xref: /freebsd/sys/geom/raid/md_promise.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/endian.h>
33 #include <sys/kernel.h>
34 #include <sys/kobj.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/systm.h>
40 #include <geom/geom.h>
41 #include "geom/raid/g_raid.h"
42 #include "g_raid_md_if.h"
43 
44 static MALLOC_DEFINE(M_MD_PROMISE, "md_promise_data", "GEOM_RAID Promise metadata");
45 
46 #define	PROMISE_MAX_DISKS	8
47 #define	PROMISE_MAX_SUBDISKS	2
48 #define	PROMISE_META_OFFSET	14
49 
50 struct promise_raid_disk {
51 	uint8_t		flags;			/* Subdisk status. */
52 #define PROMISE_F_VALID		0x01
53 #define PROMISE_F_ONLINE	0x02
54 #define PROMISE_F_ASSIGNED	0x04
55 #define PROMISE_F_SPARE		0x08
56 #define PROMISE_F_DUPLICATE	0x10
57 #define PROMISE_F_REDIR		0x20
58 #define PROMISE_F_DOWN		0x40
59 #define PROMISE_F_READY		0x80
60 
61 	uint8_t		number;			/* Position in a volume. */
62 	uint8_t		channel;		/* ATA channel number. */
63 	uint8_t		device;			/* ATA device number. */
64 	uint64_t	id __packed;		/* Subdisk ID. */
65 } __packed;
66 
67 struct promise_raid_conf {
68 	char		promise_id[24];
69 #define PROMISE_MAGIC		"Promise Technology, Inc."
70 #define FREEBSD_MAGIC		"FreeBSD ATA driver RAID "
71 
72 	uint32_t	dummy_0;
73 	uint64_t	magic_0;
74 #define PROMISE_MAGIC0(x)	(((uint64_t)(x.channel) << 48) | \
75 				((uint64_t)(x.device != 0) << 56))
76 	uint16_t	magic_1;
77 	uint32_t	magic_2;
78 	uint8_t		filler1[470];
79 
80 	uint32_t	integrity;
81 #define PROMISE_I_VALID		0x00000080
82 
83 	struct promise_raid_disk	disk;	/* This subdisk info. */
84 	uint32_t	disk_offset;		/* Subdisk offset. */
85 	uint32_t	disk_sectors;		/* Subdisk size */
86 	uint32_t	rebuild_lba;		/* Rebuild position. */
87 	uint16_t	generation;		/* Generation number. */
88 	uint8_t		status;			/* Volume status. */
89 #define PROMISE_S_VALID		0x01
90 #define PROMISE_S_ONLINE	0x02
91 #define PROMISE_S_INITED	0x04
92 #define PROMISE_S_READY		0x08
93 #define PROMISE_S_DEGRADED	0x10
94 #define PROMISE_S_MARKED	0x20
95 #define PROMISE_S_MIGRATING	0x40
96 #define PROMISE_S_FUNCTIONAL	0x80
97 
98 	uint8_t		type;			/* Voluem type. */
99 #define PROMISE_T_RAID0		0x00
100 #define PROMISE_T_RAID1		0x01
101 #define PROMISE_T_RAID3		0x02
102 #define PROMISE_T_RAID5		0x04
103 #define PROMISE_T_SPAN		0x08
104 #define PROMISE_T_JBOD		0x10
105 
106 	uint8_t		total_disks;		/* Disks in this volume. */
107 	uint8_t		stripe_shift;		/* Strip size. */
108 	uint8_t		array_width;		/* Number of RAID0 stripes. */
109 	uint8_t		array_number;		/* Global volume number. */
110 	uint32_t	total_sectors;		/* Volume size. */
111 	uint16_t	cylinders;		/* Volume geometry: C. */
112 	uint8_t		heads;			/* Volume geometry: H. */
113 	uint8_t		sectors;		/* Volume geometry: S. */
114 	uint64_t	volume_id __packed;	/* Volume ID, */
115 	struct promise_raid_disk	disks[PROMISE_MAX_DISKS];
116 						/* Subdisks in this volume. */
117 	char		name[32];		/* Volume label. */
118 
119 	uint32_t	filler2[8];
120 	uint32_t	magic_3;	/* Something related to rebuild. */
121 	uint64_t	rebuild_lba64;	/* Per-volume rebuild position. */
122 	uint32_t	magic_4;
123 	uint32_t	magic_5;
124 	uint32_t	filler3[325];
125 	uint32_t	checksum;
126 } __packed;
127 
128 struct g_raid_md_promise_perdisk {
129 	int		 pd_updated;
130 	int		 pd_subdisks;
131 	struct promise_raid_conf	*pd_meta[PROMISE_MAX_SUBDISKS];
132 };
133 
134 struct g_raid_md_promise_pervolume {
135 	struct promise_raid_conf	*pv_meta;
136 	uint64_t			 pv_id;
137 	uint16_t			 pv_generation;
138 	int				 pv_disks_present;
139 	int				 pv_started;
140 	struct callout			 pv_start_co;	/* STARTING state timer. */
141 };
142 
143 static g_raid_md_create_t g_raid_md_create_promise;
144 static g_raid_md_taste_t g_raid_md_taste_promise;
145 static g_raid_md_event_t g_raid_md_event_promise;
146 static g_raid_md_volume_event_t g_raid_md_volume_event_promise;
147 static g_raid_md_ctl_t g_raid_md_ctl_promise;
148 static g_raid_md_write_t g_raid_md_write_promise;
149 static g_raid_md_fail_disk_t g_raid_md_fail_disk_promise;
150 static g_raid_md_free_disk_t g_raid_md_free_disk_promise;
151 static g_raid_md_free_volume_t g_raid_md_free_volume_promise;
152 static g_raid_md_free_t g_raid_md_free_promise;
153 
154 static kobj_method_t g_raid_md_promise_methods[] = {
155 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_promise),
156 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_promise),
157 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_promise),
158 	KOBJMETHOD(g_raid_md_volume_event,	g_raid_md_volume_event_promise),
159 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_promise),
160 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_promise),
161 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_promise),
162 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_promise),
163 	KOBJMETHOD(g_raid_md_free_volume,	g_raid_md_free_volume_promise),
164 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_promise),
165 	{ 0, 0 }
166 };
167 
168 static struct g_raid_md_class g_raid_md_promise_class = {
169 	"Promise",
170 	g_raid_md_promise_methods,
171 	sizeof(struct g_raid_md_object),
172 	.mdc_priority = 100
173 };
174 
175 
176 static void
177 g_raid_md_promise_print(struct promise_raid_conf *meta)
178 {
179 	int i;
180 
181 	if (g_raid_debug < 1)
182 		return;
183 
184 	printf("********* ATA Promise Metadata *********\n");
185 	printf("promise_id          <%.24s>\n", meta->promise_id);
186 	printf("disk                %02x %02x %02x %02x %016jx\n",
187 	    meta->disk.flags, meta->disk.number, meta->disk.channel,
188 	    meta->disk.device, meta->disk.id);
189 	printf("disk_offset         %u\n", meta->disk_offset);
190 	printf("disk_sectors        %u\n", meta->disk_sectors);
191 	printf("rebuild_lba         %u\n", meta->rebuild_lba);
192 	printf("generation          %u\n", meta->generation);
193 	printf("status              0x%02x\n", meta->status);
194 	printf("type                %u\n", meta->type);
195 	printf("total_disks         %u\n", meta->total_disks);
196 	printf("stripe_shift        %u\n", meta->stripe_shift);
197 	printf("array_width         %u\n", meta->array_width);
198 	printf("array_number        %u\n", meta->array_number);
199 	printf("total_sectors       %u\n", meta->total_sectors);
200 	printf("cylinders           %u\n", meta->cylinders);
201 	printf("heads               %u\n", meta->heads);
202 	printf("sectors             %u\n", meta->sectors);
203 	printf("volume_id           0x%016jx\n", meta->volume_id);
204 	printf("disks:\n");
205 	for (i = 0; i < PROMISE_MAX_DISKS; i++ ) {
206 		printf("                    %02x %02x %02x %02x %016jx\n",
207 		    meta->disks[i].flags, meta->disks[i].number,
208 		    meta->disks[i].channel, meta->disks[i].device,
209 		    meta->disks[i].id);
210 	}
211 	printf("name                <%.32s>\n", meta->name);
212 	printf("magic_3             0x%08x\n", meta->magic_3);
213 	printf("rebuild_lba64       %ju\n", meta->rebuild_lba64);
214 	printf("magic_4             0x%08x\n", meta->magic_4);
215 	printf("magic_5             0x%08x\n", meta->magic_5);
216 	printf("=================================================\n");
217 }
218 
219 static struct promise_raid_conf *
220 promise_meta_copy(struct promise_raid_conf *meta)
221 {
222 	struct promise_raid_conf *nmeta;
223 
224 	nmeta = malloc(sizeof(*nmeta), M_MD_PROMISE, M_WAITOK);
225 	memcpy(nmeta, meta, sizeof(*nmeta));
226 	return (nmeta);
227 }
228 
229 static int
230 promise_meta_find_disk(struct promise_raid_conf *meta, uint64_t id)
231 {
232 	int pos;
233 
234 	for (pos = 0; pos < meta->total_disks; pos++) {
235 		if (meta->disks[pos].id == id)
236 			return (pos);
237 	}
238 	return (-1);
239 }
240 
241 static int
242 promise_meta_unused_range(struct promise_raid_conf **metaarr, int nsd,
243     uint32_t sectors, uint32_t *off, uint32_t *size)
244 {
245 	uint32_t coff, csize;
246 	int i, j;
247 
248 	sectors -= 131072;
249 	*off = 0;
250 	*size = 0;
251 	coff = 0;
252 	csize = sectors;
253 	i = 0;
254 	while (1) {
255 		for (j = 0; j < nsd; j++) {
256 			if (metaarr[j]->disk_offset >= coff) {
257 				csize = MIN(csize,
258 				    metaarr[j]->disk_offset - coff);
259 			}
260 		}
261 		if (csize > *size) {
262 			*off = coff;
263 			*size = csize;
264 		}
265 		if (i >= nsd)
266 			break;
267 		coff = metaarr[i]->disk_offset + metaarr[i]->disk_sectors;
268 		csize = sectors - coff;
269 		i++;
270 	};
271 	return ((*size > 0) ? 1 : 0);
272 }
273 
274 static int
275 promise_meta_translate_disk(struct g_raid_volume *vol, int md_disk_pos)
276 {
277 	int disk_pos, width;
278 
279 	if (md_disk_pos >= 0 && vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
280 		width = vol->v_disks_count / 2;
281 		disk_pos = (md_disk_pos / width) +
282 		    (md_disk_pos % width) * width;
283 	} else
284 		disk_pos = md_disk_pos;
285 	return (disk_pos);
286 }
287 
288 static void
289 promise_meta_get_name(struct promise_raid_conf *meta, char *buf)
290 {
291 	int i;
292 
293 	strncpy(buf, meta->name, 32);
294 	buf[32] = 0;
295 	for (i = 31; i >= 0; i--) {
296 		if (buf[i] > 0x20)
297 			break;
298 		buf[i] = 0;
299 	}
300 }
301 
302 static void
303 promise_meta_put_name(struct promise_raid_conf *meta, char *buf)
304 {
305 
306 	memset(meta->name, 0x20, 32);
307 	memcpy(meta->name, buf, MIN(strlen(buf), 32));
308 }
309 
310 static int
311 promise_meta_read(struct g_consumer *cp, struct promise_raid_conf **metaarr)
312 {
313 	struct g_provider *pp;
314 	struct promise_raid_conf *meta;
315 	char *buf;
316 	int error, i, subdisks;
317 	uint32_t checksum, *ptr;
318 
319 	pp = cp->provider;
320 	subdisks = 0;
321 next:
322 	/* Read metadata block. */
323 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize *
324 	    (63 - subdisks * PROMISE_META_OFFSET),
325 	    pp->sectorsize * 4, &error);
326 	if (buf == NULL) {
327 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
328 		    pp->name, error);
329 		return (subdisks);
330 	}
331 	meta = (struct promise_raid_conf *)buf;
332 
333 	/* Check if this is an Promise RAID struct */
334 	if (strncmp(meta->promise_id, PROMISE_MAGIC, strlen(PROMISE_MAGIC)) &&
335 	    strncmp(meta->promise_id, FREEBSD_MAGIC, strlen(FREEBSD_MAGIC))) {
336 		if (subdisks == 0)
337 			G_RAID_DEBUG(1,
338 			    "Promise signature check failed on %s", pp->name);
339 		g_free(buf);
340 		return (subdisks);
341 	}
342 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK);
343 	memcpy(meta, buf, MIN(sizeof(*meta), pp->sectorsize * 4));
344 	g_free(buf);
345 
346 	/* Check metadata checksum. */
347 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
348 		checksum += *ptr++;
349 	if (checksum != meta->checksum) {
350 		G_RAID_DEBUG(1, "Promise checksum check failed on %s", pp->name);
351 		free(meta, M_MD_PROMISE);
352 		return (subdisks);
353 	}
354 
355 	if ((meta->integrity & PROMISE_I_VALID) == 0) {
356 		G_RAID_DEBUG(1, "Promise metadata is invalid on %s", pp->name);
357 		free(meta, M_MD_PROMISE);
358 		return (subdisks);
359 	}
360 
361 	if (meta->total_disks > PROMISE_MAX_DISKS) {
362 		G_RAID_DEBUG(1, "Wrong number of disks on %s (%d)",
363 		    pp->name, meta->total_disks);
364 		free(meta, M_MD_PROMISE);
365 		return (subdisks);
366 	}
367 
368 	/* Save this part and look for next. */
369 	*metaarr = meta;
370 	metaarr++;
371 	subdisks++;
372 	if (subdisks < PROMISE_MAX_SUBDISKS)
373 		goto next;
374 
375 	return (subdisks);
376 }
377 
378 static int
379 promise_meta_write(struct g_consumer *cp,
380     struct promise_raid_conf **metaarr, int nsd)
381 {
382 	struct g_provider *pp;
383 	struct promise_raid_conf *meta;
384 	char *buf;
385 	int error, i, subdisk, fake;
386 	uint32_t checksum, *ptr, off, size;
387 
388 	pp = cp->provider;
389 	subdisk = 0;
390 	fake = 0;
391 next:
392 	buf = malloc(pp->sectorsize * 4, M_MD_PROMISE, M_WAITOK | M_ZERO);
393 	meta = NULL;
394 	if (subdisk < nsd) {
395 		meta = metaarr[subdisk];
396 	} else if (!fake && promise_meta_unused_range(metaarr, nsd,
397 	    cp->provider->mediasize / cp->provider->sectorsize,
398 	    &off, &size)) {
399 		/* Optionally add record for unused space. */
400 		meta = (struct promise_raid_conf *)buf;
401 		memcpy(&meta->promise_id[0], PROMISE_MAGIC,
402 		    sizeof(PROMISE_MAGIC) - 1);
403 		meta->dummy_0 = 0x00020000;
404 		meta->integrity = PROMISE_I_VALID;
405 		meta->disk.flags = PROMISE_F_ONLINE | PROMISE_F_VALID;
406 		meta->disk.number = 0xff;
407 		arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
408 		meta->disk_offset = off;
409 		meta->disk_sectors = size;
410 		meta->rebuild_lba = UINT32_MAX;
411 		fake = 1;
412 	}
413 	if (meta != NULL) {
414 		/* Recalculate checksum for case if metadata were changed. */
415 		meta->checksum = 0;
416 		for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
417 			checksum += *ptr++;
418 		meta->checksum = checksum;
419 		memcpy(buf, meta, MIN(pp->sectorsize * 4, sizeof(*meta)));
420 	}
421 	error = g_write_data(cp, pp->mediasize - pp->sectorsize *
422 	    (63 - subdisk * PROMISE_META_OFFSET),
423 	    buf, pp->sectorsize * 4);
424 	if (error != 0) {
425 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
426 		    pp->name, error);
427 	}
428 	free(buf, M_MD_PROMISE);
429 
430 	subdisk++;
431 	if (subdisk < PROMISE_MAX_SUBDISKS)
432 		goto next;
433 
434 	return (error);
435 }
436 
437 static int
438 promise_meta_erase(struct g_consumer *cp)
439 {
440 	struct g_provider *pp;
441 	char *buf;
442 	int error, subdisk;
443 
444 	pp = cp->provider;
445 	buf = malloc(4 * pp->sectorsize, M_MD_PROMISE, M_WAITOK | M_ZERO);
446 	for (subdisk = 0; subdisk < PROMISE_MAX_SUBDISKS; subdisk++) {
447 		error = g_write_data(cp, pp->mediasize - pp->sectorsize *
448 		    (63 - subdisk * PROMISE_META_OFFSET),
449 		    buf, 4 * pp->sectorsize);
450 		if (error != 0) {
451 			G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
452 			    pp->name, error);
453 		}
454 	}
455 	free(buf, M_MD_PROMISE);
456 	return (error);
457 }
458 
459 static int
460 promise_meta_write_spare(struct g_consumer *cp)
461 {
462 	struct promise_raid_conf *meta;
463 	int error;
464 
465 	meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
466 	memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC) - 1);
467 	meta->dummy_0 = 0x00020000;
468 	meta->integrity = PROMISE_I_VALID;
469 	meta->disk.flags = PROMISE_F_SPARE | PROMISE_F_ONLINE | PROMISE_F_VALID;
470 	meta->disk.number = 0xff;
471 	arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
472 	meta->disk_sectors = cp->provider->mediasize / cp->provider->sectorsize;
473 	meta->disk_sectors -= 131072;
474 	meta->rebuild_lba = UINT32_MAX;
475 	error = promise_meta_write(cp, &meta, 1);
476 	free(meta, M_MD_PROMISE);
477 	return (error);
478 }
479 
480 static struct g_raid_volume *
481 g_raid_md_promise_get_volume(struct g_raid_softc *sc, uint64_t id)
482 {
483 	struct g_raid_volume	*vol;
484 	struct g_raid_md_promise_pervolume *pv;
485 
486 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
487 		pv = vol->v_md_data;
488 		if (pv->pv_id == id)
489 			break;
490 	}
491 	return (vol);
492 }
493 
494 static int
495 g_raid_md_promise_purge_volumes(struct g_raid_softc *sc)
496 {
497 	struct g_raid_volume	*vol, *tvol;
498 	struct g_raid_md_promise_pervolume *pv;
499 	int i, res;
500 
501 	res = 0;
502 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tvol) {
503 		pv = vol->v_md_data;
504 		if (!pv->pv_started || vol->v_stopping)
505 			continue;
506 		for (i = 0; i < vol->v_disks_count; i++) {
507 			if (vol->v_subdisks[i].sd_state != G_RAID_SUBDISK_S_NONE)
508 				break;
509 		}
510 		if (i >= vol->v_disks_count) {
511 			g_raid_destroy_volume(vol);
512 			res = 1;
513 		}
514 	}
515 	return (res);
516 }
517 
518 static int
519 g_raid_md_promise_purge_disks(struct g_raid_softc *sc)
520 {
521 	struct g_raid_disk	*disk, *tdisk;
522 	struct g_raid_volume	*vol;
523 	struct g_raid_md_promise_perdisk *pd;
524 	int i, j, res;
525 
526 	res = 0;
527 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) {
528 		if (disk->d_state == G_RAID_DISK_S_SPARE)
529 			continue;
530 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
531 
532 		/* Scan for deleted volumes. */
533 		for (i = 0; i < pd->pd_subdisks; ) {
534 			vol = g_raid_md_promise_get_volume(sc,
535 			    pd->pd_meta[i]->volume_id);
536 			if (vol != NULL && !vol->v_stopping) {
537 				i++;
538 				continue;
539 			}
540 			free(pd->pd_meta[i], M_MD_PROMISE);
541 			for (j = i; j < pd->pd_subdisks - 1; j++)
542 				pd->pd_meta[j] = pd->pd_meta[j + 1];
543 			pd->pd_meta[PROMISE_MAX_SUBDISKS - 1] = NULL;
544 			pd->pd_subdisks--;
545 			pd->pd_updated = 1;
546 		}
547 
548 		/* If there is no metadata left - erase and delete disk. */
549 		if (pd->pd_subdisks == 0) {
550 			promise_meta_erase(disk->d_consumer);
551 			g_raid_destroy_disk(disk);
552 			res = 1;
553 		}
554 	}
555 	return (res);
556 }
557 
558 static int
559 g_raid_md_promise_supported(int level, int qual, int disks, int force)
560 {
561 
562 	if (disks > PROMISE_MAX_DISKS)
563 		return (0);
564 	switch (level) {
565 	case G_RAID_VOLUME_RL_RAID0:
566 		if (disks < 1)
567 			return (0);
568 		if (!force && disks < 2)
569 			return (0);
570 		break;
571 	case G_RAID_VOLUME_RL_RAID1:
572 		if (disks < 1)
573 			return (0);
574 		if (!force && (disks != 2))
575 			return (0);
576 		break;
577 	case G_RAID_VOLUME_RL_RAID1E:
578 		if (disks < 2)
579 			return (0);
580 		if (disks % 2 != 0)
581 			return (0);
582 		if (!force && (disks != 4))
583 			return (0);
584 		break;
585 	case G_RAID_VOLUME_RL_SINGLE:
586 		if (disks != 1)
587 			return (0);
588 		break;
589 	case G_RAID_VOLUME_RL_CONCAT:
590 		if (disks < 2)
591 			return (0);
592 		break;
593 	case G_RAID_VOLUME_RL_RAID5:
594 		if (disks < 3)
595 			return (0);
596 		break;
597 	default:
598 		return (0);
599 	}
600 	if (qual != G_RAID_VOLUME_RLQ_NONE)
601 		return (0);
602 	return (1);
603 }
604 
605 static int
606 g_raid_md_promise_start_disk(struct g_raid_disk *disk, int sdn,
607     struct g_raid_volume *vol)
608 {
609 	struct g_raid_softc *sc;
610 	struct g_raid_subdisk *sd;
611 	struct g_raid_md_promise_perdisk *pd;
612 	struct g_raid_md_promise_pervolume *pv;
613 	struct promise_raid_conf *meta;
614 	off_t size;
615 	int disk_pos, md_disk_pos, i, resurrection = 0;
616 	uint32_t eoff, esize;
617 
618 	sc = disk->d_softc;
619 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
620 
621 	pv = vol->v_md_data;
622 	meta = pv->pv_meta;
623 
624 	if (sdn >= 0) {
625 		/* Find disk position in metadata by it's serial. */
626 		md_disk_pos = promise_meta_find_disk(meta, pd->pd_meta[sdn]->disk.id);
627 		/* For RAID0+1 we need to translate order. */
628 		disk_pos = promise_meta_translate_disk(vol, md_disk_pos);
629 	} else {
630 		md_disk_pos = -1;
631 		disk_pos = -1;
632 	}
633 	if (disk_pos < 0) {
634 		G_RAID_DEBUG1(1, sc, "Disk %s is not part of the volume %s",
635 		    g_raid_get_diskname(disk), vol->v_name);
636 		/* Failed stale disk is useless for us. */
637 		if (sdn >= 0 &&
638 		    pd->pd_meta[sdn]->disk.flags & PROMISE_F_DOWN) {
639 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
640 			return (0);
641 		}
642 		/* If we were given specific metadata subdisk - erase it. */
643 		if (sdn >= 0) {
644 			free(pd->pd_meta[sdn], M_MD_PROMISE);
645 			for (i = sdn; i < pd->pd_subdisks - 1; i++)
646 				pd->pd_meta[i] = pd->pd_meta[i + 1];
647 			pd->pd_meta[PROMISE_MAX_SUBDISKS - 1] = NULL;
648 			pd->pd_subdisks--;
649 		}
650 		/* If we are in the start process, that's all for now. */
651 		if (!pv->pv_started)
652 			goto nofit;
653 		/*
654 		 * If we have already started - try to get use of the disk.
655 		 * Try to replace OFFLINE disks first, then FAILED.
656 		 */
657 		promise_meta_unused_range(pd->pd_meta, pd->pd_subdisks,
658 		    disk->d_consumer->provider->mediasize /
659 		    disk->d_consumer->provider->sectorsize,
660 		    &eoff, &esize);
661 		if (esize == 0) {
662 			G_RAID_DEBUG1(1, sc, "No free space on disk %s",
663 			    g_raid_get_diskname(disk));
664 			goto nofit;
665 		}
666 		size = INT64_MAX;
667 		for (i = 0; i < vol->v_disks_count; i++) {
668 			sd = &vol->v_subdisks[i];
669 			if (sd->sd_state != G_RAID_SUBDISK_S_NONE)
670 				size = sd->sd_size;
671 			if (sd->sd_state <= G_RAID_SUBDISK_S_FAILED &&
672 			    (disk_pos < 0 ||
673 			     vol->v_subdisks[i].sd_state < sd->sd_state))
674 				disk_pos = i;
675 		}
676 		if (disk_pos >= 0 &&
677 		    vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT &&
678 		    (off_t)esize * 512 < size) {
679 			G_RAID_DEBUG1(1, sc, "Disk %s free space "
680 			    "is too small (%ju < %ju)",
681 			    g_raid_get_diskname(disk),
682 			    (off_t)esize * 512, size);
683 			disk_pos = -1;
684 		}
685 		if (disk_pos >= 0) {
686 			if (vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT)
687 				esize = size / 512;
688 			/* For RAID0+1 we need to translate order. */
689 			md_disk_pos = promise_meta_translate_disk(vol, disk_pos);
690 		} else {
691 nofit:
692 			if (pd->pd_subdisks == 0) {
693 				g_raid_change_disk_state(disk,
694 				    G_RAID_DISK_S_SPARE);
695 			}
696 			return (0);
697 		}
698 		G_RAID_DEBUG1(1, sc, "Disk %s takes pos %d in the volume %s",
699 		    g_raid_get_diskname(disk), disk_pos, vol->v_name);
700 		resurrection = 1;
701 	}
702 
703 	sd = &vol->v_subdisks[disk_pos];
704 
705 	if (resurrection && sd->sd_disk != NULL) {
706 		g_raid_change_disk_state(sd->sd_disk,
707 		    G_RAID_DISK_S_STALE_FAILED);
708 		TAILQ_REMOVE(&sd->sd_disk->d_subdisks,
709 		    sd, sd_next);
710 	}
711 	vol->v_subdisks[disk_pos].sd_disk = disk;
712 	TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
713 
714 	/* Welcome the new disk. */
715 	if (resurrection)
716 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
717 	else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN)
718 		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
719 	else
720 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
721 
722 	if (resurrection) {
723 		sd->sd_offset = (off_t)eoff * 512;
724 		sd->sd_size = (off_t)esize * 512;
725 	} else {
726 		sd->sd_offset = (off_t)pd->pd_meta[sdn]->disk_offset * 512;
727 		sd->sd_size = (off_t)pd->pd_meta[sdn]->disk_sectors * 512;
728 	}
729 
730 	if (resurrection) {
731 		/* Stale disk, almost same as new. */
732 		g_raid_change_subdisk_state(sd,
733 		    G_RAID_SUBDISK_S_NEW);
734 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) {
735 		/* Failed disk. */
736 		g_raid_change_subdisk_state(sd,
737 		    G_RAID_SUBDISK_S_FAILED);
738 	} else if (meta->disks[md_disk_pos].flags & PROMISE_F_REDIR) {
739 		/* Rebuilding disk. */
740 		g_raid_change_subdisk_state(sd,
741 		    G_RAID_SUBDISK_S_REBUILD);
742 		if (pd->pd_meta[sdn]->generation != meta->generation)
743 			sd->sd_rebuild_pos = 0;
744 		else {
745 			sd->sd_rebuild_pos =
746 			    (off_t)pd->pd_meta[sdn]->rebuild_lba * 512;
747 		}
748 	} else if (!(meta->disks[md_disk_pos].flags & PROMISE_F_ONLINE)) {
749 		/* Rebuilding disk. */
750 		g_raid_change_subdisk_state(sd,
751 		    G_RAID_SUBDISK_S_NEW);
752 	} else if (pd->pd_meta[sdn]->generation != meta->generation ||
753 	    (meta->status & PROMISE_S_MARKED)) {
754 		/* Stale disk or dirty volume (unclean shutdown). */
755 		g_raid_change_subdisk_state(sd,
756 		    G_RAID_SUBDISK_S_STALE);
757 	} else {
758 		/* Up to date disk. */
759 		g_raid_change_subdisk_state(sd,
760 		    G_RAID_SUBDISK_S_ACTIVE);
761 	}
762 	g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
763 	    G_RAID_EVENT_SUBDISK);
764 
765 	return (resurrection);
766 }
767 
768 static void
769 g_raid_md_promise_refill(struct g_raid_softc *sc)
770 {
771 	struct g_raid_volume *vol;
772 	struct g_raid_subdisk *sd;
773 	struct g_raid_disk *disk;
774 	struct g_raid_md_object *md;
775 	struct g_raid_md_promise_perdisk *pd;
776 	struct g_raid_md_promise_pervolume *pv;
777 	int update, updated, i, bad;
778 
779 	md = sc->sc_md;
780 restart:
781 	updated = 0;
782 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
783 		pv = vol->v_md_data;
784 		if (!pv->pv_started || vol->v_stopping)
785 			continue;
786 
787 		/* Search for subdisk that needs replacement. */
788 		bad = 0;
789 		for (i = 0; i < vol->v_disks_count; i++) {
790 			sd = &vol->v_subdisks[i];
791 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
792 			    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
793 			        bad = 1;
794 		}
795 		if (!bad)
796 			continue;
797 
798 		G_RAID_DEBUG1(1, sc, "Volume %s is not complete, "
799 		    "trying to refill.", vol->v_name);
800 
801 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
802 			/* Skip failed. */
803 			if (disk->d_state < G_RAID_DISK_S_SPARE)
804 				continue;
805 			/* Skip already used by this volume. */
806 			for (i = 0; i < vol->v_disks_count; i++) {
807 				sd = &vol->v_subdisks[i];
808 				if (sd->sd_disk == disk)
809 					break;
810 			}
811 			if (i < vol->v_disks_count)
812 				continue;
813 
814 			/* Try to use disk if it has empty extents. */
815 			pd = disk->d_md_data;
816 			if (pd->pd_subdisks < PROMISE_MAX_SUBDISKS) {
817 				update =
818 				    g_raid_md_promise_start_disk(disk, -1, vol);
819 			} else
820 				update = 0;
821 			if (update) {
822 				updated = 1;
823 				g_raid_md_write_promise(md, vol, NULL, disk);
824 				break;
825 			}
826 		}
827 	}
828 	if (updated)
829 		goto restart;
830 }
831 
832 static void
833 g_raid_md_promise_start(struct g_raid_volume *vol)
834 {
835 	struct g_raid_softc *sc;
836 	struct g_raid_subdisk *sd;
837 	struct g_raid_disk *disk;
838 	struct g_raid_md_object *md;
839 	struct g_raid_md_promise_perdisk *pd;
840 	struct g_raid_md_promise_pervolume *pv;
841 	struct promise_raid_conf *meta;
842 	int i;
843 
844 	sc = vol->v_softc;
845 	md = sc->sc_md;
846 	pv = vol->v_md_data;
847 	meta = pv->pv_meta;
848 
849 	if (meta->type == PROMISE_T_RAID0)
850 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
851 	else if (meta->type == PROMISE_T_RAID1) {
852 		if (meta->array_width == 1)
853 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
854 		else
855 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
856 	} else if (meta->type == PROMISE_T_RAID3)
857 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID3;
858 	else if (meta->type == PROMISE_T_RAID5)
859 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
860 	else if (meta->type == PROMISE_T_SPAN)
861 		vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
862 	else if (meta->type == PROMISE_T_JBOD)
863 		vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
864 	else
865 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
866 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
867 	vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ
868 	vol->v_disks_count = meta->total_disks;
869 	vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ
870 	vol->v_sectorsize = 512; //ZZZ
871 	for (i = 0; i < vol->v_disks_count; i++) {
872 		sd = &vol->v_subdisks[i];
873 		sd->sd_offset = (off_t)meta->disk_offset * 512; //ZZZ
874 		sd->sd_size = (off_t)meta->disk_sectors * 512; //ZZZ
875 	}
876 	g_raid_start_volume(vol);
877 
878 	/* Make all disks found till the moment take their places. */
879 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
880 		pd = disk->d_md_data;
881 		for (i = 0; i < pd->pd_subdisks; i++) {
882 			if (pd->pd_meta[i]->volume_id == meta->volume_id)
883 				g_raid_md_promise_start_disk(disk, i, vol);
884 		}
885 	}
886 
887 	pv->pv_started = 1;
888 	callout_stop(&pv->pv_start_co);
889 	G_RAID_DEBUG1(0, sc, "Volume started.");
890 	g_raid_md_write_promise(md, vol, NULL, NULL);
891 
892 	/* Pickup any STALE/SPARE disks to refill array if needed. */
893 	g_raid_md_promise_refill(sc);
894 
895 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
896 }
897 
898 static void
899 g_raid_promise_go(void *arg)
900 {
901 	struct g_raid_volume *vol;
902 	struct g_raid_softc *sc;
903 	struct g_raid_md_promise_pervolume *pv;
904 
905 	vol = arg;
906 	pv = vol->v_md_data;
907 	sc = vol->v_softc;
908 	if (!pv->pv_started) {
909 		G_RAID_DEBUG1(0, sc, "Force volume start due to timeout.");
910 		g_raid_event_send(vol, G_RAID_VOLUME_E_STARTMD,
911 		    G_RAID_EVENT_VOLUME);
912 	}
913 }
914 
915 static void
916 g_raid_md_promise_new_disk(struct g_raid_disk *disk)
917 {
918 	struct g_raid_softc *sc;
919 	struct g_raid_md_object *md;
920 	struct promise_raid_conf *pdmeta;
921 	struct g_raid_md_promise_perdisk *pd;
922 	struct g_raid_md_promise_pervolume *pv;
923 	struct g_raid_volume *vol;
924 	int i;
925 	char buf[33];
926 
927 	sc = disk->d_softc;
928 	md = sc->sc_md;
929 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
930 
931 	if (pd->pd_subdisks == 0) {
932 		g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
933 		g_raid_md_promise_refill(sc);
934 		return;
935 	}
936 
937 	for (i = 0; i < pd->pd_subdisks; i++) {
938 		pdmeta = pd->pd_meta[i];
939 
940 		/* Look for volume with matching ID. */
941 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
942 		if (vol == NULL) {
943 			promise_meta_get_name(pdmeta, buf);
944 			vol = g_raid_create_volume(sc, buf, pdmeta->array_number);
945 			pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
946 			pv->pv_id = pdmeta->volume_id;
947 			vol->v_md_data = pv;
948 			callout_init(&pv->pv_start_co, 1);
949 			callout_reset(&pv->pv_start_co,
950 			    g_raid_start_timeout * hz,
951 			    g_raid_promise_go, vol);
952 		} else
953 			pv = vol->v_md_data;
954 
955 		/* If we haven't started yet - check metadata freshness. */
956 		if (pv->pv_meta == NULL || !pv->pv_started) {
957 			if (pv->pv_meta == NULL ||
958 			    ((int16_t)(pdmeta->generation - pv->pv_generation)) > 0) {
959 				G_RAID_DEBUG1(1, sc, "Newer disk");
960 				if (pv->pv_meta != NULL)
961 					free(pv->pv_meta, M_MD_PROMISE);
962 				pv->pv_meta = promise_meta_copy(pdmeta);
963 				pv->pv_generation = pv->pv_meta->generation;
964 				pv->pv_disks_present = 1;
965 			} else if (pdmeta->generation == pv->pv_generation) {
966 				pv->pv_disks_present++;
967 				G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
968 				    pv->pv_disks_present,
969 				    pv->pv_meta->total_disks);
970 			} else {
971 				G_RAID_DEBUG1(1, sc, "Older disk");
972 			}
973 		}
974 	}
975 
976 	for (i = 0; i < pd->pd_subdisks; i++) {
977 		pdmeta = pd->pd_meta[i];
978 
979 		/* Look for volume with matching ID. */
980 		vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
981 		if (vol == NULL)
982 			continue;
983 		pv = vol->v_md_data;
984 
985 		if (pv->pv_started) {
986 			if (g_raid_md_promise_start_disk(disk, i, vol))
987 				g_raid_md_write_promise(md, vol, NULL, NULL);
988 		} else {
989 			/* If we collected all needed disks - start array. */
990 			if (pv->pv_disks_present == pv->pv_meta->total_disks)
991 				g_raid_md_promise_start(vol);
992 		}
993 	}
994 }
995 
996 static int
997 g_raid_md_create_promise(struct g_raid_md_object *md, struct g_class *mp,
998     struct g_geom **gp)
999 {
1000 	struct g_geom *geom;
1001 	struct g_raid_softc *sc;
1002 
1003 	/* Search for existing node. */
1004 	LIST_FOREACH(geom, &mp->geom, geom) {
1005 		sc = geom->softc;
1006 		if (sc == NULL)
1007 			continue;
1008 		if (sc->sc_stopping != 0)
1009 			continue;
1010 		if (sc->sc_md->mdo_class != md->mdo_class)
1011 			continue;
1012 		break;
1013 	}
1014 	if (geom != NULL) {
1015 		*gp = geom;
1016 		return (G_RAID_MD_TASTE_EXISTING);
1017 	}
1018 
1019 	/* Create new one if not found. */
1020 	sc = g_raid_create_node(mp, "Promise", md);
1021 	if (sc == NULL)
1022 		return (G_RAID_MD_TASTE_FAIL);
1023 	md->mdo_softc = sc;
1024 	*gp = sc->sc_geom;
1025 	return (G_RAID_MD_TASTE_NEW);
1026 }
1027 
1028 static int
1029 g_raid_md_taste_promise(struct g_raid_md_object *md, struct g_class *mp,
1030                               struct g_consumer *cp, struct g_geom **gp)
1031 {
1032 	struct g_consumer *rcp;
1033 	struct g_provider *pp;
1034 	struct g_raid_softc *sc;
1035 	struct g_raid_disk *disk;
1036 	struct promise_raid_conf *meta, *metaarr[4];
1037 	struct g_raid_md_promise_perdisk *pd;
1038 	struct g_geom *geom;
1039 	int error, i, j, result, len, subdisks;
1040 	char name[16];
1041 	uint16_t vendor;
1042 
1043 	G_RAID_DEBUG(1, "Tasting Promise on %s", cp->provider->name);
1044 	pp = cp->provider;
1045 
1046 	/* Read metadata from device. */
1047 	meta = NULL;
1048 	vendor = 0xffff;
1049 	if (g_access(cp, 1, 0, 0) != 0)
1050 		return (G_RAID_MD_TASTE_FAIL);
1051 	g_topology_unlock();
1052 	len = 2;
1053 	if (pp->geom->rank == 1)
1054 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1055 	subdisks = promise_meta_read(cp, metaarr);
1056 	g_topology_lock();
1057 	g_access(cp, -1, 0, 0);
1058 	if (subdisks == 0) {
1059 		if (g_raid_aggressive_spare) {
1060 			if (vendor == 0x105a || vendor == 0x1002) {
1061 				G_RAID_DEBUG(1,
1062 				    "No Promise metadata, forcing spare.");
1063 				goto search;
1064 			} else {
1065 				G_RAID_DEBUG(1,
1066 				    "Promise/ATI vendor mismatch "
1067 				    "0x%04x != 0x105a/0x1002",
1068 				    vendor);
1069 			}
1070 		}
1071 		return (G_RAID_MD_TASTE_FAIL);
1072 	}
1073 
1074 	/* Metadata valid. Print it. */
1075 	for (i = 0; i < subdisks; i++)
1076 		g_raid_md_promise_print(metaarr[i]);
1077 
1078 	/* Purge meaningless (empty/spare) records. */
1079 	for (i = 0; i < subdisks; ) {
1080 		if (metaarr[i]->disk.flags & PROMISE_F_ASSIGNED) {
1081 			i++;
1082 			continue;
1083 		}
1084 		free(metaarr[i], M_MD_PROMISE);
1085 		for (j = i; j < subdisks - 1; j++)
1086 			metaarr[i] = metaarr[j + 1];
1087 		metaarr[PROMISE_MAX_SUBDISKS - 1] = NULL;
1088 		subdisks--;
1089 	}
1090 
1091 search:
1092 	/* Search for matching node. */
1093 	sc = NULL;
1094 	LIST_FOREACH(geom, &mp->geom, geom) {
1095 		sc = geom->softc;
1096 		if (sc == NULL)
1097 			continue;
1098 		if (sc->sc_stopping != 0)
1099 			continue;
1100 		if (sc->sc_md->mdo_class != md->mdo_class)
1101 			continue;
1102 		break;
1103 	}
1104 
1105 	/* Found matching node. */
1106 	if (geom != NULL) {
1107 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1108 		result = G_RAID_MD_TASTE_EXISTING;
1109 
1110 	} else { /* Not found matching node -- create one. */
1111 		result = G_RAID_MD_TASTE_NEW;
1112 		snprintf(name, sizeof(name), "Promise");
1113 		sc = g_raid_create_node(mp, name, md);
1114 		md->mdo_softc = sc;
1115 		geom = sc->sc_geom;
1116 	}
1117 
1118 	rcp = g_new_consumer(geom);
1119 	g_attach(rcp, pp);
1120 	if (g_access(rcp, 1, 1, 1) != 0)
1121 		; //goto fail1;
1122 
1123 	g_topology_unlock();
1124 	sx_xlock(&sc->sc_lock);
1125 
1126 	pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1127 	pd->pd_subdisks = subdisks;
1128 	for (i = 0; i < subdisks; i++)
1129 		pd->pd_meta[i] = metaarr[i];
1130 	disk = g_raid_create_disk(sc);
1131 	disk->d_md_data = (void *)pd;
1132 	disk->d_consumer = rcp;
1133 	rcp->private = disk;
1134 
1135 	/* Read kernel dumping information. */
1136 	disk->d_kd.offset = 0;
1137 	disk->d_kd.length = OFF_MAX;
1138 	len = sizeof(disk->d_kd);
1139 	error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd);
1140 	if (disk->d_kd.di.dumper == NULL)
1141 		G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.",
1142 		    rcp->provider->name, error);
1143 
1144 	g_raid_md_promise_new_disk(disk);
1145 
1146 	sx_xunlock(&sc->sc_lock);
1147 	g_topology_lock();
1148 	*gp = geom;
1149 	return (result);
1150 }
1151 
1152 static int
1153 g_raid_md_event_promise(struct g_raid_md_object *md,
1154     struct g_raid_disk *disk, u_int event)
1155 {
1156 	struct g_raid_softc *sc;
1157 
1158 	sc = md->mdo_softc;
1159 	if (disk == NULL)
1160 		return (-1);
1161 	switch (event) {
1162 	case G_RAID_DISK_E_DISCONNECTED:
1163 		/* Delete disk. */
1164 		g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1165 		g_raid_destroy_disk(disk);
1166 		g_raid_md_promise_purge_volumes(sc);
1167 
1168 		/* Write updated metadata to all disks. */
1169 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1170 
1171 		/* Check if anything left. */
1172 		if (g_raid_ndisks(sc, -1) == 0)
1173 			g_raid_destroy_node(sc, 0);
1174 		else
1175 			g_raid_md_promise_refill(sc);
1176 		return (0);
1177 	}
1178 	return (-2);
1179 }
1180 
1181 static int
1182 g_raid_md_volume_event_promise(struct g_raid_md_object *md,
1183     struct g_raid_volume *vol, u_int event)
1184 {
1185 	struct g_raid_md_promise_pervolume *pv;
1186 
1187 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1188 	switch (event) {
1189 	case G_RAID_VOLUME_E_STARTMD:
1190 		if (!pv->pv_started)
1191 			g_raid_md_promise_start(vol);
1192 		return (0);
1193 	}
1194 	return (-2);
1195 }
1196 
1197 static int
1198 g_raid_md_ctl_promise(struct g_raid_md_object *md,
1199     struct gctl_req *req)
1200 {
1201 	struct g_raid_softc *sc;
1202 	struct g_raid_volume *vol, *vol1;
1203 	struct g_raid_subdisk *sd;
1204 	struct g_raid_disk *disk, *disks[PROMISE_MAX_DISKS];
1205 	struct g_raid_md_promise_perdisk *pd;
1206 	struct g_raid_md_promise_pervolume *pv;
1207 	struct g_consumer *cp;
1208 	struct g_provider *pp;
1209 	char arg[16];
1210 	const char *verb, *volname, *levelname, *diskname;
1211 	char *tmp;
1212 	int *nargs, *force;
1213 	off_t size, sectorsize, strip;
1214 	intmax_t *sizearg, *striparg;
1215 	uint32_t offs[PROMISE_MAX_DISKS], esize;
1216 	int numdisks, i, len, level, qual;
1217 	int error;
1218 
1219 	sc = md->mdo_softc;
1220 	verb = gctl_get_param(req, "verb", NULL);
1221 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1222 	error = 0;
1223 	if (strcmp(verb, "label") == 0) {
1224 
1225 		if (*nargs < 4) {
1226 			gctl_error(req, "Invalid number of arguments.");
1227 			return (-1);
1228 		}
1229 		volname = gctl_get_asciiparam(req, "arg1");
1230 		if (volname == NULL) {
1231 			gctl_error(req, "No volume name.");
1232 			return (-2);
1233 		}
1234 		levelname = gctl_get_asciiparam(req, "arg2");
1235 		if (levelname == NULL) {
1236 			gctl_error(req, "No RAID level.");
1237 			return (-3);
1238 		}
1239 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1240 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1241 			return (-4);
1242 		}
1243 		numdisks = *nargs - 3;
1244 		force = gctl_get_paraml(req, "force", sizeof(*force));
1245 		if (!g_raid_md_promise_supported(level, qual, numdisks,
1246 		    force ? *force : 0)) {
1247 			gctl_error(req, "Unsupported RAID level "
1248 			    "(0x%02x/0x%02x), or number of disks (%d).",
1249 			    level, qual, numdisks);
1250 			return (-5);
1251 		}
1252 
1253 		/* Search for disks, connect them and probe. */
1254 		size = INT64_MAX;
1255 		sectorsize = 0;
1256 		bzero(disks, sizeof(disks));
1257 		bzero(offs, sizeof(offs));
1258 		for (i = 0; i < numdisks; i++) {
1259 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1260 			diskname = gctl_get_asciiparam(req, arg);
1261 			if (diskname == NULL) {
1262 				gctl_error(req, "No disk name (%s).", arg);
1263 				error = -6;
1264 				break;
1265 			}
1266 			if (strcmp(diskname, "NONE") == 0)
1267 				continue;
1268 
1269 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1270 				if (disk->d_consumer != NULL &&
1271 				    disk->d_consumer->provider != NULL &&
1272 				    strcmp(disk->d_consumer->provider->name,
1273 				     diskname) == 0)
1274 					break;
1275 			}
1276 			if (disk != NULL) {
1277 				if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
1278 					gctl_error(req, "Disk '%s' is in a "
1279 					    "wrong state (%s).", diskname,
1280 					    g_raid_disk_state2str(disk->d_state));
1281 					error = -7;
1282 					break;
1283 				}
1284 				pd = disk->d_md_data;
1285 				if (pd->pd_subdisks >= PROMISE_MAX_SUBDISKS) {
1286 					gctl_error(req, "Disk '%s' already "
1287 					    "used by %d volumes.",
1288 					    diskname, pd->pd_subdisks);
1289 					error = -7;
1290 					break;
1291 				}
1292 				pp = disk->d_consumer->provider;
1293 				disks[i] = disk;
1294 				promise_meta_unused_range(pd->pd_meta,
1295 				    pd->pd_subdisks,
1296 				    pp->mediasize / pp->sectorsize,
1297 				    &offs[i], &esize);
1298 				size = MIN(size, (off_t)esize * pp->sectorsize);
1299 				sectorsize = MAX(sectorsize, pp->sectorsize);
1300 				continue;
1301 			}
1302 
1303 			g_topology_lock();
1304 			cp = g_raid_open_consumer(sc, diskname);
1305 			if (cp == NULL) {
1306 				gctl_error(req, "Can't open disk '%s'.",
1307 				    diskname);
1308 				g_topology_unlock();
1309 				error = -8;
1310 				break;
1311 			}
1312 			pp = cp->provider;
1313 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1314 			disk = g_raid_create_disk(sc);
1315 			disk->d_md_data = (void *)pd;
1316 			disk->d_consumer = cp;
1317 			disks[i] = disk;
1318 			cp->private = disk;
1319 			g_topology_unlock();
1320 
1321 			/* Read kernel dumping information. */
1322 			disk->d_kd.offset = 0;
1323 			disk->d_kd.length = OFF_MAX;
1324 			len = sizeof(disk->d_kd);
1325 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1326 			if (disk->d_kd.di.dumper == NULL)
1327 				G_RAID_DEBUG1(2, sc,
1328 				    "Dumping not supported by %s.",
1329 				    cp->provider->name);
1330 
1331 			/* Reserve some space for metadata. */
1332 			size = MIN(size, pp->mediasize - 131072llu * pp->sectorsize);
1333 			sectorsize = MAX(sectorsize, pp->sectorsize);
1334 		}
1335 		if (error != 0) {
1336 			for (i = 0; i < numdisks; i++) {
1337 				if (disks[i] != NULL &&
1338 				    disks[i]->d_state == G_RAID_DISK_S_NONE)
1339 					g_raid_destroy_disk(disks[i]);
1340 			}
1341 			return (error);
1342 		}
1343 
1344 		if (sectorsize <= 0) {
1345 			gctl_error(req, "Can't get sector size.");
1346 			return (-8);
1347 		}
1348 
1349 		/* Handle size argument. */
1350 		len = sizeof(*sizearg);
1351 		sizearg = gctl_get_param(req, "size", &len);
1352 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1353 		    *sizearg > 0) {
1354 			if (*sizearg > size) {
1355 				gctl_error(req, "Size too big %lld > %lld.",
1356 				    (long long)*sizearg, (long long)size);
1357 				return (-9);
1358 			}
1359 			size = *sizearg;
1360 		}
1361 
1362 		/* Handle strip argument. */
1363 		strip = 131072;
1364 		len = sizeof(*striparg);
1365 		striparg = gctl_get_param(req, "strip", &len);
1366 		if (striparg != NULL && len == sizeof(*striparg) &&
1367 		    *striparg > 0) {
1368 			if (*striparg < sectorsize) {
1369 				gctl_error(req, "Strip size too small.");
1370 				return (-10);
1371 			}
1372 			if (*striparg % sectorsize != 0) {
1373 				gctl_error(req, "Incorrect strip size.");
1374 				return (-11);
1375 			}
1376 			strip = *striparg;
1377 		}
1378 
1379 		/* Round size down to strip or sector. */
1380 		if (level == G_RAID_VOLUME_RL_RAID1 ||
1381 		    level == G_RAID_VOLUME_RL_SINGLE ||
1382 		    level == G_RAID_VOLUME_RL_CONCAT)
1383 			size -= (size % sectorsize);
1384 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1385 		    (numdisks & 1) != 0)
1386 			size -= (size % (2 * strip));
1387 		else
1388 			size -= (size % strip);
1389 		if (size <= 0) {
1390 			gctl_error(req, "Size too small.");
1391 			return (-13);
1392 		}
1393 		if (size > 0xffffffffllu * sectorsize) {
1394 			gctl_error(req, "Size too big.");
1395 			return (-14);
1396 		}
1397 
1398 		/* We have all we need, create things: volume, ... */
1399 		pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1400 		arc4rand(&pv->pv_id, sizeof(pv->pv_id), 0);
1401 		pv->pv_generation = 0;
1402 		pv->pv_started = 1;
1403 		vol = g_raid_create_volume(sc, volname, -1);
1404 		vol->v_md_data = pv;
1405 		vol->v_raid_level = level;
1406 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
1407 		vol->v_strip_size = strip;
1408 		vol->v_disks_count = numdisks;
1409 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1410 		    level == G_RAID_VOLUME_RL_CONCAT ||
1411 		    level == G_RAID_VOLUME_RL_SINGLE)
1412 			vol->v_mediasize = size * numdisks;
1413 		else if (level == G_RAID_VOLUME_RL_RAID1)
1414 			vol->v_mediasize = size;
1415 		else if (level == G_RAID_VOLUME_RL_RAID3 ||
1416 		    level == G_RAID_VOLUME_RL_RAID5)
1417 			vol->v_mediasize = size * (numdisks - 1);
1418 		else { /* RAID1E */
1419 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1420 			    strip;
1421 		}
1422 		vol->v_sectorsize = sectorsize;
1423 		g_raid_start_volume(vol);
1424 
1425 		/* , and subdisks. */
1426 		for (i = 0; i < numdisks; i++) {
1427 			disk = disks[i];
1428 			sd = &vol->v_subdisks[i];
1429 			sd->sd_disk = disk;
1430 			sd->sd_offset = (off_t)offs[i] * 512;
1431 			sd->sd_size = size;
1432 			if (disk == NULL)
1433 				continue;
1434 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1435 			g_raid_change_disk_state(disk,
1436 			    G_RAID_DISK_S_ACTIVE);
1437 			g_raid_change_subdisk_state(sd,
1438 			    G_RAID_SUBDISK_S_ACTIVE);
1439 			g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1440 			    G_RAID_EVENT_SUBDISK);
1441 		}
1442 
1443 		/* Write metadata based on created entities. */
1444 		G_RAID_DEBUG1(0, sc, "Array started.");
1445 		g_raid_md_write_promise(md, vol, NULL, NULL);
1446 
1447 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1448 		g_raid_md_promise_refill(sc);
1449 
1450 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1451 		    G_RAID_EVENT_VOLUME);
1452 		return (0);
1453 	}
1454 	if (strcmp(verb, "add") == 0) {
1455 
1456 		gctl_error(req, "`add` command is not applicable, "
1457 		    "use `label` instead.");
1458 		return (-99);
1459 	}
1460 	if (strcmp(verb, "delete") == 0) {
1461 
1462 		/* Full node destruction. */
1463 		if (*nargs == 1) {
1464 			/* Check if some volume is still open. */
1465 			force = gctl_get_paraml(req, "force", sizeof(*force));
1466 			if (force != NULL && *force == 0 &&
1467 			    g_raid_nopens(sc) != 0) {
1468 				gctl_error(req, "Some volume is still open.");
1469 				return (-4);
1470 			}
1471 
1472 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1473 				if (disk->d_consumer)
1474 					promise_meta_erase(disk->d_consumer);
1475 			}
1476 			g_raid_destroy_node(sc, 0);
1477 			return (0);
1478 		}
1479 
1480 		/* Destroy specified volume. If it was last - all node. */
1481 		if (*nargs != 2) {
1482 			gctl_error(req, "Invalid number of arguments.");
1483 			return (-1);
1484 		}
1485 		volname = gctl_get_asciiparam(req, "arg1");
1486 		if (volname == NULL) {
1487 			gctl_error(req, "No volume name.");
1488 			return (-2);
1489 		}
1490 
1491 		/* Search for volume. */
1492 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1493 			if (strcmp(vol->v_name, volname) == 0)
1494 				break;
1495 		}
1496 		if (vol == NULL) {
1497 			i = strtol(volname, &tmp, 10);
1498 			if (verb != volname && tmp[0] == 0) {
1499 				TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1500 					if (vol->v_global_id == i)
1501 						break;
1502 				}
1503 			}
1504 		}
1505 		if (vol == NULL) {
1506 			gctl_error(req, "Volume '%s' not found.", volname);
1507 			return (-3);
1508 		}
1509 
1510 		/* Check if volume is still open. */
1511 		force = gctl_get_paraml(req, "force", sizeof(*force));
1512 		if (force != NULL && *force == 0 &&
1513 		    vol->v_provider_open != 0) {
1514 			gctl_error(req, "Volume is still open.");
1515 			return (-4);
1516 		}
1517 
1518 		/* Destroy volume and potentially node. */
1519 		i = 0;
1520 		TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
1521 			i++;
1522 		if (i >= 2) {
1523 			g_raid_destroy_volume(vol);
1524 			g_raid_md_promise_purge_disks(sc);
1525 			g_raid_md_write_promise(md, NULL, NULL, NULL);
1526 		} else {
1527 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1528 				if (disk->d_consumer)
1529 					promise_meta_erase(disk->d_consumer);
1530 			}
1531 			g_raid_destroy_node(sc, 0);
1532 		}
1533 		return (0);
1534 	}
1535 	if (strcmp(verb, "remove") == 0 ||
1536 	    strcmp(verb, "fail") == 0) {
1537 		if (*nargs < 2) {
1538 			gctl_error(req, "Invalid number of arguments.");
1539 			return (-1);
1540 		}
1541 		for (i = 1; i < *nargs; i++) {
1542 			snprintf(arg, sizeof(arg), "arg%d", i);
1543 			diskname = gctl_get_asciiparam(req, arg);
1544 			if (diskname == NULL) {
1545 				gctl_error(req, "No disk name (%s).", arg);
1546 				error = -2;
1547 				break;
1548 			}
1549 			if (strncmp(diskname, "/dev/", 5) == 0)
1550 				diskname += 5;
1551 
1552 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1553 				if (disk->d_consumer != NULL &&
1554 				    disk->d_consumer->provider != NULL &&
1555 				    strcmp(disk->d_consumer->provider->name,
1556 				     diskname) == 0)
1557 					break;
1558 			}
1559 			if (disk == NULL) {
1560 				gctl_error(req, "Disk '%s' not found.",
1561 				    diskname);
1562 				error = -3;
1563 				break;
1564 			}
1565 
1566 			if (strcmp(verb, "fail") == 0) {
1567 				g_raid_md_fail_disk_promise(md, NULL, disk);
1568 				continue;
1569 			}
1570 
1571 			/* Erase metadata on deleting disk and destroy it. */
1572 			promise_meta_erase(disk->d_consumer);
1573 			g_raid_destroy_disk(disk);
1574 		}
1575 		g_raid_md_promise_purge_volumes(sc);
1576 
1577 		/* Write updated metadata to remaining disks. */
1578 		g_raid_md_write_promise(md, NULL, NULL, NULL);
1579 
1580 		/* Check if anything left. */
1581 		if (g_raid_ndisks(sc, -1) == 0)
1582 			g_raid_destroy_node(sc, 0);
1583 		else
1584 			g_raid_md_promise_refill(sc);
1585 		return (error);
1586 	}
1587 	if (strcmp(verb, "insert") == 0) {
1588 		if (*nargs < 2) {
1589 			gctl_error(req, "Invalid number of arguments.");
1590 			return (-1);
1591 		}
1592 		for (i = 1; i < *nargs; i++) {
1593 			/* Get disk name. */
1594 			snprintf(arg, sizeof(arg), "arg%d", i);
1595 			diskname = gctl_get_asciiparam(req, arg);
1596 			if (diskname == NULL) {
1597 				gctl_error(req, "No disk name (%s).", arg);
1598 				error = -3;
1599 				break;
1600 			}
1601 
1602 			/* Try to find provider with specified name. */
1603 			g_topology_lock();
1604 			cp = g_raid_open_consumer(sc, diskname);
1605 			if (cp == NULL) {
1606 				gctl_error(req, "Can't open disk '%s'.",
1607 				    diskname);
1608 				g_topology_unlock();
1609 				error = -4;
1610 				break;
1611 			}
1612 			g_topology_unlock();
1613 
1614 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1615 
1616 			disk = g_raid_create_disk(sc);
1617 			disk->d_consumer = cp;
1618 			disk->d_md_data = (void *)pd;
1619 			cp->private = disk;
1620 
1621 			/* Read kernel dumping information. */
1622 			disk->d_kd.offset = 0;
1623 			disk->d_kd.length = OFF_MAX;
1624 			len = sizeof(disk->d_kd);
1625 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1626 			if (disk->d_kd.di.dumper == NULL)
1627 				G_RAID_DEBUG1(2, sc,
1628 				    "Dumping not supported by %s.",
1629 				    cp->provider->name);
1630 
1631 			/* Welcome the "new" disk. */
1632 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1633 			promise_meta_write_spare(cp);
1634 			g_raid_md_promise_refill(sc);
1635 		}
1636 		return (error);
1637 	}
1638 	return (-100);
1639 }
1640 
1641 static int
1642 g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1643     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1644 {
1645 	struct g_raid_softc *sc;
1646 	struct g_raid_volume *vol;
1647 	struct g_raid_subdisk *sd;
1648 	struct g_raid_disk *disk;
1649 	struct g_raid_md_promise_perdisk *pd;
1650 	struct g_raid_md_promise_pervolume *pv;
1651 	struct promise_raid_conf *meta;
1652 	off_t rebuild_lba64;
1653 	int i, j, pos, rebuild;
1654 
1655 	sc = md->mdo_softc;
1656 
1657 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1658 		return (0);
1659 
1660 	/* Generate new per-volume metadata for affected volumes. */
1661 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1662 		if (vol->v_stopping)
1663 			continue;
1664 
1665 		/* Skip volumes not related to specified targets. */
1666 		if (tvol != NULL && vol != tvol)
1667 			continue;
1668 		if (tsd != NULL && vol != tsd->sd_volume)
1669 			continue;
1670 		if (tdisk != NULL) {
1671 			for (i = 0; i < vol->v_disks_count; i++) {
1672 				if (vol->v_subdisks[i].sd_disk == tdisk)
1673 					break;
1674 			}
1675 			if (i >= vol->v_disks_count)
1676 				continue;
1677 		}
1678 
1679 		pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1680 		pv->pv_generation++;
1681 
1682 		meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
1683 		if (pv->pv_meta != NULL)
1684 			memcpy(meta, pv->pv_meta, sizeof(*meta));
1685 		memcpy(meta->promise_id, PROMISE_MAGIC,
1686 		    sizeof(PROMISE_MAGIC) - 1);
1687 		meta->dummy_0 = 0x00020000;
1688 		meta->integrity = PROMISE_I_VALID;
1689 
1690 		meta->generation = pv->pv_generation;
1691 		meta->status = PROMISE_S_VALID | PROMISE_S_ONLINE |
1692 		    PROMISE_S_INITED | PROMISE_S_READY;
1693 		if (vol->v_state <= G_RAID_VOLUME_S_DEGRADED)
1694 			meta->status |= PROMISE_S_DEGRADED;
1695 		if (vol->v_dirty)
1696 			meta->status |= PROMISE_S_MARKED; /* XXX: INVENTED! */
1697 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0 ||
1698 		    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1699 			meta->type = PROMISE_T_RAID0;
1700 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1701 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1702 			meta->type = PROMISE_T_RAID1;
1703 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3)
1704 			meta->type = PROMISE_T_RAID3;
1705 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1706 			meta->type = PROMISE_T_RAID5;
1707 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT)
1708 			meta->type = PROMISE_T_SPAN;
1709 		else
1710 			meta->type = PROMISE_T_JBOD;
1711 		meta->total_disks = vol->v_disks_count;
1712 		meta->stripe_shift = ffs(vol->v_strip_size / 1024);
1713 		meta->array_width = vol->v_disks_count;
1714 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1715 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1716 			meta->array_width /= 2;
1717 		meta->array_number = vol->v_global_id;
1718 		meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1719 		meta->cylinders = meta->total_sectors / (255 * 63) - 1;
1720 		meta->heads = 254;
1721 		meta->sectors = 63;
1722 		meta->volume_id = pv->pv_id;
1723 		rebuild_lba64 = UINT64_MAX;
1724 		rebuild = 0;
1725 		for (i = 0; i < vol->v_disks_count; i++) {
1726 			sd = &vol->v_subdisks[i];
1727 			/* For RAID0+1 we need to translate order. */
1728 			pos = promise_meta_translate_disk(vol, i);
1729 			meta->disks[pos].flags = PROMISE_F_VALID |
1730 			    PROMISE_F_ASSIGNED;
1731 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE) {
1732 				meta->disks[pos].flags |= 0;
1733 			} else if (sd->sd_state == G_RAID_SUBDISK_S_FAILED) {
1734 				meta->disks[pos].flags |=
1735 				    PROMISE_F_DOWN | PROMISE_F_REDIR;
1736 			} else if (sd->sd_state <= G_RAID_SUBDISK_S_REBUILD) {
1737 				meta->disks[pos].flags |=
1738 				    PROMISE_F_ONLINE | PROMISE_F_REDIR;
1739 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1740 					rebuild_lba64 = MIN(rebuild_lba64,
1741 					    sd->sd_rebuild_pos / 512);
1742 				} else
1743 					rebuild_lba64 = 0;
1744 				rebuild = 1;
1745 			} else {
1746 				meta->disks[pos].flags |= PROMISE_F_ONLINE;
1747 				if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) {
1748 					meta->status |= PROMISE_S_MARKED;
1749 					if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
1750 						rebuild_lba64 = MIN(rebuild_lba64,
1751 						    sd->sd_rebuild_pos / 512);
1752 					} else
1753 						rebuild_lba64 = 0;
1754 				}
1755 			}
1756 			if (pv->pv_meta != NULL) {
1757 				meta->disks[pos].id = pv->pv_meta->disks[pos].id;
1758 			} else {
1759 				meta->disks[pos].number = i * 2;
1760 				arc4rand(&meta->disks[pos].id,
1761 				    sizeof(meta->disks[pos].id), 0);
1762 			}
1763 		}
1764 		promise_meta_put_name(meta, vol->v_name);
1765 
1766 		/* Try to mimic AMD BIOS rebuild/resync behavior. */
1767 		if (rebuild_lba64 != UINT64_MAX) {
1768 			if (rebuild)
1769 				meta->magic_3 = 0x03040010UL; /* Rebuild? */
1770 			else
1771 				meta->magic_3 = 0x03040008UL; /* Resync? */
1772 			/* Translate from per-disk to per-volume LBA. */
1773 			if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1774 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1775 				rebuild_lba64 *= meta->array_width;
1776 			} else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1777 			    vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) {
1778 				rebuild_lba64 *= meta->array_width - 1;
1779 			} else
1780 				rebuild_lba64 = 0;
1781 		} else
1782 			meta->magic_3 = 0x03000000UL;
1783 		meta->rebuild_lba64 = rebuild_lba64;
1784 		meta->magic_4 = 0x04010101UL;
1785 
1786 		/* Replace per-volume metadata with new. */
1787 		if (pv->pv_meta != NULL)
1788 			free(pv->pv_meta, M_MD_PROMISE);
1789 		pv->pv_meta = meta;
1790 
1791 		/* Copy new metadata to the disks, adding or replacing old. */
1792 		for (i = 0; i < vol->v_disks_count; i++) {
1793 			sd = &vol->v_subdisks[i];
1794 			disk = sd->sd_disk;
1795 			if (disk == NULL)
1796 				continue;
1797 			/* For RAID0+1 we need to translate order. */
1798 			pos = promise_meta_translate_disk(vol, i);
1799 			pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1800 			for (j = 0; j < pd->pd_subdisks; j++) {
1801 				if (pd->pd_meta[j]->volume_id == meta->volume_id)
1802 					break;
1803 			}
1804 			if (j == pd->pd_subdisks)
1805 				pd->pd_subdisks++;
1806 			if (pd->pd_meta[j] != NULL)
1807 				free(pd->pd_meta[j], M_MD_PROMISE);
1808 			pd->pd_meta[j] = promise_meta_copy(meta);
1809 			pd->pd_meta[j]->disk = meta->disks[pos];
1810 			pd->pd_meta[j]->disk.number = pos;
1811 			pd->pd_meta[j]->disk_offset = sd->sd_offset / 512;
1812 			pd->pd_meta[j]->disk_sectors = sd->sd_size / 512;
1813 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1814 				pd->pd_meta[j]->rebuild_lba =
1815 				    sd->sd_rebuild_pos / 512;
1816 			} else if (sd->sd_state < G_RAID_SUBDISK_S_REBUILD)
1817 				pd->pd_meta[j]->rebuild_lba = 0;
1818 			else
1819 				pd->pd_meta[j]->rebuild_lba = UINT32_MAX;
1820 			pd->pd_updated = 1;
1821 		}
1822 	}
1823 
1824 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1825 		pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1826 		if (disk->d_state != G_RAID_DISK_S_ACTIVE)
1827 			continue;
1828 		if (!pd->pd_updated)
1829 			continue;
1830 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1831 		    g_raid_get_diskname(disk));
1832 		for (i = 0; i < pd->pd_subdisks; i++)
1833 			g_raid_md_promise_print(pd->pd_meta[i]);
1834 		promise_meta_write(disk->d_consumer,
1835 		    pd->pd_meta, pd->pd_subdisks);
1836 		pd->pd_updated = 0;
1837 	}
1838 
1839 	return (0);
1840 }
1841 
1842 static int
1843 g_raid_md_fail_disk_promise(struct g_raid_md_object *md,
1844     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1845 {
1846 	struct g_raid_softc *sc;
1847 	struct g_raid_md_promise_perdisk *pd;
1848 	struct g_raid_subdisk *sd;
1849 	int i, pos;
1850 
1851 	sc = md->mdo_softc;
1852 	pd = (struct g_raid_md_promise_perdisk *)tdisk->d_md_data;
1853 
1854 	/* We can't fail disk that is not a part of array now. */
1855 	if (tdisk->d_state != G_RAID_DISK_S_ACTIVE)
1856 		return (-1);
1857 
1858 	/*
1859 	 * Mark disk as failed in metadata and try to write that metadata
1860 	 * to the disk itself to prevent it's later resurrection as STALE.
1861 	 */
1862 	if (pd->pd_subdisks > 0 && tdisk->d_consumer != NULL)
1863 		G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1864 		    g_raid_get_diskname(tdisk));
1865 	for (i = 0; i < pd->pd_subdisks; i++) {
1866 		pd->pd_meta[i]->disk.flags |=
1867 		    PROMISE_F_DOWN | PROMISE_F_REDIR;
1868 		pos = pd->pd_meta[i]->disk.number;
1869 		if (pos >= 0 && pos < PROMISE_MAX_DISKS) {
1870 			pd->pd_meta[i]->disks[pos].flags |=
1871 			    PROMISE_F_DOWN | PROMISE_F_REDIR;
1872 		}
1873 		g_raid_md_promise_print(pd->pd_meta[i]);
1874 	}
1875 	if (tdisk->d_consumer != NULL)
1876 		promise_meta_write(tdisk->d_consumer,
1877 		    pd->pd_meta, pd->pd_subdisks);
1878 
1879 	/* Change states. */
1880 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1881 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1882 		g_raid_change_subdisk_state(sd,
1883 		    G_RAID_SUBDISK_S_FAILED);
1884 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1885 		    G_RAID_EVENT_SUBDISK);
1886 	}
1887 
1888 	/* Write updated metadata to remaining disks. */
1889 	g_raid_md_write_promise(md, NULL, NULL, tdisk);
1890 
1891 	g_raid_md_promise_refill(sc);
1892 	return (0);
1893 }
1894 
1895 static int
1896 g_raid_md_free_disk_promise(struct g_raid_md_object *md,
1897     struct g_raid_disk *disk)
1898 {
1899 	struct g_raid_md_promise_perdisk *pd;
1900 	int i;
1901 
1902 	pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1903 	for (i = 0; i < pd->pd_subdisks; i++) {
1904 		if (pd->pd_meta[i] != NULL) {
1905 			free(pd->pd_meta[i], M_MD_PROMISE);
1906 			pd->pd_meta[i] = NULL;
1907 		}
1908 	}
1909 	free(pd, M_MD_PROMISE);
1910 	disk->d_md_data = NULL;
1911 	return (0);
1912 }
1913 
1914 static int
1915 g_raid_md_free_volume_promise(struct g_raid_md_object *md,
1916     struct g_raid_volume *vol)
1917 {
1918 	struct g_raid_md_promise_pervolume *pv;
1919 
1920 	pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1921 	if (pv && pv->pv_meta != NULL) {
1922 		free(pv->pv_meta, M_MD_PROMISE);
1923 		pv->pv_meta = NULL;
1924 	}
1925 	if (pv && !pv->pv_started) {
1926 		pv->pv_started = 1;
1927 		callout_stop(&pv->pv_start_co);
1928 	}
1929 	return (0);
1930 }
1931 
1932 static int
1933 g_raid_md_free_promise(struct g_raid_md_object *md)
1934 {
1935 
1936 	return (0);
1937 }
1938 
1939 G_RAID_MD_DECLARE(g_raid_md_promise);
1940