {
1928 		pd->pd_meta[i]->disk.flags |=
1929 		    PROMISE_F_DOWN | PROMISE_F_REDIR;
1930 		pos = pd->pd_meta[i]->disk.number;
1931 		if (pos >= 0 && pos < PROMISE_MAX_DISKS) {
1932 			pd->pd_meta[i]->disks[pos].flags |=
1933 			    PROMISE_F_DOWN | PROMISE_F_REDIR;
1934 		}
1935 		g_raid_md_promise_print(pd->pd_meta[i]);
1936 	}
1937 	if (tdisk->d_consumer != NULL)
1938 		promise_meta_write(tdisk->d_consumer,
1939 		    pd->pd_meta, pd->pd_subdisks);
1940 
1941 	/* Change states. */
1942 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1943 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1944 		g_raid_change_subdisk_state(sd,
1945 		    G_RAID_SUBDISK_S_FAILED);
1946 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1947 		    G_RAID_EVENT_SUBDISK);
1948 	}
1949 
1950 	/* Write updated metadata to remaining disks. */
1951 	g_raid_md_write_promise(md, NULL, NULL, tdisk);
1952 
1953 	g_raid_md_promise_refill(sc);
1954 	return (0);
1955 }
1956 
1957 static int
1958 g_raid_md_free_disk_promise(struct g_raid_md_object *md,
1959     struct g_raid_disk *disk)
1960 {
1961 	struct g_raid_md_promise_perdisk *pd;
1962 	int i;
1963 
1964 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1965 	for (i = 0; i < pd->pd_subdisks; i++) {
1966 		if (pd->pd_meta[i] != NULL) {
1967 			free(pd->pd_meta[i], M_MD_PROMISE);
1968 			pd->pd_meta[i] = NULL;
1969 		}
1970 	}
1971 	free(pd, M_MD_PROMISE);
1972 	disk->d_md_data = NULL;
1973 	return (0);
1974 }
1975 
1976 static int
1977 g_raid_md_free_volume_promise(struct g_raid_md_object *md,
1978     struct g_raid_volume *vol)
1979 {
1980 	struct g_raid_md_promise_pervolume *pv;
1981 
1982 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1983 	if (pv && pv->pv_meta != NULL) {
1984 		free(pv->pv_meta, M_MD_PROMISE);
1985 		pv->pv_meta = NULL;
1986 	}
1987 	if (pv && !pv->pv_started) {
1988 		pv->pv_started = 1;
1989 		callout_stop(&pv->pv_start_co);
1990 	}
1991 	free(pv, M_MD_PROMISE);
1992 	vol->v_md_data = NULL;
1993 	return (0);
1994 }
1995 
1996 static int
1997 g_raid_md_free_promise(struct g_raid_md_object *md)
1998 {
1999 
2000 	return (0);
2001 }
2002 
2003 G_RAID_MD_DECLARE(promise, "Promise");
2004