xref: /freebsd/sys/geom/raid/md_sii.c (revision 40a8ac8f62b535d30349faf28cf47106b7041b83)
1 /*-
2  * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
3  * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include <sys/param.h>
32 #include <sys/bio.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/kobj.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/systm.h>
41 #include <sys/taskqueue.h>
42 #include <geom/geom.h>
43 #include "geom/raid/g_raid.h"
44 #include "g_raid_md_if.h"
45 
46 static MALLOC_DEFINE(M_MD_SII, "md_sii_data", "GEOM_RAID SiI metadata");
47 
48 struct sii_raid_conf {
49 	uint16_t	ata_params_00_53[54];
50 	uint64_t	total_sectors;		/* 54 - 57 */
51 	uint16_t	ata_params_58_81[72];
52 	uint16_t	product_id;		/* 130 */
53 	uint16_t	vendor_id;		/* 131 */
54 	uint16_t	version_minor;		/* 132 */
55 	uint16_t	version_major;		/* 133 */
56 	uint8_t		timestamp[6];		/* 134 - 136 */
57 	uint16_t	strip_sectors;		/* 137 */
58 	uint16_t	dummy_2;
59 	uint8_t		disk_number;		/* 139 */
60 	uint8_t		type;
61 #define SII_T_RAID0             0x00
62 #define SII_T_RAID1             0x01
63 #define SII_T_RAID01            0x02
64 #define SII_T_SPARE             0x03
65 #define SII_T_CONCAT            0x04
66 #define SII_T_RAID5             0x10
67 #define SII_T_RESERVED          0xfd
68 #define SII_T_JBOD              0xff
69 
70 	uint8_t		raid0_disks;		/* 140 */
71 	uint8_t		raid0_ident;
72 	uint8_t		raid1_disks;		/* 141 */
73 	uint8_t		raid1_ident;
74 	uint64_t	rebuild_lba;		/* 142 - 145 */
75 	uint32_t	generation;		/* 146 - 147 */
76 	uint8_t		disk_status;		/* 148 */
77 #define SII_S_CURRENT           0x01
78 #define SII_S_REBUILD           0x02
79 #define SII_S_DROPPED           0x03
80 #define SII_S_REMOVED           0x04
81 
82 	uint8_t		raid_status;
83 #define SII_S_ONLINE            0x01
84 #define SII_S_AVAILABLE         0x02
85 
86 	uint8_t		raid_location;		/* 149 */
87 	uint8_t		disk_location;
88 	uint8_t		auto_rebuild;		/* 150 */
89 #define SII_R_REBUILD           0x00
90 #define SII_R_NOREBUILD         0xff
91 
92 	uint8_t		dummy_3;
93 	uint8_t		name[16];		/* 151 - 158 */
94 	uint16_t	checksum;		/* 159 */
95 	uint16_t	ata_params_160_255[96];
96 } __packed;
97 
98 struct g_raid_md_sii_perdisk {
99 	struct sii_raid_conf	*pd_meta;
100 	int			 pd_disk_pos;
101 	off_t			 pd_disk_size;
102 };
103 
104 struct g_raid_md_sii_object {
105 	struct g_raid_md_object	 mdio_base;
106 	uint8_t			 mdio_timestamp[6];
107 	uint8_t			 mdio_location;
108 	uint32_t		 mdio_generation;
109 	struct sii_raid_conf	*mdio_meta;
110 	struct callout		 mdio_start_co;	/* STARTING state timer. */
111 	int			 mdio_total_disks;
112 	int			 mdio_disks_present;
113 	int			 mdio_started;
114 	int			 mdio_incomplete;
115 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
116 };
117 
118 static g_raid_md_create_t g_raid_md_create_sii;
119 static g_raid_md_taste_t g_raid_md_taste_sii;
120 static g_raid_md_event_t g_raid_md_event_sii;
121 static g_raid_md_ctl_t g_raid_md_ctl_sii;
122 static g_raid_md_write_t g_raid_md_write_sii;
123 static g_raid_md_fail_disk_t g_raid_md_fail_disk_sii;
124 static g_raid_md_free_disk_t g_raid_md_free_disk_sii;
125 static g_raid_md_free_t g_raid_md_free_sii;
126 
127 static kobj_method_t g_raid_md_sii_methods[] = {
128 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_sii),
129 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_sii),
130 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_sii),
131 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_sii),
132 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_sii),
133 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_sii),
134 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_sii),
135 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_sii),
136 	{ 0, 0 }
137 };
138 
139 static struct g_raid_md_class g_raid_md_sii_class = {
140 	"SiI",
141 	g_raid_md_sii_methods,
142 	sizeof(struct g_raid_md_sii_object),
143 	.mdc_enable = 1,
144 	.mdc_priority = 100
145 };
146 
147 static void
148 g_raid_md_sii_print(struct sii_raid_conf *meta)
149 {
150 
151 	if (g_raid_debug < 1)
152 		return;
153 
154 	printf("********* ATA SiI RAID Metadata *********\n");
155 	printf("total_sectors       %llu\n",
156 	    (long long unsigned)meta->total_sectors);
157 	printf("product_id          0x%04x\n", meta->product_id);
158 	printf("vendor_id           0x%04x\n", meta->vendor_id);
159 	printf("version_minor       0x%04x\n", meta->version_minor);
160 	printf("version_major       0x%04x\n", meta->version_major);
161 	printf("timestamp           0x%02x%02x%02x%02x%02x%02x\n",
162 	    meta->timestamp[5], meta->timestamp[4], meta->timestamp[3],
163 	    meta->timestamp[2], meta->timestamp[1], meta->timestamp[0]);
164 	printf("strip_sectors       %d\n", meta->strip_sectors);
165 	printf("disk_number         %d\n", meta->disk_number);
166 	printf("type                0x%02x\n", meta->type);
167 	printf("raid0_disks         %d\n", meta->raid0_disks);
168 	printf("raid0_ident         %d\n", meta->raid0_ident);
169 	printf("raid1_disks         %d\n", meta->raid1_disks);
170 	printf("raid1_ident         %d\n", meta->raid1_ident);
171 	printf("rebuild_lba         %llu\n",
172 	    (long long unsigned)meta->rebuild_lba);
173 	printf("generation          %d\n", meta->generation);
174 	printf("disk_status         %d\n", meta->disk_status);
175 	printf("raid_status         %d\n", meta->raid_status);
176 	printf("raid_location       %d\n", meta->raid_location);
177 	printf("disk_location       %d\n", meta->disk_location);
178 	printf("auto_rebuild        %d\n", meta->auto_rebuild);
179 	printf("name                <%.16s>\n", meta->name);
180 	printf("checksum            0x%04x\n", meta->checksum);
181 	printf("=================================================\n");
182 }
183 
184 static struct sii_raid_conf *
185 sii_meta_copy(struct sii_raid_conf *meta)
186 {
187 	struct sii_raid_conf *nmeta;
188 
189 	nmeta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK);
190 	memcpy(nmeta, meta, sizeof(*meta));
191 	return (nmeta);
192 }
193 
194 static int
195 sii_meta_total_disks(struct sii_raid_conf *meta)
196 {
197 
198 	switch (meta->type) {
199 	case SII_T_RAID0:
200 	case SII_T_RAID5:
201 	case SII_T_CONCAT:
202 		return (meta->raid0_disks);
203 	case SII_T_RAID1:
204 		return (meta->raid1_disks);
205 	case SII_T_RAID01:
206 		return (meta->raid0_disks * meta->raid1_disks);
207 	case SII_T_SPARE:
208 	case SII_T_JBOD:
209 		return (1);
210 	}
211 	return (0);
212 }
213 
214 static int
215 sii_meta_disk_pos(struct sii_raid_conf *meta, struct sii_raid_conf *pdmeta)
216 {
217 
218 	if (pdmeta->type == SII_T_SPARE)
219 		return (-3);
220 
221 	if (memcmp(&meta->timestamp, &pdmeta->timestamp, 6) != 0)
222 		return (-1);
223 
224 	switch (pdmeta->type) {
225 	case SII_T_RAID0:
226 	case SII_T_RAID1:
227 	case SII_T_RAID5:
228 	case SII_T_CONCAT:
229 		return (pdmeta->disk_number);
230 	case SII_T_RAID01:
231 		return (pdmeta->raid1_ident * pdmeta->raid1_disks +
232 		    pdmeta->raid0_ident);
233 	case SII_T_JBOD:
234 		return (0);
235 	}
236 	return (-1);
237 }
238 
239 static void
240 sii_meta_get_name(struct sii_raid_conf *meta, char *buf)
241 {
242 	int i;
243 
244 	strncpy(buf, meta->name, 16);
245 	buf[16] = 0;
246 	for (i = 15; i >= 0; i--) {
247 		if (buf[i] > 0x20)
248 			break;
249 		buf[i] = 0;
250 	}
251 }
252 
253 static void
254 sii_meta_put_name(struct sii_raid_conf *meta, char *buf)
255 {
256 
257 	memset(meta->name, 0x20, 16);
258 	memcpy(meta->name, buf, MIN(strlen(buf), 16));
259 }
260 
261 static struct sii_raid_conf *
262 sii_meta_read(struct g_consumer *cp)
263 {
264 	struct g_provider *pp;
265 	struct sii_raid_conf *meta;
266 	char *buf;
267 	int error, i;
268 	uint16_t checksum, *ptr;
269 
270 	pp = cp->provider;
271 
272 	/* Read the anchor sector. */
273 	buf = g_read_data(cp,
274 	    pp->mediasize - pp->sectorsize, pp->sectorsize, &error);
275 	if (buf == NULL) {
276 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
277 		    pp->name, error);
278 		return (NULL);
279 	}
280 	meta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK);
281 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
282 	g_free(buf);
283 
284 	/* Check vendor ID. */
285 	if (meta->vendor_id != 0x1095) {
286 		G_RAID_DEBUG(1, "SiI vendor ID check failed on %s (0x%04x)",
287 		    pp->name, meta->vendor_id);
288 		free(meta, M_MD_SII);
289 		return (NULL);
290 	}
291 
292 	/* Check metadata major version. */
293 	if (meta->version_major != 2) {
294 		G_RAID_DEBUG(1, "SiI version check failed on %s (%d.%d)",
295 		    pp->name, meta->version_major, meta->version_minor);
296 		free(meta, M_MD_SII);
297 		return (NULL);
298 	}
299 
300 	/* Check metadata checksum. */
301 	for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i <= 159; i++)
302 		checksum += *ptr++;
303 	if (checksum != 0) {
304 		G_RAID_DEBUG(1, "SiI checksum check failed on %s", pp->name);
305 		free(meta, M_MD_SII);
306 		return (NULL);
307 	}
308 
309 	/* Check raid type. */
310 	if (meta->type != SII_T_RAID0 && meta->type != SII_T_RAID1 &&
311 	    meta->type != SII_T_RAID01 && meta->type != SII_T_SPARE &&
312 	    meta->type != SII_T_RAID5 && meta->type != SII_T_CONCAT &&
313 	    meta->type != SII_T_JBOD) {
314 		G_RAID_DEBUG(1, "SiI unknown RAID level on %s (0x%02x)",
315 		    pp->name, meta->type);
316 		free(meta, M_MD_SII);
317 		return (NULL);
318 	}
319 
320 	return (meta);
321 }
322 
323 static int
324 sii_meta_write(struct g_consumer *cp, struct sii_raid_conf *meta)
325 {
326 	struct g_provider *pp;
327 	char *buf;
328 	int error, i;
329 	uint16_t checksum, *ptr;
330 
331 	pp = cp->provider;
332 
333 	/* Recalculate checksum for case if metadata were changed. */
334 	meta->checksum = 0;
335 	for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 159; i++)
336 		checksum += *ptr++;
337 	meta->checksum -= checksum;
338 
339 	/* Create and fill buffer. */
340 	buf = malloc(pp->sectorsize, M_MD_SII, M_WAITOK | M_ZERO);
341 	memcpy(buf, meta, sizeof(*meta));
342 
343 	/* Write 4 copies of metadata. */
344 	for (i = 0; i < 4; i++) {
345 		error = g_write_data(cp,
346 		    pp->mediasize - (pp->sectorsize * (1 + 0x200 * i)),
347 		    buf, pp->sectorsize);
348 		if (error != 0) {
349 			G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
350 			    pp->name, error);
351 			break;
352 		}
353 	}
354 
355 	free(buf, M_MD_SII);
356 	return (error);
357 }
358 
359 static int
360 sii_meta_erase(struct g_consumer *cp)
361 {
362 	struct g_provider *pp;
363 	char *buf;
364 	int error, i;
365 
366 	pp = cp->provider;
367 	buf = malloc(pp->sectorsize, M_MD_SII, M_WAITOK | M_ZERO);
368 	/* Write 4 copies of metadata. */
369 	for (i = 0; i < 4; i++) {
370 		error = g_write_data(cp,
371 		    pp->mediasize - (pp->sectorsize * (1 + 0x200 * i)),
372 		    buf, pp->sectorsize);
373 		if (error != 0) {
374 			G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
375 			    pp->name, error);
376 		}
377 	}
378 	free(buf, M_MD_SII);
379 	return (error);
380 }
381 
382 static int
383 sii_meta_write_spare(struct g_consumer *cp)
384 {
385 	struct sii_raid_conf *meta;
386 	int error;
387 
388 	meta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK | M_ZERO);
389 	meta->total_sectors = cp->provider->mediasize /
390 	    cp->provider->sectorsize - 0x800;
391 	meta->vendor_id = 0x1095;
392 	meta->version_minor = 0;
393 	meta->version_major = 2;
394 	meta->timestamp[0] = arc4random();
395 	meta->timestamp[1] = arc4random();
396 	meta->timestamp[2] = arc4random();
397 	meta->timestamp[3] = arc4random();
398 	meta->timestamp[4] = arc4random();
399 	meta->timestamp[5] = arc4random();
400 	meta->type = SII_T_SPARE;
401 	meta->generation = 1;
402 	meta->raid1_ident = 0xff;
403 	meta->raid_location = arc4random();
404 	error = sii_meta_write(cp, meta);
405 	free(meta, M_MD_SII);
406 	return (error);
407 }
408 
409 static struct g_raid_disk *
410 g_raid_md_sii_get_disk(struct g_raid_softc *sc, int id)
411 {
412 	struct g_raid_disk	*disk;
413 	struct g_raid_md_sii_perdisk *pd;
414 
415 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
416 		pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
417 		if (pd->pd_disk_pos == id)
418 			break;
419 	}
420 	return (disk);
421 }
422 
423 static int
424 g_raid_md_sii_supported(int level, int qual, int disks, int force)
425 {
426 
427 	if (disks > 8)
428 		return (0);
429 	switch (level) {
430 	case G_RAID_VOLUME_RL_RAID0:
431 		if (disks < 1)
432 			return (0);
433 		if (!force && (disks < 2 || disks > 6))
434 			return (0);
435 		break;
436 	case G_RAID_VOLUME_RL_RAID1:
437 		if (disks < 1)
438 			return (0);
439 		if (!force && (disks != 2))
440 			return (0);
441 		break;
442 	case G_RAID_VOLUME_RL_RAID1E:
443 		if (disks < 2)
444 			return (0);
445 		if (disks % 2 != 0)
446 			return (0);
447 		if (!force && (disks < 4))
448 			return (0);
449 		break;
450 	case G_RAID_VOLUME_RL_SINGLE:
451 		if (disks != 1)
452 			return (0);
453 		break;
454 	case G_RAID_VOLUME_RL_CONCAT:
455 		if (disks < 2)
456 			return (0);
457 		break;
458 	case G_RAID_VOLUME_RL_RAID5:
459 		if (disks < 3)
460 			return (0);
461 		if (qual != G_RAID_VOLUME_RLQ_R5LS)
462 			return (0);
463 		break;
464 	default:
465 		return (0);
466 	}
467 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
468 		return (0);
469 	return (1);
470 }
471 
472 static int
473 g_raid_md_sii_start_disk(struct g_raid_disk *disk)
474 {
475 	struct g_raid_softc *sc;
476 	struct g_raid_subdisk *sd, *tmpsd;
477 	struct g_raid_disk *olddisk, *tmpdisk;
478 	struct g_raid_md_object *md;
479 	struct g_raid_md_sii_object *mdi;
480 	struct g_raid_md_sii_perdisk *pd, *oldpd;
481 	struct sii_raid_conf *meta;
482 	int disk_pos, resurrection = 0;
483 
484 	sc = disk->d_softc;
485 	md = sc->sc_md;
486 	mdi = (struct g_raid_md_sii_object *)md;
487 	meta = mdi->mdio_meta;
488 	pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
489 	olddisk = NULL;
490 
491 	/* Find disk position in metadata by it's serial. */
492 	if (pd->pd_meta != NULL)
493 		disk_pos = sii_meta_disk_pos(meta, pd->pd_meta);
494 	else
495 		disk_pos = -3;
496 	if (disk_pos < 0) {
497 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
498 		/* If we are in the start process, that's all for now. */
499 		if (!mdi->mdio_started)
500 			goto nofit;
501 		/*
502 		 * If we have already started - try to get use of the disk.
503 		 * Try to replace OFFLINE disks first, then FAILED.
504 		 */
505 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
506 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
507 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
508 				continue;
509 			/* Make sure this disk is big enough. */
510 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
511 				if (sd->sd_offset + sd->sd_size + 512 >
512 				    pd->pd_disk_size) {
513 					G_RAID_DEBUG1(1, sc,
514 					    "Disk too small (%ju < %ju)",
515 					    pd->pd_disk_size,
516 					    sd->sd_offset + sd->sd_size + 512);
517 					break;
518 				}
519 			}
520 			if (sd != NULL)
521 				continue;
522 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
523 				olddisk = tmpdisk;
524 				break;
525 			} else if (olddisk == NULL)
526 				olddisk = tmpdisk;
527 		}
528 		if (olddisk == NULL) {
529 nofit:
530 			if (disk_pos == -3 || pd->pd_disk_pos == -3) {
531 				g_raid_change_disk_state(disk,
532 				    G_RAID_DISK_S_SPARE);
533 				return (1);
534 			} else {
535 				g_raid_change_disk_state(disk,
536 				    G_RAID_DISK_S_STALE);
537 				return (0);
538 			}
539 		}
540 		oldpd = (struct g_raid_md_sii_perdisk *)olddisk->d_md_data;
541 		disk_pos = oldpd->pd_disk_pos;
542 		resurrection = 1;
543 	}
544 
545 	if (olddisk == NULL) {
546 		/* Find placeholder by position. */
547 		olddisk = g_raid_md_sii_get_disk(sc, disk_pos);
548 		if (olddisk == NULL)
549 			panic("No disk at position %d!", disk_pos);
550 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
551 			G_RAID_DEBUG1(1, sc, "More then one disk for pos %d",
552 			    disk_pos);
553 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
554 			return (0);
555 		}
556 		oldpd = (struct g_raid_md_sii_perdisk *)olddisk->d_md_data;
557 	}
558 
559 	/* Replace failed disk or placeholder with new disk. */
560 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
561 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
562 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
563 		sd->sd_disk = disk;
564 	}
565 	oldpd->pd_disk_pos = -2;
566 	pd->pd_disk_pos = disk_pos;
567 
568 	/* If it was placeholder -- destroy it. */
569 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
570 		g_raid_destroy_disk(olddisk);
571 	} else {
572 		/* Otherwise, make it STALE_FAILED. */
573 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
574 	}
575 
576 	/* Welcome the new disk. */
577 	if (resurrection)
578 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
579 	else if (pd->pd_meta->disk_status == SII_S_CURRENT ||
580 	    pd->pd_meta->disk_status == SII_S_REBUILD)
581 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
582 	else
583 		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
584 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
585 
586 		/*
587 		 * Different disks may have different sizes,
588 		 * in concat mode. Update from real disk size.
589 		 */
590 		if (meta->type == SII_T_CONCAT || meta->type == SII_T_JBOD)
591 			sd->sd_size = pd->pd_disk_size - 0x800 * 512;
592 
593 		if (resurrection) {
594 			/* New or ex-spare disk. */
595 			g_raid_change_subdisk_state(sd,
596 			    G_RAID_SUBDISK_S_NEW);
597 		} else if (pd->pd_meta->disk_status == SII_S_REBUILD) {
598 			/* Rebuilding disk. */
599 			g_raid_change_subdisk_state(sd,
600 			    G_RAID_SUBDISK_S_REBUILD);
601 			if (pd->pd_meta->generation == meta->generation)
602 				sd->sd_rebuild_pos = pd->pd_meta->rebuild_lba * 512;
603 			else
604 				sd->sd_rebuild_pos = 0;
605 		} else if (pd->pd_meta->disk_status == SII_S_CURRENT) {
606 			if (pd->pd_meta->raid_status == SII_S_ONLINE ||
607 			    pd->pd_meta->generation != meta->generation) {
608 				/* Dirty or resyncing disk. */
609 				g_raid_change_subdisk_state(sd,
610 				    G_RAID_SUBDISK_S_STALE);
611 			} else {
612 				/* Up to date disk. */
613 				g_raid_change_subdisk_state(sd,
614 				    G_RAID_SUBDISK_S_ACTIVE);
615 			}
616 		} else {
617 			g_raid_change_subdisk_state(sd,
618 			    G_RAID_SUBDISK_S_FAILED);
619 		}
620 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
621 		    G_RAID_EVENT_SUBDISK);
622 	}
623 
624 	/* Update status of our need for spare. */
625 	if (mdi->mdio_started) {
626 		mdi->mdio_incomplete =
627 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
628 		     mdi->mdio_total_disks);
629 	}
630 
631 	return (resurrection);
632 }
633 
634 static void
635 g_disk_md_sii_retaste(void *arg, int pending)
636 {
637 
638 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
639 	g_retaste(&g_raid_class);
640 	free(arg, M_MD_SII);
641 }
642 
643 static void
644 g_raid_md_sii_refill(struct g_raid_softc *sc)
645 {
646 	struct g_raid_md_object *md;
647 	struct g_raid_md_sii_object *mdi;
648 	struct g_raid_disk *disk;
649 	struct task *task;
650 	int update, na;
651 
652 	md = sc->sc_md;
653 	mdi = (struct g_raid_md_sii_object *)md;
654 	update = 0;
655 	do {
656 		/* Make sure we miss anything. */
657 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
658 		if (na == mdi->mdio_total_disks)
659 			break;
660 
661 		G_RAID_DEBUG1(1, md->mdo_softc,
662 		    "Array is not complete (%d of %d), "
663 		    "trying to refill.", na, mdi->mdio_total_disks);
664 
665 		/* Try to get use some of STALE disks. */
666 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
667 			if (disk->d_state == G_RAID_DISK_S_STALE) {
668 				update += g_raid_md_sii_start_disk(disk);
669 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
670 					break;
671 			}
672 		}
673 		if (disk != NULL)
674 			continue;
675 
676 		/* Try to get use some of SPARE disks. */
677 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
678 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
679 				update += g_raid_md_sii_start_disk(disk);
680 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
681 					break;
682 			}
683 		}
684 	} while (disk != NULL);
685 
686 	/* Write new metadata if we changed something. */
687 	if (update)
688 		g_raid_md_write_sii(md, NULL, NULL, NULL);
689 
690 	/* Update status of our need for spare. */
691 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
692 	    mdi->mdio_total_disks);
693 
694 	/* Request retaste hoping to find spare. */
695 	if (mdi->mdio_incomplete) {
696 		task = malloc(sizeof(struct task),
697 		    M_MD_SII, M_WAITOK | M_ZERO);
698 		TASK_INIT(task, 0, g_disk_md_sii_retaste, task);
699 		taskqueue_enqueue(taskqueue_swi, task);
700 	}
701 }
702 
703 static void
704 g_raid_md_sii_start(struct g_raid_softc *sc)
705 {
706 	struct g_raid_md_object *md;
707 	struct g_raid_md_sii_object *mdi;
708 	struct g_raid_md_sii_perdisk *pd;
709 	struct sii_raid_conf *meta;
710 	struct g_raid_volume *vol;
711 	struct g_raid_subdisk *sd;
712 	struct g_raid_disk *disk, *best;
713 	off_t size;
714 	int j, disk_pos;
715 	uint32_t gendiff, bestgendiff;
716 	char buf[17];
717 
718 	md = sc->sc_md;
719 	mdi = (struct g_raid_md_sii_object *)md;
720 	meta = mdi->mdio_meta;
721 
722 	/* Create volumes and subdisks. */
723 	sii_meta_get_name(meta, buf);
724 	vol = g_raid_create_volume(sc, buf, -1);
725 	vol->v_mediasize = (off_t)meta->total_sectors * 512;
726 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
727 	if (meta->type == SII_T_RAID0) {
728 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
729 		size = vol->v_mediasize / mdi->mdio_total_disks;
730 	} else if (meta->type == SII_T_RAID1) {
731 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
732 		size = vol->v_mediasize;
733 	} else if (meta->type == SII_T_RAID01) {
734 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
735 		size = vol->v_mediasize / (mdi->mdio_total_disks / 2);
736 	} else if (meta->type == SII_T_CONCAT) {
737 		if (mdi->mdio_total_disks == 1)
738 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
739 		else
740 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
741 		size = 0;
742 	} else if (meta->type == SII_T_RAID5) {
743 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
744 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LS;
745 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
746 	} else if (meta->type == SII_T_JBOD) {
747 		vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
748 		size = 0;
749 	} else {
750 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
751 		size = 0;
752 	}
753 	vol->v_strip_size = meta->strip_sectors * 512; //ZZZ
754 	vol->v_disks_count = mdi->mdio_total_disks;
755 	vol->v_sectorsize = 512; //ZZZ
756 	for (j = 0; j < vol->v_disks_count; j++) {
757 		sd = &vol->v_subdisks[j];
758 		sd->sd_offset = 0;
759 		sd->sd_size = size;
760 	}
761 	g_raid_start_volume(vol);
762 
763 	/* Create disk placeholders to store data for later writing. */
764 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
765 		pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO);
766 		pd->pd_disk_pos = disk_pos;
767 		disk = g_raid_create_disk(sc);
768 		disk->d_md_data = (void *)pd;
769 		disk->d_state = G_RAID_DISK_S_OFFLINE;
770 		sd = &vol->v_subdisks[disk_pos];
771 		sd->sd_disk = disk;
772 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
773 	}
774 
775 	/*
776 	 * Make all disks found till the moment take their places
777 	 * in order of their generation numbers.
778 	 */
779 	do {
780 		best = NULL;
781 		bestgendiff = 0xffffffff;
782 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
783 			if (disk->d_state != G_RAID_DISK_S_NONE)
784 				continue;
785 			pd = disk->d_md_data;
786 			if (pd->pd_meta == NULL)
787 				gendiff = 0xfffffffe;
788 			else
789 				gendiff = meta->generation -
790 				    pd->pd_meta->generation;
791 			if (gendiff < bestgendiff) {
792 				best = disk;
793 				bestgendiff = gendiff;
794 			}
795 		}
796 		if (best != NULL)
797 			g_raid_md_sii_start_disk(best);
798 	} while (best != NULL);
799 
800 	mdi->mdio_started = 1;
801 	G_RAID_DEBUG1(0, sc, "Array started.");
802 	g_raid_md_write_sii(md, NULL, NULL, NULL);
803 
804 	/* Pickup any STALE/SPARE disks to refill array if needed. */
805 	g_raid_md_sii_refill(sc);
806 
807 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
808 
809 	callout_stop(&mdi->mdio_start_co);
810 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
811 	root_mount_rel(mdi->mdio_rootmount);
812 	mdi->mdio_rootmount = NULL;
813 }
814 
815 static void
816 g_raid_md_sii_new_disk(struct g_raid_disk *disk)
817 {
818 	struct g_raid_softc *sc;
819 	struct g_raid_md_object *md;
820 	struct g_raid_md_sii_object *mdi;
821 	struct sii_raid_conf *pdmeta;
822 	struct g_raid_md_sii_perdisk *pd;
823 
824 	sc = disk->d_softc;
825 	md = sc->sc_md;
826 	mdi = (struct g_raid_md_sii_object *)md;
827 	pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
828 	pdmeta = pd->pd_meta;
829 
830 	if (mdi->mdio_started) {
831 		if (g_raid_md_sii_start_disk(disk))
832 			g_raid_md_write_sii(md, NULL, NULL, NULL);
833 	} else {
834 		if (mdi->mdio_meta == NULL ||
835 		    ((int32_t)(pdmeta->generation - mdi->mdio_generation)) > 0) {
836 			G_RAID_DEBUG1(1, sc, "Newer disk");
837 			if (mdi->mdio_meta != NULL)
838 				free(mdi->mdio_meta, M_MD_SII);
839 			mdi->mdio_meta = sii_meta_copy(pdmeta);
840 			mdi->mdio_generation = mdi->mdio_meta->generation;
841 			mdi->mdio_total_disks = sii_meta_total_disks(pdmeta);
842 			mdi->mdio_disks_present = 1;
843 		} else if (pdmeta->generation == mdi->mdio_generation) {
844 			mdi->mdio_disks_present++;
845 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
846 			    mdi->mdio_disks_present,
847 			    mdi->mdio_total_disks);
848 		} else {
849 			G_RAID_DEBUG1(1, sc, "Older disk");
850 		}
851 
852 		/* If we collected all needed disks - start array. */
853 		if (mdi->mdio_disks_present == mdi->mdio_total_disks)
854 			g_raid_md_sii_start(sc);
855 	}
856 }
857 
858 static void
859 g_raid_sii_go(void *arg)
860 {
861 	struct g_raid_softc *sc;
862 	struct g_raid_md_object *md;
863 	struct g_raid_md_sii_object *mdi;
864 
865 	sc = arg;
866 	md = sc->sc_md;
867 	mdi = (struct g_raid_md_sii_object *)md;
868 	if (!mdi->mdio_started) {
869 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
870 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
871 	}
872 }
873 
874 static int
875 g_raid_md_create_sii(struct g_raid_md_object *md, struct g_class *mp,
876     struct g_geom **gp)
877 {
878 	struct g_raid_softc *sc;
879 	struct g_raid_md_sii_object *mdi;
880 	char name[32];
881 
882 	mdi = (struct g_raid_md_sii_object *)md;
883 	mdi->mdio_timestamp[5] = arc4random();
884 	mdi->mdio_timestamp[4] = arc4random();
885 	mdi->mdio_timestamp[3] = arc4random();
886 	mdi->mdio_timestamp[2] = arc4random();
887 	mdi->mdio_timestamp[1] = arc4random();
888 	mdi->mdio_timestamp[0] = arc4random();
889 	mdi->mdio_location = arc4random();
890 	mdi->mdio_generation = 0;
891 	snprintf(name, sizeof(name), "SiI-%02x%02x%02x%02x%02x%02x",
892 	    mdi->mdio_timestamp[5], mdi->mdio_timestamp[4],
893 	    mdi->mdio_timestamp[3], mdi->mdio_timestamp[2],
894 	    mdi->mdio_timestamp[1], mdi->mdio_timestamp[0]);
895 	sc = g_raid_create_node(mp, name, md);
896 	if (sc == NULL)
897 		return (G_RAID_MD_TASTE_FAIL);
898 	md->mdo_softc = sc;
899 	*gp = sc->sc_geom;
900 	return (G_RAID_MD_TASTE_NEW);
901 }
902 
903 static int
904 g_raid_md_taste_sii(struct g_raid_md_object *md, struct g_class *mp,
905                               struct g_consumer *cp, struct g_geom **gp)
906 {
907 	struct g_consumer *rcp;
908 	struct g_provider *pp;
909 	struct g_raid_md_sii_object *mdi, *mdi1;
910 	struct g_raid_softc *sc;
911 	struct g_raid_disk *disk;
912 	struct sii_raid_conf *meta;
913 	struct g_raid_md_sii_perdisk *pd;
914 	struct g_geom *geom;
915 	int disk_pos, result, spare, len;
916 	char name[32];
917 	uint16_t vendor;
918 
919 	G_RAID_DEBUG(1, "Tasting SiI on %s", cp->provider->name);
920 	mdi = (struct g_raid_md_sii_object *)md;
921 	pp = cp->provider;
922 
923 	/* Read metadata from device. */
924 	meta = NULL;
925 	vendor = 0xffff;
926 	g_topology_unlock();
927 	len = 2;
928 	if (pp->geom->rank == 1)
929 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
930 	meta = sii_meta_read(cp);
931 	g_topology_lock();
932 	if (meta == NULL) {
933 		if (g_raid_aggressive_spare) {
934 			if (vendor == 0x1095) {
935 				G_RAID_DEBUG(1,
936 				    "No SiI metadata, forcing spare.");
937 				spare = 2;
938 				goto search;
939 			} else {
940 				G_RAID_DEBUG(1,
941 				    "SiI vendor mismatch 0x%04x != 0x1095",
942 				    vendor);
943 			}
944 		}
945 		return (G_RAID_MD_TASTE_FAIL);
946 	}
947 
948 	/* Check this disk position in obtained metadata. */
949 	disk_pos = sii_meta_disk_pos(meta, meta);
950 	if (disk_pos == -1) {
951 		G_RAID_DEBUG(1, "SiI disk position not found");
952 		goto fail1;
953 	}
954 
955 	/* Metadata valid. Print it. */
956 	g_raid_md_sii_print(meta);
957 	G_RAID_DEBUG(1, "SiI disk position %d", disk_pos);
958 	spare = (meta->type == SII_T_SPARE) ? 1 : 0;
959 
960 search:
961 	/* Search for matching node. */
962 	sc = NULL;
963 	mdi1 = NULL;
964 	LIST_FOREACH(geom, &mp->geom, geom) {
965 		sc = geom->softc;
966 		if (sc == NULL)
967 			continue;
968 		if (sc->sc_stopping != 0)
969 			continue;
970 		if (sc->sc_md->mdo_class != md->mdo_class)
971 			continue;
972 		mdi1 = (struct g_raid_md_sii_object *)sc->sc_md;
973 		if (spare) {
974 			if (mdi1->mdio_incomplete)
975 				break;
976 		} else {
977 			if (mdi1->mdio_location == meta->raid_location &&
978 			    memcmp(&mdi1->mdio_timestamp,
979 			     &meta->timestamp, 6) == 0)
980 				break;
981 		}
982 	}
983 
984 	/* Found matching node. */
985 	if (geom != NULL) {
986 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
987 		result = G_RAID_MD_TASTE_EXISTING;
988 
989 	} else if (spare) { /* Not found needy node -- left for later. */
990 		G_RAID_DEBUG(1, "Spare is not needed at this time");
991 		goto fail1;
992 
993 	} else { /* Not found matching node -- create one. */
994 		result = G_RAID_MD_TASTE_NEW;
995 		memcpy(&mdi->mdio_timestamp, &meta->timestamp, 6);
996 		mdi->mdio_location = meta->raid_location;
997 		snprintf(name, sizeof(name), "SiI-%02x%02x%02x%02x%02x%02x",
998 		    mdi->mdio_timestamp[5], mdi->mdio_timestamp[4],
999 		    mdi->mdio_timestamp[3], mdi->mdio_timestamp[2],
1000 		    mdi->mdio_timestamp[1], mdi->mdio_timestamp[0]);
1001 		sc = g_raid_create_node(mp, name, md);
1002 		md->mdo_softc = sc;
1003 		geom = sc->sc_geom;
1004 		callout_init(&mdi->mdio_start_co, 1);
1005 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
1006 		    g_raid_sii_go, sc);
1007 		mdi->mdio_rootmount = root_mount_hold("GRAID-SiI");
1008 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
1009 	}
1010 
1011 	/* There is no return after this point, so we close passed consumer. */
1012 	g_access(cp, -1, 0, 0);
1013 
1014 	rcp = g_new_consumer(geom);
1015 	rcp->flags |= G_CF_DIRECT_RECEIVE;
1016 	g_attach(rcp, pp);
1017 	if (g_access(rcp, 1, 1, 1) != 0)
1018 		; //goto fail1;
1019 
1020 	g_topology_unlock();
1021 	sx_xlock(&sc->sc_lock);
1022 
1023 	pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO);
1024 	pd->pd_meta = meta;
1025 	if (spare == 2) {
1026 		pd->pd_disk_pos = -3;
1027 	} else {
1028 		pd->pd_disk_pos = -1;
1029 	}
1030 	pd->pd_disk_size = pp->mediasize;
1031 	disk = g_raid_create_disk(sc);
1032 	disk->d_md_data = (void *)pd;
1033 	disk->d_consumer = rcp;
1034 	rcp->private = disk;
1035 
1036 	g_raid_get_disk_info(disk);
1037 
1038 	g_raid_md_sii_new_disk(disk);
1039 
1040 	sx_xunlock(&sc->sc_lock);
1041 	g_topology_lock();
1042 	*gp = geom;
1043 	return (result);
1044 fail1:
1045 	free(meta, M_MD_SII);
1046 	return (G_RAID_MD_TASTE_FAIL);
1047 }
1048 
1049 static int
1050 g_raid_md_event_sii(struct g_raid_md_object *md,
1051     struct g_raid_disk *disk, u_int event)
1052 {
1053 	struct g_raid_softc *sc;
1054 	struct g_raid_subdisk *sd;
1055 	struct g_raid_md_sii_object *mdi;
1056 	struct g_raid_md_sii_perdisk *pd;
1057 
1058 	sc = md->mdo_softc;
1059 	mdi = (struct g_raid_md_sii_object *)md;
1060 	if (disk == NULL) {
1061 		switch (event) {
1062 		case G_RAID_NODE_E_START:
1063 			if (!mdi->mdio_started)
1064 				g_raid_md_sii_start(sc);
1065 			return (0);
1066 		}
1067 		return (-1);
1068 	}
1069 	pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
1070 	switch (event) {
1071 	case G_RAID_DISK_E_DISCONNECTED:
1072 		/* If disk was assigned, just update statuses. */
1073 		if (pd->pd_disk_pos >= 0) {
1074 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1075 			if (disk->d_consumer) {
1076 				g_raid_kill_consumer(sc, disk->d_consumer);
1077 				disk->d_consumer = NULL;
1078 			}
1079 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1080 				g_raid_change_subdisk_state(sd,
1081 				    G_RAID_SUBDISK_S_NONE);
1082 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1083 				    G_RAID_EVENT_SUBDISK);
1084 			}
1085 		} else {
1086 			/* Otherwise -- delete. */
1087 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1088 			g_raid_destroy_disk(disk);
1089 		}
1090 
1091 		/* Write updated metadata to all disks. */
1092 		g_raid_md_write_sii(md, NULL, NULL, NULL);
1093 
1094 		/* Check if anything left except placeholders. */
1095 		if (g_raid_ndisks(sc, -1) ==
1096 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1097 			g_raid_destroy_node(sc, 0);
1098 		else
1099 			g_raid_md_sii_refill(sc);
1100 		return (0);
1101 	}
1102 	return (-2);
1103 }
1104 
1105 static int
1106 g_raid_md_ctl_sii(struct g_raid_md_object *md,
1107     struct gctl_req *req)
1108 {
1109 	struct g_raid_softc *sc;
1110 	struct g_raid_volume *vol;
1111 	struct g_raid_subdisk *sd;
1112 	struct g_raid_disk *disk;
1113 	struct g_raid_md_sii_object *mdi;
1114 	struct g_raid_md_sii_perdisk *pd;
1115 	struct g_consumer *cp;
1116 	struct g_provider *pp;
1117 	char arg[16];
1118 	const char *verb, *volname, *levelname, *diskname;
1119 	int *nargs, *force;
1120 	off_t size, sectorsize, strip;
1121 	intmax_t *sizearg, *striparg;
1122 	int numdisks, i, len, level, qual, update;
1123 	int error;
1124 
1125 	sc = md->mdo_softc;
1126 	mdi = (struct g_raid_md_sii_object *)md;
1127 	verb = gctl_get_param(req, "verb", NULL);
1128 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1129 	error = 0;
1130 	if (strcmp(verb, "label") == 0) {
1131 
1132 		if (*nargs < 4) {
1133 			gctl_error(req, "Invalid number of arguments.");
1134 			return (-1);
1135 		}
1136 		volname = gctl_get_asciiparam(req, "arg1");
1137 		if (volname == NULL) {
1138 			gctl_error(req, "No volume name.");
1139 			return (-2);
1140 		}
1141 		levelname = gctl_get_asciiparam(req, "arg2");
1142 		if (levelname == NULL) {
1143 			gctl_error(req, "No RAID level.");
1144 			return (-3);
1145 		}
1146 		if (strcasecmp(levelname, "RAID5") == 0)
1147 			levelname = "RAID5-LS";
1148 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1149 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1150 			return (-4);
1151 		}
1152 		numdisks = *nargs - 3;
1153 		force = gctl_get_paraml(req, "force", sizeof(*force));
1154 		if (!g_raid_md_sii_supported(level, qual, numdisks,
1155 		    force ? *force : 0)) {
1156 			gctl_error(req, "Unsupported RAID level "
1157 			    "(0x%02x/0x%02x), or number of disks (%d).",
1158 			    level, qual, numdisks);
1159 			return (-5);
1160 		}
1161 
1162 		/* Search for disks, connect them and probe. */
1163 		size = 0x7fffffffffffffffllu;
1164 		sectorsize = 0;
1165 		for (i = 0; i < numdisks; i++) {
1166 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1167 			diskname = gctl_get_asciiparam(req, arg);
1168 			if (diskname == NULL) {
1169 				gctl_error(req, "No disk name (%s).", arg);
1170 				error = -6;
1171 				break;
1172 			}
1173 			if (strcmp(diskname, "NONE") == 0) {
1174 				cp = NULL;
1175 				pp = NULL;
1176 			} else {
1177 				g_topology_lock();
1178 				cp = g_raid_open_consumer(sc, diskname);
1179 				if (cp == NULL) {
1180 					gctl_error(req, "Can't open '%s'.",
1181 					    diskname);
1182 					g_topology_unlock();
1183 					error = -7;
1184 					break;
1185 				}
1186 				pp = cp->provider;
1187 			}
1188 			pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO);
1189 			pd->pd_disk_pos = i;
1190 			disk = g_raid_create_disk(sc);
1191 			disk->d_md_data = (void *)pd;
1192 			disk->d_consumer = cp;
1193 			if (cp == NULL)
1194 				continue;
1195 			cp->private = disk;
1196 			g_topology_unlock();
1197 
1198 			g_raid_get_disk_info(disk);
1199 
1200 			pd->pd_disk_size = pp->mediasize;
1201 			if (size > pp->mediasize)
1202 				size = pp->mediasize;
1203 			if (sectorsize < pp->sectorsize)
1204 				sectorsize = pp->sectorsize;
1205 		}
1206 		if (error != 0)
1207 			return (error);
1208 
1209 		if (sectorsize <= 0) {
1210 			gctl_error(req, "Can't get sector size.");
1211 			return (-8);
1212 		}
1213 
1214 		/* Reserve space for metadata. */
1215 		size -= 0x800 * sectorsize;
1216 
1217 		/* Handle size argument. */
1218 		len = sizeof(*sizearg);
1219 		sizearg = gctl_get_param(req, "size", &len);
1220 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1221 		    *sizearg > 0) {
1222 			if (*sizearg > size) {
1223 				gctl_error(req, "Size too big %lld > %lld.",
1224 				    (long long)*sizearg, (long long)size);
1225 				return (-9);
1226 			}
1227 			size = *sizearg;
1228 		}
1229 
1230 		/* Handle strip argument. */
1231 		strip = 131072;
1232 		len = sizeof(*striparg);
1233 		striparg = gctl_get_param(req, "strip", &len);
1234 		if (striparg != NULL && len == sizeof(*striparg) &&
1235 		    *striparg > 0) {
1236 			if (*striparg < sectorsize) {
1237 				gctl_error(req, "Strip size too small.");
1238 				return (-10);
1239 			}
1240 			if (*striparg % sectorsize != 0) {
1241 				gctl_error(req, "Incorrect strip size.");
1242 				return (-11);
1243 			}
1244 			if (strip > 65535 * sectorsize) {
1245 				gctl_error(req, "Strip size too big.");
1246 				return (-12);
1247 			}
1248 			strip = *striparg;
1249 		}
1250 
1251 		/* Round size down to strip or sector. */
1252 		if (level == G_RAID_VOLUME_RL_RAID1)
1253 			size -= (size % sectorsize);
1254 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1255 		    (numdisks & 1) != 0)
1256 			size -= (size % (2 * strip));
1257 		else
1258 			size -= (size % strip);
1259 		if (size <= 0) {
1260 			gctl_error(req, "Size too small.");
1261 			return (-13);
1262 		}
1263 		if (size > 0xffffffffffffllu * sectorsize) {
1264 			gctl_error(req, "Size too big.");
1265 			return (-14);
1266 		}
1267 
1268 		/* We have all we need, create things: volume, ... */
1269 		mdi->mdio_total_disks = numdisks;
1270 		mdi->mdio_started = 1;
1271 		vol = g_raid_create_volume(sc, volname, -1);
1272 		vol->v_md_data = (void *)(intptr_t)0;
1273 		vol->v_raid_level = level;
1274 		vol->v_raid_level_qualifier = qual;
1275 		vol->v_strip_size = strip;
1276 		vol->v_disks_count = numdisks;
1277 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1278 		    level == G_RAID_VOLUME_RL_CONCAT ||
1279 		    level == G_RAID_VOLUME_RL_SINGLE)
1280 			vol->v_mediasize = size * numdisks;
1281 		else if (level == G_RAID_VOLUME_RL_RAID1)
1282 			vol->v_mediasize = size;
1283 		else if (level == G_RAID_VOLUME_RL_RAID5)
1284 			vol->v_mediasize = size * (numdisks - 1);
1285 		else { /* RAID1E */
1286 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1287 			    strip;
1288 		}
1289 		vol->v_sectorsize = sectorsize;
1290 		g_raid_start_volume(vol);
1291 
1292 		/* , and subdisks. */
1293 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1294 			pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
1295 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1296 			sd->sd_disk = disk;
1297 			sd->sd_offset = 0;
1298 			sd->sd_size = size;
1299 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1300 			if (sd->sd_disk->d_consumer != NULL) {
1301 				g_raid_change_disk_state(disk,
1302 				    G_RAID_DISK_S_ACTIVE);
1303 				g_raid_change_subdisk_state(sd,
1304 				    G_RAID_SUBDISK_S_ACTIVE);
1305 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1306 				    G_RAID_EVENT_SUBDISK);
1307 			} else {
1308 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1309 			}
1310 		}
1311 
1312 		/* Write metadata based on created entities. */
1313 		G_RAID_DEBUG1(0, sc, "Array started.");
1314 		g_raid_md_write_sii(md, NULL, NULL, NULL);
1315 
1316 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1317 		g_raid_md_sii_refill(sc);
1318 
1319 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1320 		    G_RAID_EVENT_VOLUME);
1321 		return (0);
1322 	}
1323 	if (strcmp(verb, "delete") == 0) {
1324 
1325 		/* Check if some volume is still open. */
1326 		force = gctl_get_paraml(req, "force", sizeof(*force));
1327 		if (force != NULL && *force == 0 &&
1328 		    g_raid_nopens(sc) != 0) {
1329 			gctl_error(req, "Some volume is still open.");
1330 			return (-4);
1331 		}
1332 
1333 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1334 			if (disk->d_consumer)
1335 				sii_meta_erase(disk->d_consumer);
1336 		}
1337 		g_raid_destroy_node(sc, 0);
1338 		return (0);
1339 	}
1340 	if (strcmp(verb, "remove") == 0 ||
1341 	    strcmp(verb, "fail") == 0) {
1342 		if (*nargs < 2) {
1343 			gctl_error(req, "Invalid number of arguments.");
1344 			return (-1);
1345 		}
1346 		for (i = 1; i < *nargs; i++) {
1347 			snprintf(arg, sizeof(arg), "arg%d", i);
1348 			diskname = gctl_get_asciiparam(req, arg);
1349 			if (diskname == NULL) {
1350 				gctl_error(req, "No disk name (%s).", arg);
1351 				error = -2;
1352 				break;
1353 			}
1354 			if (strncmp(diskname, "/dev/", 5) == 0)
1355 				diskname += 5;
1356 
1357 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1358 				if (disk->d_consumer != NULL &&
1359 				    disk->d_consumer->provider != NULL &&
1360 				    strcmp(disk->d_consumer->provider->name,
1361 				     diskname) == 0)
1362 					break;
1363 			}
1364 			if (disk == NULL) {
1365 				gctl_error(req, "Disk '%s' not found.",
1366 				    diskname);
1367 				error = -3;
1368 				break;
1369 			}
1370 
1371 			if (strcmp(verb, "fail") == 0) {
1372 				g_raid_md_fail_disk_sii(md, NULL, disk);
1373 				continue;
1374 			}
1375 
1376 			pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
1377 
1378 			/* Erase metadata on deleting disk. */
1379 			sii_meta_erase(disk->d_consumer);
1380 
1381 			/* If disk was assigned, just update statuses. */
1382 			if (pd->pd_disk_pos >= 0) {
1383 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1384 				g_raid_kill_consumer(sc, disk->d_consumer);
1385 				disk->d_consumer = NULL;
1386 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1387 					g_raid_change_subdisk_state(sd,
1388 					    G_RAID_SUBDISK_S_NONE);
1389 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1390 					    G_RAID_EVENT_SUBDISK);
1391 				}
1392 			} else {
1393 				/* Otherwise -- delete. */
1394 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1395 				g_raid_destroy_disk(disk);
1396 			}
1397 		}
1398 
1399 		/* Write updated metadata to remaining disks. */
1400 		g_raid_md_write_sii(md, NULL, NULL, NULL);
1401 
1402 		/* Check if anything left except placeholders. */
1403 		if (g_raid_ndisks(sc, -1) ==
1404 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1405 			g_raid_destroy_node(sc, 0);
1406 		else
1407 			g_raid_md_sii_refill(sc);
1408 		return (error);
1409 	}
1410 	if (strcmp(verb, "insert") == 0) {
1411 		if (*nargs < 2) {
1412 			gctl_error(req, "Invalid number of arguments.");
1413 			return (-1);
1414 		}
1415 		update = 0;
1416 		for (i = 1; i < *nargs; i++) {
1417 			/* Get disk name. */
1418 			snprintf(arg, sizeof(arg), "arg%d", i);
1419 			diskname = gctl_get_asciiparam(req, arg);
1420 			if (diskname == NULL) {
1421 				gctl_error(req, "No disk name (%s).", arg);
1422 				error = -3;
1423 				break;
1424 			}
1425 
1426 			/* Try to find provider with specified name. */
1427 			g_topology_lock();
1428 			cp = g_raid_open_consumer(sc, diskname);
1429 			if (cp == NULL) {
1430 				gctl_error(req, "Can't open disk '%s'.",
1431 				    diskname);
1432 				g_topology_unlock();
1433 				error = -4;
1434 				break;
1435 			}
1436 			pp = cp->provider;
1437 
1438 			pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO);
1439 			pd->pd_disk_pos = -3;
1440 			pd->pd_disk_size = pp->mediasize;
1441 
1442 			disk = g_raid_create_disk(sc);
1443 			disk->d_consumer = cp;
1444 			disk->d_md_data = (void *)pd;
1445 			cp->private = disk;
1446 			g_topology_unlock();
1447 
1448 			g_raid_get_disk_info(disk);
1449 
1450 			/* Welcome the "new" disk. */
1451 			update += g_raid_md_sii_start_disk(disk);
1452 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
1453 				sii_meta_write_spare(cp);
1454 				g_raid_destroy_disk(disk);
1455 			} else if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
1456 				gctl_error(req, "Disk '%s' doesn't fit.",
1457 				    diskname);
1458 				g_raid_destroy_disk(disk);
1459 				error = -8;
1460 				break;
1461 			}
1462 		}
1463 
1464 		/* Write new metadata if we changed something. */
1465 		if (update)
1466 			g_raid_md_write_sii(md, NULL, NULL, NULL);
1467 		return (error);
1468 	}
1469 	gctl_error(req, "Command '%s' is not supported.", verb);
1470 	return (-100);
1471 }
1472 
1473 static int
1474 g_raid_md_write_sii(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1475     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1476 {
1477 	struct g_raid_softc *sc;
1478 	struct g_raid_volume *vol;
1479 	struct g_raid_subdisk *sd;
1480 	struct g_raid_disk *disk;
1481 	struct g_raid_md_sii_object *mdi;
1482 	struct g_raid_md_sii_perdisk *pd;
1483 	struct sii_raid_conf *meta;
1484 	int i;
1485 
1486 	sc = md->mdo_softc;
1487 	mdi = (struct g_raid_md_sii_object *)md;
1488 
1489 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1490 		return (0);
1491 
1492 	/* Bump generation. Newly written metadata may differ from previous. */
1493 	mdi->mdio_generation++;
1494 
1495 	/* There is only one volume. */
1496 	vol = TAILQ_FIRST(&sc->sc_volumes);
1497 
1498 	/* Fill global fields. */
1499 	meta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK | M_ZERO);
1500 	if (mdi->mdio_meta)
1501 		memcpy(meta, mdi->mdio_meta, sizeof(*meta));
1502 	meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1503 	meta->vendor_id = 0x1095;
1504 	meta->version_minor = 0;
1505 	meta->version_major = 2;
1506 	memcpy(&meta->timestamp, &mdi->mdio_timestamp, 6);
1507 	meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize;
1508 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) {
1509 		meta->type = SII_T_RAID0;
1510 		meta->raid0_disks = vol->v_disks_count;
1511 		meta->raid1_disks = 0xff;
1512 	} else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) {
1513 		meta->type = SII_T_RAID1;
1514 		meta->raid0_disks = 0xff;
1515 		meta->raid1_disks = vol->v_disks_count;
1516 	} else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1517 		meta->type = SII_T_RAID01;
1518 		meta->raid0_disks = vol->v_disks_count / 2;
1519 		meta->raid1_disks = 2;
1520 	} else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1521 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE) {
1522 		meta->type = SII_T_JBOD;
1523 		meta->raid0_disks = vol->v_disks_count;
1524 		meta->raid1_disks = 0xff;
1525 	} else {
1526 		meta->type = SII_T_RAID5;
1527 		meta->raid0_disks = vol->v_disks_count;
1528 		meta->raid1_disks = 0xff;
1529 	}
1530 	meta->generation = mdi->mdio_generation;
1531 	meta->raid_status = vol->v_dirty ? SII_S_ONLINE : SII_S_AVAILABLE;
1532 	for (i = 0; i < vol->v_disks_count; i++) {
1533 		sd = &vol->v_subdisks[i];
1534 		if (sd->sd_state == G_RAID_SUBDISK_S_STALE ||
1535 		    sd->sd_state == G_RAID_SUBDISK_S_RESYNC)
1536 			meta->raid_status = SII_S_ONLINE;
1537 	}
1538 	meta->raid_location = mdi->mdio_location;
1539 	sii_meta_put_name(meta, vol->v_name);
1540 
1541 	/* We are done. Print meta data and store them to disks. */
1542 	if (mdi->mdio_meta != NULL)
1543 		free(mdi->mdio_meta, M_MD_SII);
1544 	mdi->mdio_meta = meta;
1545 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1546 		pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
1547 		if (disk->d_state != G_RAID_DISK_S_ACTIVE)
1548 			continue;
1549 		if (pd->pd_meta != NULL) {
1550 			free(pd->pd_meta, M_MD_SII);
1551 			pd->pd_meta = NULL;
1552 		}
1553 		pd->pd_meta = sii_meta_copy(meta);
1554 		if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1555 			if (sd->sd_state < G_RAID_SUBDISK_S_NEW)
1556 				pd->pd_meta->disk_status = SII_S_DROPPED;
1557 			else if (sd->sd_state < G_RAID_SUBDISK_S_STALE) {
1558 				pd->pd_meta->disk_status = SII_S_REBUILD;
1559 				pd->pd_meta->rebuild_lba =
1560 				    sd->sd_rebuild_pos / vol->v_sectorsize;
1561 			} else
1562 				pd->pd_meta->disk_status = SII_S_CURRENT;
1563 			if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) {
1564 				pd->pd_meta->disk_number = sd->sd_pos;
1565 				pd->pd_meta->raid0_ident = 0xff;
1566 				pd->pd_meta->raid1_ident = 0;
1567 			} else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1568 				pd->pd_meta->disk_number = sd->sd_pos / meta->raid1_disks;
1569 				pd->pd_meta->raid0_ident = sd->sd_pos % meta->raid1_disks;
1570 				pd->pd_meta->raid1_ident = sd->sd_pos / meta->raid1_disks;
1571 			} else {
1572 				pd->pd_meta->disk_number = sd->sd_pos;
1573 				pd->pd_meta->raid0_ident = 0;
1574 				pd->pd_meta->raid1_ident = 0xff;
1575 			}
1576 		}
1577 		G_RAID_DEBUG(1, "Writing SiI metadata to %s",
1578 		    g_raid_get_diskname(disk));
1579 		g_raid_md_sii_print(pd->pd_meta);
1580 		sii_meta_write(disk->d_consumer, pd->pd_meta);
1581 	}
1582 	return (0);
1583 }
1584 
1585 static int
1586 g_raid_md_fail_disk_sii(struct g_raid_md_object *md,
1587     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1588 {
1589 	struct g_raid_softc *sc;
1590 	struct g_raid_md_sii_perdisk *pd;
1591 	struct g_raid_subdisk *sd;
1592 
1593 	sc = md->mdo_softc;
1594 	pd = (struct g_raid_md_sii_perdisk *)tdisk->d_md_data;
1595 
1596 	/* We can't fail disk that is not a part of array now. */
1597 	if (pd->pd_disk_pos < 0)
1598 		return (-1);
1599 
1600 	/*
1601 	 * Mark disk as failed in metadata and try to write that metadata
1602 	 * to the disk itself to prevent it's later resurrection as STALE.
1603 	 */
1604 	if (tdisk->d_consumer) {
1605 		if (pd->pd_meta) {
1606 			pd->pd_meta->disk_status = SII_S_REMOVED;
1607 			sii_meta_write(tdisk->d_consumer, pd->pd_meta);
1608 		} else
1609 			sii_meta_erase(tdisk->d_consumer);
1610 	}
1611 
1612 	/* Change states. */
1613 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1614 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1615 		g_raid_change_subdisk_state(sd,
1616 		    G_RAID_SUBDISK_S_FAILED);
1617 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1618 		    G_RAID_EVENT_SUBDISK);
1619 	}
1620 
1621 	/* Write updated metadata to remaining disks. */
1622 	g_raid_md_write_sii(md, NULL, NULL, tdisk);
1623 
1624 	/* Check if anything left except placeholders. */
1625 	if (g_raid_ndisks(sc, -1) ==
1626 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1627 		g_raid_destroy_node(sc, 0);
1628 	else
1629 		g_raid_md_sii_refill(sc);
1630 	return (0);
1631 }
1632 
1633 static int
1634 g_raid_md_free_disk_sii(struct g_raid_md_object *md,
1635     struct g_raid_disk *disk)
1636 {
1637 	struct g_raid_md_sii_perdisk *pd;
1638 
1639 	pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data;
1640 	if (pd->pd_meta != NULL) {
1641 		free(pd->pd_meta, M_MD_SII);
1642 		pd->pd_meta = NULL;
1643 	}
1644 	free(pd, M_MD_SII);
1645 	disk->d_md_data = NULL;
1646 	return (0);
1647 }
1648 
1649 static int
1650 g_raid_md_free_sii(struct g_raid_md_object *md)
1651 {
1652 	struct g_raid_md_sii_object *mdi;
1653 
1654 	mdi = (struct g_raid_md_sii_object *)md;
1655 	if (!mdi->mdio_started) {
1656 		mdi->mdio_started = 0;
1657 		callout_stop(&mdi->mdio_start_co);
1658 		G_RAID_DEBUG1(1, md->mdo_softc,
1659 		    "root_mount_rel %p", mdi->mdio_rootmount);
1660 		root_mount_rel(mdi->mdio_rootmount);
1661 		mdi->mdio_rootmount = NULL;
1662 	}
1663 	if (mdi->mdio_meta != NULL) {
1664 		free(mdi->mdio_meta, M_MD_SII);
1665 		mdi->mdio_meta = NULL;
1666 	}
1667 	return (0);
1668 }
1669 
1670 G_RAID_MD_DECLARE(sii, "SiI");
1671