xref: /freebsd/sys/geom/raid/md_intel.c (revision ab2043b81eaba0d7d7769b4a58b2b6d17bc464a3)
1 /*-
2  * Copyright (c) 2010 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_INTEL, "md_intel_data", "GEOM_RAID Intel metadata");
47 
48 struct intel_raid_map {
49 	uint32_t	offset;
50 	uint32_t	disk_sectors;
51 	uint32_t	stripe_count;
52 	uint16_t	strip_sectors;
53 	uint8_t		status;
54 #define INTEL_S_READY           0x00
55 #define INTEL_S_UNINITIALIZED   0x01
56 #define INTEL_S_DEGRADED        0x02
57 #define INTEL_S_FAILURE         0x03
58 
59 	uint8_t		type;
60 #define INTEL_T_RAID0           0x00
61 #define INTEL_T_RAID1           0x01
62 #define INTEL_T_RAID5           0x05
63 
64 	uint8_t		total_disks;
65 	uint8_t		total_domains;
66 	uint8_t		failed_disk_num;
67 	uint8_t		ddf;
68 	uint32_t	offset_hi;
69 	uint32_t	disk_sectors_hi;
70 	uint32_t	stripe_count_hi;
71 	uint32_t	filler_2[4];
72 	uint32_t	disk_idx[1];	/* total_disks entries. */
73 #define INTEL_DI_IDX	0x00ffffff
74 #define INTEL_DI_RBLD	0x01000000
75 } __packed;
76 
77 struct intel_raid_vol {
78 	uint8_t		name[16];
79 	u_int64_t	total_sectors __packed;
80 	uint32_t	state;
81 #define INTEL_ST_BOOTABLE		0x00000001
82 #define INTEL_ST_BOOT_DEVICE		0x00000002
83 #define INTEL_ST_READ_COALESCING	0x00000004
84 #define INTEL_ST_WRITE_COALESCING	0x00000008
85 #define INTEL_ST_LAST_SHUTDOWN_DIRTY	0x00000010
86 #define INTEL_ST_HIDDEN_AT_BOOT		0x00000020
87 #define INTEL_ST_CURRENTLY_HIDDEN	0x00000040
88 #define INTEL_ST_VERIFY_AND_FIX		0x00000080
89 #define INTEL_ST_MAP_STATE_UNINIT	0x00000100
90 #define INTEL_ST_NO_AUTO_RECOVERY	0x00000200
91 #define INTEL_ST_CLONE_N_GO		0x00000400
92 #define INTEL_ST_CLONE_MAN_SYNC		0x00000800
93 #define INTEL_ST_CNG_MASTER_DISK_NUM	0x00001000
94 	uint32_t	reserved;
95 	uint8_t		migr_priority;
96 	uint8_t		num_sub_vols;
97 	uint8_t		tid;
98 	uint8_t		cng_master_disk;
99 	uint16_t	cache_policy;
100 	uint8_t		cng_state;
101 	uint8_t		cng_sub_state;
102 	uint32_t	filler_0[10];
103 
104 	uint32_t	curr_migr_unit;
105 	uint32_t	checkpoint_id;
106 	uint8_t		migr_state;
107 	uint8_t		migr_type;
108 #define INTEL_MT_INIT		0
109 #define INTEL_MT_REBUILD	1
110 #define INTEL_MT_VERIFY		2
111 #define INTEL_MT_GEN_MIGR	3
112 #define INTEL_MT_STATE_CHANGE	4
113 #define INTEL_MT_REPAIR		5
114 	uint8_t		dirty;
115 	uint8_t		fs_state;
116 	uint16_t	verify_errors;
117 	uint16_t	bad_blocks;
118 	uint32_t	curr_migr_unit_hi;
119 	uint32_t	filler_1[3];
120 	struct intel_raid_map map[1];	/* 2 entries if migr_state != 0. */
121 } __packed;
122 
123 struct intel_raid_disk {
124 #define INTEL_SERIAL_LEN	16
125 	uint8_t		serial[INTEL_SERIAL_LEN];
126 	uint32_t	sectors;
127 	uint32_t	id;
128 	uint32_t	flags;
129 #define INTEL_F_SPARE		0x01
130 #define INTEL_F_ASSIGNED	0x02
131 #define INTEL_F_FAILED		0x04
132 #define INTEL_F_ONLINE		0x08
133 	uint32_t	owner_cfg_num;
134 	uint32_t	sectors_hi;
135 	uint32_t	filler[3];
136 } __packed;
137 
138 struct intel_raid_conf {
139 	uint8_t		intel_id[24];
140 #define INTEL_MAGIC             "Intel Raid ISM Cfg Sig. "
141 
142 	uint8_t		version[6];
143 #define INTEL_VERSION_1000	"1.0.00"	/* RAID0 */
144 #define INTEL_VERSION_1100	"1.1.00"	/* RAID1 */
145 #define INTEL_VERSION_1200	"1.2.00"	/* Many volumes */
146 #define INTEL_VERSION_1201	"1.2.01"	/* 3 or 4 disks */
147 #define INTEL_VERSION_1202	"1.2.02"	/* RAID5 */
148 #define INTEL_VERSION_1204	"1.2.04"	/* 5 or 6 disks */
149 #define INTEL_VERSION_1206	"1.2.06"	/* CNG */
150 #define INTEL_VERSION_1300	"1.3.00"	/* Attributes */
151 
152 	uint8_t		dummy_0[2];
153 	uint32_t	checksum;
154 	uint32_t	config_size;
155 	uint32_t	config_id;
156 	uint32_t	generation;
157 	uint32_t	error_log_size;
158 	uint32_t	attributes;
159 #define INTEL_ATTR_RAID0	0x00000001
160 #define INTEL_ATTR_RAID1	0x00000002
161 #define INTEL_ATTR_RAID10	0x00000004
162 #define INTEL_ATTR_RAID1E	0x00000008
163 #define INTEL_ATTR_RAID5	0x00000010
164 #define INTEL_ATTR_RAIDCNG	0x00000020
165 #define INTEL_ATTR_2TB		0x20000000
166 #define INTEL_ATTR_PM		0x40000000
167 #define INTEL_ATTR_CHECKSUM	0x80000000
168 
169 	uint8_t		total_disks;
170 	uint8_t		total_volumes;
171 	uint8_t		dummy_2[2];
172 	uint32_t	filler_0[39];
173 	struct intel_raid_disk	disk[1];	/* total_disks entries. */
174 	/* Here goes total_volumes of struct intel_raid_vol. */
175 } __packed;
176 
177 #define INTEL_MAX_MD_SIZE(ndisks)				\
178     (sizeof(struct intel_raid_conf) +				\
179      sizeof(struct intel_raid_disk) * (ndisks - 1) +		\
180      sizeof(struct intel_raid_vol) * 2 +			\
181      sizeof(struct intel_raid_map) * 2 +			\
182      sizeof(uint32_t) * (ndisks - 1) * 4)
183 
184 struct g_raid_md_intel_perdisk {
185 	struct intel_raid_conf	*pd_meta;
186 	int			 pd_disk_pos;
187 	struct intel_raid_disk	 pd_disk_meta;
188 };
189 
190 struct g_raid_md_intel_object {
191 	struct g_raid_md_object	 mdio_base;
192 	uint32_t		 mdio_config_id;
193 	uint32_t		 mdio_generation;
194 	struct intel_raid_conf	*mdio_meta;
195 	struct callout		 mdio_start_co;	/* STARTING state timer. */
196 	int			 mdio_disks_present;
197 	int			 mdio_started;
198 	int			 mdio_incomplete;
199 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
200 };
201 
202 static g_raid_md_create_t g_raid_md_create_intel;
203 static g_raid_md_taste_t g_raid_md_taste_intel;
204 static g_raid_md_event_t g_raid_md_event_intel;
205 static g_raid_md_ctl_t g_raid_md_ctl_intel;
206 static g_raid_md_write_t g_raid_md_write_intel;
207 static g_raid_md_fail_disk_t g_raid_md_fail_disk_intel;
208 static g_raid_md_free_disk_t g_raid_md_free_disk_intel;
209 static g_raid_md_free_t g_raid_md_free_intel;
210 
211 static kobj_method_t g_raid_md_intel_methods[] = {
212 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_intel),
213 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_intel),
214 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_intel),
215 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_intel),
216 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_intel),
217 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_intel),
218 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_intel),
219 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_intel),
220 	{ 0, 0 }
221 };
222 
223 static struct g_raid_md_class g_raid_md_intel_class = {
224 	"Intel",
225 	g_raid_md_intel_methods,
226 	sizeof(struct g_raid_md_intel_object),
227 	.mdc_enable = 1,
228 	.mdc_priority = 100
229 };
230 
231 
232 static struct intel_raid_map *
233 intel_get_map(struct intel_raid_vol *mvol, int i)
234 {
235 	struct intel_raid_map *mmap;
236 
237 	if (i > (mvol->migr_state ? 1 : 0))
238 		return (NULL);
239 	mmap = &mvol->map[0];
240 	for (; i > 0; i--) {
241 		mmap = (struct intel_raid_map *)
242 		    &mmap->disk_idx[mmap->total_disks];
243 	}
244 	return ((struct intel_raid_map *)mmap);
245 }
246 
247 static struct intel_raid_vol *
248 intel_get_volume(struct intel_raid_conf *meta, int i)
249 {
250 	struct intel_raid_vol *mvol;
251 	struct intel_raid_map *mmap;
252 
253 	if (i > 1)
254 		return (NULL);
255 	mvol = (struct intel_raid_vol *)&meta->disk[meta->total_disks];
256 	for (; i > 0; i--) {
257 		mmap = intel_get_map(mvol, mvol->migr_state ? 1 : 0);
258 		mvol = (struct intel_raid_vol *)
259 		    &mmap->disk_idx[mmap->total_disks];
260 	}
261 	return (mvol);
262 }
263 
264 static off_t
265 intel_get_map_offset(struct intel_raid_map *mmap)
266 {
267 	off_t offset = (off_t)mmap->offset_hi << 32;
268 
269 	offset += mmap->offset;
270 	return (offset);
271 }
272 
273 static void
274 intel_set_map_offset(struct intel_raid_map *mmap, off_t offset)
275 {
276 
277 	mmap->offset = offset & 0xffffffff;
278 	mmap->offset_hi = offset >> 32;
279 }
280 
281 static off_t
282 intel_get_map_disk_sectors(struct intel_raid_map *mmap)
283 {
284 	off_t disk_sectors = (off_t)mmap->disk_sectors_hi << 32;
285 
286 	disk_sectors += mmap->disk_sectors;
287 	return (disk_sectors);
288 }
289 
290 static void
291 intel_set_map_disk_sectors(struct intel_raid_map *mmap, off_t disk_sectors)
292 {
293 
294 	mmap->disk_sectors = disk_sectors & 0xffffffff;
295 	mmap->disk_sectors_hi = disk_sectors >> 32;
296 }
297 
298 static void
299 intel_set_map_stripe_count(struct intel_raid_map *mmap, off_t stripe_count)
300 {
301 
302 	mmap->stripe_count = stripe_count & 0xffffffff;
303 	mmap->stripe_count_hi = stripe_count >> 32;
304 }
305 
306 static off_t
307 intel_get_disk_sectors(struct intel_raid_disk *disk)
308 {
309 	off_t sectors = (off_t)disk->sectors_hi << 32;
310 
311 	sectors += disk->sectors;
312 	return (sectors);
313 }
314 
315 static void
316 intel_set_disk_sectors(struct intel_raid_disk *disk, off_t sectors)
317 {
318 
319 	disk->sectors = sectors & 0xffffffff;
320 	disk->sectors_hi = sectors >> 32;
321 }
322 
323 static off_t
324 intel_get_vol_curr_migr_unit(struct intel_raid_vol *vol)
325 {
326 	off_t curr_migr_unit = (off_t)vol->curr_migr_unit_hi << 32;
327 
328 	curr_migr_unit += vol->curr_migr_unit;
329 	return (curr_migr_unit);
330 }
331 
332 static void
333 intel_set_vol_curr_migr_unit(struct intel_raid_vol *vol, off_t curr_migr_unit)
334 {
335 
336 	vol->curr_migr_unit = curr_migr_unit & 0xffffffff;
337 	vol->curr_migr_unit_hi = curr_migr_unit >> 32;
338 }
339 
340 static void
341 g_raid_md_intel_print(struct intel_raid_conf *meta)
342 {
343 	struct intel_raid_vol *mvol;
344 	struct intel_raid_map *mmap;
345 	int i, j, k;
346 
347 	if (g_raid_debug < 1)
348 		return;
349 
350 	printf("********* ATA Intel MatrixRAID Metadata *********\n");
351 	printf("intel_id            <%.24s>\n", meta->intel_id);
352 	printf("version             <%.6s>\n", meta->version);
353 	printf("checksum            0x%08x\n", meta->checksum);
354 	printf("config_size         0x%08x\n", meta->config_size);
355 	printf("config_id           0x%08x\n", meta->config_id);
356 	printf("generation          0x%08x\n", meta->generation);
357 	printf("attributes          0x%08x\n", meta->attributes);
358 	printf("total_disks         %u\n", meta->total_disks);
359 	printf("total_volumes       %u\n", meta->total_volumes);
360 	printf("DISK#   serial disk_sectors disk_sectors_hi disk_id flags\n");
361 	for (i = 0; i < meta->total_disks; i++ ) {
362 		printf("    %d   <%.16s> %u %u 0x%08x 0x%08x\n", i,
363 		    meta->disk[i].serial, meta->disk[i].sectors,
364 		    meta->disk[i].sectors_hi,
365 		    meta->disk[i].id, meta->disk[i].flags);
366 	}
367 	for (i = 0; i < meta->total_volumes; i++) {
368 		mvol = intel_get_volume(meta, i);
369 		printf(" ****** Volume %d ******\n", i);
370 		printf(" name               %.16s\n", mvol->name);
371 		printf(" total_sectors      %ju\n", mvol->total_sectors);
372 		printf(" state              %u\n", mvol->state);
373 		printf(" reserved           %u\n", mvol->reserved);
374 		printf(" curr_migr_unit     %u\n", mvol->curr_migr_unit);
375 		printf(" curr_migr_unit_hi  %u\n", mvol->curr_migr_unit_hi);
376 		printf(" checkpoint_id      %u\n", mvol->checkpoint_id);
377 		printf(" migr_state         %u\n", mvol->migr_state);
378 		printf(" migr_type          %u\n", mvol->migr_type);
379 		printf(" dirty              %u\n", mvol->dirty);
380 
381 		for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) {
382 			printf("  *** Map %d ***\n", j);
383 			mmap = intel_get_map(mvol, j);
384 			printf("  offset            %u\n", mmap->offset);
385 			printf("  offset_hi         %u\n", mmap->offset_hi);
386 			printf("  disk_sectors      %u\n", mmap->disk_sectors);
387 			printf("  disk_sectors_hi   %u\n", mmap->disk_sectors_hi);
388 			printf("  stripe_count      %u\n", mmap->stripe_count);
389 			printf("  stripe_count_hi   %u\n", mmap->stripe_count_hi);
390 			printf("  strip_sectors     %u\n", mmap->strip_sectors);
391 			printf("  status            %u\n", mmap->status);
392 			printf("  type              %u\n", mmap->type);
393 			printf("  total_disks       %u\n", mmap->total_disks);
394 			printf("  total_domains     %u\n", mmap->total_domains);
395 			printf("  failed_disk_num   %u\n", mmap->failed_disk_num);
396 			printf("  ddf               %u\n", mmap->ddf);
397 			printf("  disk_idx         ");
398 			for (k = 0; k < mmap->total_disks; k++)
399 				printf(" 0x%08x", mmap->disk_idx[k]);
400 			printf("\n");
401 		}
402 	}
403 	printf("=================================================\n");
404 }
405 
406 static struct intel_raid_conf *
407 intel_meta_copy(struct intel_raid_conf *meta)
408 {
409 	struct intel_raid_conf *nmeta;
410 
411 	nmeta = malloc(meta->config_size, M_MD_INTEL, M_WAITOK);
412 	memcpy(nmeta, meta, meta->config_size);
413 	return (nmeta);
414 }
415 
416 static int
417 intel_meta_find_disk(struct intel_raid_conf *meta, char *serial)
418 {
419 	int pos;
420 
421 	for (pos = 0; pos < meta->total_disks; pos++) {
422 		if (strncmp(meta->disk[pos].serial,
423 		    serial, INTEL_SERIAL_LEN) == 0)
424 			return (pos);
425 	}
426 	return (-1);
427 }
428 
429 static struct intel_raid_conf *
430 intel_meta_read(struct g_consumer *cp)
431 {
432 	struct g_provider *pp;
433 	struct intel_raid_conf *meta;
434 	struct intel_raid_vol *mvol;
435 	struct intel_raid_map *mmap;
436 	char *buf;
437 	int error, i, j, k, left, size;
438 	uint32_t checksum, *ptr;
439 
440 	pp = cp->provider;
441 
442 	/* Read the anchor sector. */
443 	buf = g_read_data(cp,
444 	    pp->mediasize - pp->sectorsize * 2, pp->sectorsize, &error);
445 	if (buf == NULL) {
446 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
447 		    pp->name, error);
448 		return (NULL);
449 	}
450 	meta = (struct intel_raid_conf *)buf;
451 
452 	/* Check if this is an Intel RAID struct */
453 	if (strncmp(meta->intel_id, INTEL_MAGIC, strlen(INTEL_MAGIC))) {
454 		G_RAID_DEBUG(1, "Intel signature check failed on %s", pp->name);
455 		g_free(buf);
456 		return (NULL);
457 	}
458 	if (meta->config_size > 65536 ||
459 	    meta->config_size < sizeof(struct intel_raid_conf)) {
460 		G_RAID_DEBUG(1, "Intel metadata size looks wrong: %d",
461 		    meta->config_size);
462 		g_free(buf);
463 		return (NULL);
464 	}
465 	size = meta->config_size;
466 	meta = malloc(size, M_MD_INTEL, M_WAITOK);
467 	memcpy(meta, buf, min(size, pp->sectorsize));
468 	g_free(buf);
469 
470 	/* Read all the rest, if needed. */
471 	if (meta->config_size > pp->sectorsize) {
472 		left = (meta->config_size - 1) / pp->sectorsize;
473 		buf = g_read_data(cp,
474 		    pp->mediasize - pp->sectorsize * (2 + left),
475 		    pp->sectorsize * left, &error);
476 		if (buf == NULL) {
477 			G_RAID_DEBUG(1, "Cannot read remaining metadata"
478 			    " part from %s (error=%d).",
479 			    pp->name, error);
480 			free(meta, M_MD_INTEL);
481 			return (NULL);
482 		}
483 		memcpy(((char *)meta) + pp->sectorsize, buf,
484 		    pp->sectorsize * left);
485 		g_free(buf);
486 	}
487 
488 	/* Check metadata checksum. */
489 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0;
490 	    i < (meta->config_size / sizeof(uint32_t)); i++) {
491 		checksum += *ptr++;
492 	}
493 	checksum -= meta->checksum;
494 	if (checksum != meta->checksum) {
495 		G_RAID_DEBUG(1, "Intel checksum check failed on %s", pp->name);
496 		free(meta, M_MD_INTEL);
497 		return (NULL);
498 	}
499 
500 	/* Validate metadata size. */
501 	size = sizeof(struct intel_raid_conf) +
502 	    sizeof(struct intel_raid_disk) * (meta->total_disks - 1) +
503 	    sizeof(struct intel_raid_vol) * meta->total_volumes;
504 	if (size > meta->config_size) {
505 badsize:
506 		G_RAID_DEBUG(1, "Intel metadata size incorrect %d < %d",
507 		    meta->config_size, size);
508 		free(meta, M_MD_INTEL);
509 		return (NULL);
510 	}
511 	for (i = 0; i < meta->total_volumes; i++) {
512 		mvol = intel_get_volume(meta, i);
513 		mmap = intel_get_map(mvol, 0);
514 		size += 4 * (mmap->total_disks - 1);
515 		if (size > meta->config_size)
516 			goto badsize;
517 		if (mvol->migr_state) {
518 			size += sizeof(struct intel_raid_map);
519 			if (size > meta->config_size)
520 				goto badsize;
521 			mmap = intel_get_map(mvol, 1);
522 			size += 4 * (mmap->total_disks - 1);
523 			if (size > meta->config_size)
524 				goto badsize;
525 		}
526 	}
527 
528 	/* Validate disk indexes. */
529 	for (i = 0; i < meta->total_volumes; i++) {
530 		mvol = intel_get_volume(meta, i);
531 		for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) {
532 			mmap = intel_get_map(mvol, j);
533 			for (k = 0; k < mmap->total_disks; k++) {
534 				if ((mmap->disk_idx[k] & INTEL_DI_IDX) >
535 				    meta->total_disks) {
536 					G_RAID_DEBUG(1, "Intel metadata disk"
537 					    " index %d too big (>%d)",
538 					    mmap->disk_idx[k] & INTEL_DI_IDX,
539 					    meta->total_disks);
540 					free(meta, M_MD_INTEL);
541 					return (NULL);
542 				}
543 			}
544 		}
545 	}
546 
547 	/* Validate migration types. */
548 	for (i = 0; i < meta->total_volumes; i++) {
549 		mvol = intel_get_volume(meta, i);
550 		if (mvol->migr_state &&
551 		    mvol->migr_type != INTEL_MT_INIT &&
552 		    mvol->migr_type != INTEL_MT_REBUILD &&
553 		    mvol->migr_type != INTEL_MT_VERIFY &&
554 		    mvol->migr_type != INTEL_MT_REPAIR) {
555 			G_RAID_DEBUG(1, "Intel metadata has unsupported"
556 			    " migration type %d", mvol->migr_type);
557 			free(meta, M_MD_INTEL);
558 			return (NULL);
559 		}
560 	}
561 
562 	return (meta);
563 }
564 
565 static int
566 intel_meta_write(struct g_consumer *cp, struct intel_raid_conf *meta)
567 {
568 	struct g_provider *pp;
569 	char *buf;
570 	int error, i, sectors;
571 	uint32_t checksum, *ptr;
572 
573 	pp = cp->provider;
574 
575 	/* Recalculate checksum for case if metadata were changed. */
576 	meta->checksum = 0;
577 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0;
578 	    i < (meta->config_size / sizeof(uint32_t)); i++) {
579 		checksum += *ptr++;
580 	}
581 	meta->checksum = checksum;
582 
583 	/* Create and fill buffer. */
584 	sectors = (meta->config_size + pp->sectorsize - 1) / pp->sectorsize;
585 	buf = malloc(sectors * pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO);
586 	if (sectors > 1) {
587 		memcpy(buf, ((char *)meta) + pp->sectorsize,
588 		    (sectors - 1) * pp->sectorsize);
589 	}
590 	memcpy(buf + (sectors - 1) * pp->sectorsize, meta, pp->sectorsize);
591 
592 	error = g_write_data(cp,
593 	    pp->mediasize - pp->sectorsize * (1 + sectors),
594 	    buf, pp->sectorsize * sectors);
595 	if (error != 0) {
596 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
597 		    pp->name, error);
598 	}
599 
600 	free(buf, M_MD_INTEL);
601 	return (error);
602 }
603 
604 static int
605 intel_meta_erase(struct g_consumer *cp)
606 {
607 	struct g_provider *pp;
608 	char *buf;
609 	int error;
610 
611 	pp = cp->provider;
612 	buf = malloc(pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO);
613 	error = g_write_data(cp,
614 	    pp->mediasize - 2 * pp->sectorsize,
615 	    buf, pp->sectorsize);
616 	if (error != 0) {
617 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
618 		    pp->name, error);
619 	}
620 	free(buf, M_MD_INTEL);
621 	return (error);
622 }
623 
624 static int
625 intel_meta_write_spare(struct g_consumer *cp, struct intel_raid_disk *d)
626 {
627 	struct intel_raid_conf *meta;
628 	int error;
629 
630 	/* Fill anchor and single disk. */
631 	meta = malloc(INTEL_MAX_MD_SIZE(1), M_MD_INTEL, M_WAITOK | M_ZERO);
632 	memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1);
633 	memcpy(&meta->version[0], INTEL_VERSION_1000,
634 	    sizeof(INTEL_VERSION_1000) - 1);
635 	meta->config_size = INTEL_MAX_MD_SIZE(1);
636 	meta->config_id = arc4random();
637 	meta->generation = 1;
638 	meta->total_disks = 1;
639 	meta->disk[0] = *d;
640 	error = intel_meta_write(cp, meta);
641 	free(meta, M_MD_INTEL);
642 	return (error);
643 }
644 
645 static struct g_raid_disk *
646 g_raid_md_intel_get_disk(struct g_raid_softc *sc, int id)
647 {
648 	struct g_raid_disk	*disk;
649 	struct g_raid_md_intel_perdisk *pd;
650 
651 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
652 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
653 		if (pd->pd_disk_pos == id)
654 			break;
655 	}
656 	return (disk);
657 }
658 
659 static int
660 g_raid_md_intel_supported(int level, int qual, int disks, int force)
661 {
662 
663 	switch (level) {
664 	case G_RAID_VOLUME_RL_RAID0:
665 		if (disks < 1)
666 			return (0);
667 		if (!force && (disks < 2 || disks > 6))
668 			return (0);
669 		break;
670 	case G_RAID_VOLUME_RL_RAID1:
671 		if (disks < 1)
672 			return (0);
673 		if (!force && (disks != 2))
674 			return (0);
675 		break;
676 	case G_RAID_VOLUME_RL_RAID1E:
677 		if (disks < 2)
678 			return (0);
679 		if (!force && (disks != 4))
680 			return (0);
681 		break;
682 	case G_RAID_VOLUME_RL_RAID5:
683 		if (disks < 3)
684 			return (0);
685 		if (!force && disks > 6)
686 			return (0);
687 		if (qual != G_RAID_VOLUME_RLQ_R5LA)
688 			return (0);
689 		break;
690 	default:
691 		return (0);
692 	}
693 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
694 		return (0);
695 	return (1);
696 }
697 
698 static struct g_raid_volume *
699 g_raid_md_intel_get_volume(struct g_raid_softc *sc, int id)
700 {
701 	struct g_raid_volume	*mvol;
702 
703 	TAILQ_FOREACH(mvol, &sc->sc_volumes, v_next) {
704 		if ((intptr_t)(mvol->v_md_data) == id)
705 			break;
706 	}
707 	return (mvol);
708 }
709 
710 static int
711 g_raid_md_intel_start_disk(struct g_raid_disk *disk)
712 {
713 	struct g_raid_softc *sc;
714 	struct g_raid_subdisk *sd, *tmpsd;
715 	struct g_raid_disk *olddisk, *tmpdisk;
716 	struct g_raid_md_object *md;
717 	struct g_raid_md_intel_object *mdi;
718 	struct g_raid_md_intel_perdisk *pd, *oldpd;
719 	struct intel_raid_conf *meta;
720 	struct intel_raid_vol *mvol;
721 	struct intel_raid_map *mmap0, *mmap1;
722 	int disk_pos, resurrection = 0;
723 
724 	sc = disk->d_softc;
725 	md = sc->sc_md;
726 	mdi = (struct g_raid_md_intel_object *)md;
727 	meta = mdi->mdio_meta;
728 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
729 	olddisk = NULL;
730 
731 	/* Find disk position in metadata by it's serial. */
732 	disk_pos = intel_meta_find_disk(meta, pd->pd_disk_meta.serial);
733 	if (disk_pos < 0) {
734 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
735 		/* Failed stale disk is useless for us. */
736 		if (pd->pd_disk_meta.flags & INTEL_F_FAILED) {
737 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
738 			return (0);
739 		}
740 		/* If we are in the start process, that's all for now. */
741 		if (!mdi->mdio_started)
742 			goto nofit;
743 		/*
744 		 * If we have already started - try to get use of the disk.
745 		 * Try to replace OFFLINE disks first, then FAILED.
746 		 */
747 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
748 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
749 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
750 				continue;
751 			/* Make sure this disk is big enough. */
752 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
753 				off_t disk_sectors =
754 				    intel_get_disk_sectors(&pd->pd_disk_meta);
755 
756 				if (sd->sd_offset + sd->sd_size + 4096 >
757 				    disk_sectors * 512) {
758 					G_RAID_DEBUG1(1, sc,
759 					    "Disk too small (%llu < %llu)",
760 					    (unsigned long long)
761 					    disk_sectors * 512,
762 					    (unsigned long long)
763 					    sd->sd_offset + sd->sd_size + 4096);
764 					break;
765 				}
766 			}
767 			if (sd != NULL)
768 				continue;
769 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
770 				olddisk = tmpdisk;
771 				break;
772 			} else if (olddisk == NULL)
773 				olddisk = tmpdisk;
774 		}
775 		if (olddisk == NULL) {
776 nofit:
777 			if (pd->pd_disk_meta.flags & INTEL_F_SPARE) {
778 				g_raid_change_disk_state(disk,
779 				    G_RAID_DISK_S_SPARE);
780 				return (1);
781 			} else {
782 				g_raid_change_disk_state(disk,
783 				    G_RAID_DISK_S_STALE);
784 				return (0);
785 			}
786 		}
787 		oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data;
788 		disk_pos = oldpd->pd_disk_pos;
789 		resurrection = 1;
790 	}
791 
792 	if (olddisk == NULL) {
793 		/* Find placeholder by position. */
794 		olddisk = g_raid_md_intel_get_disk(sc, disk_pos);
795 		if (olddisk == NULL)
796 			panic("No disk at position %d!", disk_pos);
797 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
798 			G_RAID_DEBUG1(1, sc, "More then one disk for pos %d",
799 			    disk_pos);
800 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
801 			return (0);
802 		}
803 		oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data;
804 	}
805 
806 	/* Replace failed disk or placeholder with new disk. */
807 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
808 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
809 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
810 		sd->sd_disk = disk;
811 	}
812 	oldpd->pd_disk_pos = -2;
813 	pd->pd_disk_pos = disk_pos;
814 
815 	/* If it was placeholder -- destroy it. */
816 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
817 		g_raid_destroy_disk(olddisk);
818 	} else {
819 		/* Otherwise, make it STALE_FAILED. */
820 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
821 		/* Update global metadata just in case. */
822 		memcpy(&meta->disk[disk_pos], &pd->pd_disk_meta,
823 		    sizeof(struct intel_raid_disk));
824 	}
825 
826 	/* Welcome the new disk. */
827 	if (resurrection)
828 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
829 	else if (meta->disk[disk_pos].flags & INTEL_F_FAILED)
830 		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
831 	else if (meta->disk[disk_pos].flags & INTEL_F_SPARE)
832 		g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
833 	else
834 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
835 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
836 		mvol = intel_get_volume(meta,
837 		    (uintptr_t)(sd->sd_volume->v_md_data));
838 		mmap0 = intel_get_map(mvol, 0);
839 		if (mvol->migr_state)
840 			mmap1 = intel_get_map(mvol, 1);
841 		else
842 			mmap1 = mmap0;
843 
844 		if (resurrection) {
845 			/* Stale disk, almost same as new. */
846 			g_raid_change_subdisk_state(sd,
847 			    G_RAID_SUBDISK_S_NEW);
848 		} else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) {
849 			/* Failed disk, almost useless. */
850 			g_raid_change_subdisk_state(sd,
851 			    G_RAID_SUBDISK_S_FAILED);
852 		} else if (mvol->migr_state == 0) {
853 			if (mmap0->status == INTEL_S_UNINITIALIZED) {
854 				/* Freshly created uninitialized volume. */
855 				g_raid_change_subdisk_state(sd,
856 				    G_RAID_SUBDISK_S_UNINITIALIZED);
857 			} else if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
858 				/* Freshly inserted disk. */
859 				g_raid_change_subdisk_state(sd,
860 				    G_RAID_SUBDISK_S_NEW);
861 			} else if (mvol->dirty) {
862 				/* Dirty volume (unclean shutdown). */
863 				g_raid_change_subdisk_state(sd,
864 				    G_RAID_SUBDISK_S_STALE);
865 			} else {
866 				/* Up to date disk. */
867 				g_raid_change_subdisk_state(sd,
868 				    G_RAID_SUBDISK_S_ACTIVE);
869 			}
870 		} else if (mvol->migr_type == INTEL_MT_INIT ||
871 			   mvol->migr_type == INTEL_MT_REBUILD) {
872 			if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
873 				/* Freshly inserted disk. */
874 				g_raid_change_subdisk_state(sd,
875 				    G_RAID_SUBDISK_S_NEW);
876 			} else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
877 				/* Rebuilding disk. */
878 				g_raid_change_subdisk_state(sd,
879 				    G_RAID_SUBDISK_S_REBUILD);
880 				if (mvol->dirty) {
881 					sd->sd_rebuild_pos = 0;
882 				} else {
883 					sd->sd_rebuild_pos =
884 					    intel_get_vol_curr_migr_unit(mvol) *
885 					    sd->sd_volume->v_strip_size *
886 					    mmap0->total_domains;
887 				}
888 			} else if (mvol->dirty) {
889 				/* Dirty volume (unclean shutdown). */
890 				g_raid_change_subdisk_state(sd,
891 				    G_RAID_SUBDISK_S_STALE);
892 			} else {
893 				/* Up to date disk. */
894 				g_raid_change_subdisk_state(sd,
895 				    G_RAID_SUBDISK_S_ACTIVE);
896 			}
897 		} else if (mvol->migr_type == INTEL_MT_VERIFY ||
898 			   mvol->migr_type == INTEL_MT_REPAIR) {
899 			if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
900 				/* Freshly inserted disk. */
901 				g_raid_change_subdisk_state(sd,
902 				    G_RAID_SUBDISK_S_NEW);
903 			} else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
904 				/* Resyncing disk. */
905 				g_raid_change_subdisk_state(sd,
906 				    G_RAID_SUBDISK_S_RESYNC);
907 				if (mvol->dirty) {
908 					sd->sd_rebuild_pos = 0;
909 				} else {
910 					sd->sd_rebuild_pos =
911 					    intel_get_vol_curr_migr_unit(mvol) *
912 					    sd->sd_volume->v_strip_size *
913 					    mmap0->total_domains;
914 				}
915 			} else if (mvol->dirty) {
916 				/* Dirty volume (unclean shutdown). */
917 				g_raid_change_subdisk_state(sd,
918 				    G_RAID_SUBDISK_S_STALE);
919 			} else {
920 				/* Up to date disk. */
921 				g_raid_change_subdisk_state(sd,
922 				    G_RAID_SUBDISK_S_ACTIVE);
923 			}
924 		}
925 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
926 		    G_RAID_EVENT_SUBDISK);
927 	}
928 
929 	/* Update status of our need for spare. */
930 	if (mdi->mdio_started) {
931 		mdi->mdio_incomplete =
932 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
933 		     meta->total_disks);
934 	}
935 
936 	return (resurrection);
937 }
938 
939 static void
940 g_disk_md_intel_retaste(void *arg, int pending)
941 {
942 
943 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
944 	g_retaste(&g_raid_class);
945 	free(arg, M_MD_INTEL);
946 }
947 
948 static void
949 g_raid_md_intel_refill(struct g_raid_softc *sc)
950 {
951 	struct g_raid_md_object *md;
952 	struct g_raid_md_intel_object *mdi;
953 	struct intel_raid_conf *meta;
954 	struct g_raid_disk *disk;
955 	struct task *task;
956 	int update, na;
957 
958 	md = sc->sc_md;
959 	mdi = (struct g_raid_md_intel_object *)md;
960 	meta = mdi->mdio_meta;
961 	update = 0;
962 	do {
963 		/* Make sure we miss anything. */
964 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
965 		if (na == meta->total_disks)
966 			break;
967 
968 		G_RAID_DEBUG1(1, md->mdo_softc,
969 		    "Array is not complete (%d of %d), "
970 		    "trying to refill.", na, meta->total_disks);
971 
972 		/* Try to get use some of STALE disks. */
973 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
974 			if (disk->d_state == G_RAID_DISK_S_STALE) {
975 				update += g_raid_md_intel_start_disk(disk);
976 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
977 					break;
978 			}
979 		}
980 		if (disk != NULL)
981 			continue;
982 
983 		/* Try to get use some of SPARE disks. */
984 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
985 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
986 				update += g_raid_md_intel_start_disk(disk);
987 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
988 					break;
989 			}
990 		}
991 	} while (disk != NULL);
992 
993 	/* Write new metadata if we changed something. */
994 	if (update) {
995 		g_raid_md_write_intel(md, NULL, NULL, NULL);
996 		meta = mdi->mdio_meta;
997 	}
998 
999 	/* Update status of our need for spare. */
1000 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
1001 	    meta->total_disks);
1002 
1003 	/* Request retaste hoping to find spare. */
1004 	if (mdi->mdio_incomplete) {
1005 		task = malloc(sizeof(struct task),
1006 		    M_MD_INTEL, M_WAITOK | M_ZERO);
1007 		TASK_INIT(task, 0, g_disk_md_intel_retaste, task);
1008 		taskqueue_enqueue(taskqueue_swi, task);
1009 	}
1010 }
1011 
1012 static void
1013 g_raid_md_intel_start(struct g_raid_softc *sc)
1014 {
1015 	struct g_raid_md_object *md;
1016 	struct g_raid_md_intel_object *mdi;
1017 	struct g_raid_md_intel_perdisk *pd;
1018 	struct intel_raid_conf *meta;
1019 	struct intel_raid_vol *mvol;
1020 	struct intel_raid_map *mmap;
1021 	struct g_raid_volume *vol;
1022 	struct g_raid_subdisk *sd;
1023 	struct g_raid_disk *disk;
1024 	int i, j, disk_pos;
1025 
1026 	md = sc->sc_md;
1027 	mdi = (struct g_raid_md_intel_object *)md;
1028 	meta = mdi->mdio_meta;
1029 
1030 	/* Create volumes and subdisks. */
1031 	for (i = 0; i < meta->total_volumes; i++) {
1032 		mvol = intel_get_volume(meta, i);
1033 		mmap = intel_get_map(mvol, 0);
1034 		vol = g_raid_create_volume(sc, mvol->name, -1);
1035 		vol->v_md_data = (void *)(intptr_t)i;
1036 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
1037 		if (mmap->type == INTEL_T_RAID0)
1038 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
1039 		else if (mmap->type == INTEL_T_RAID1 &&
1040 		    mmap->total_domains >= 2 &&
1041 		    mmap->total_domains <= mmap->total_disks) {
1042 			/* Assume total_domains is correct. */
1043 			if (mmap->total_domains == mmap->total_disks)
1044 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
1045 			else
1046 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
1047 		} else if (mmap->type == INTEL_T_RAID1) {
1048 			/* total_domains looks wrong. */
1049 			if (mmap->total_disks <= 2)
1050 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
1051 			else
1052 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
1053 		} else if (mmap->type == INTEL_T_RAID5) {
1054 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
1055 			vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
1056 		} else
1057 			vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1058 		vol->v_strip_size = (u_int)mmap->strip_sectors * 512; //ZZZ
1059 		vol->v_disks_count = mmap->total_disks;
1060 		vol->v_mediasize = (off_t)mvol->total_sectors * 512; //ZZZ
1061 		vol->v_sectorsize = 512; //ZZZ
1062 		for (j = 0; j < vol->v_disks_count; j++) {
1063 			sd = &vol->v_subdisks[j];
1064 			sd->sd_offset = intel_get_map_offset(mmap) * 512; //ZZZ
1065 			sd->sd_size = intel_get_map_disk_sectors(mmap) * 512; //ZZZ
1066 		}
1067 		g_raid_start_volume(vol);
1068 	}
1069 
1070 	/* Create disk placeholders to store data for later writing. */
1071 	for (disk_pos = 0; disk_pos < meta->total_disks; disk_pos++) {
1072 		pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
1073 		pd->pd_disk_pos = disk_pos;
1074 		pd->pd_disk_meta = meta->disk[disk_pos];
1075 		disk = g_raid_create_disk(sc);
1076 		disk->d_md_data = (void *)pd;
1077 		disk->d_state = G_RAID_DISK_S_OFFLINE;
1078 		for (i = 0; i < meta->total_volumes; i++) {
1079 			mvol = intel_get_volume(meta, i);
1080 			mmap = intel_get_map(mvol, 0);
1081 			for (j = 0; j < mmap->total_disks; j++) {
1082 				if ((mmap->disk_idx[j] & INTEL_DI_IDX) == disk_pos)
1083 					break;
1084 			}
1085 			if (j == mmap->total_disks)
1086 				continue;
1087 			vol = g_raid_md_intel_get_volume(sc, i);
1088 			sd = &vol->v_subdisks[j];
1089 			sd->sd_disk = disk;
1090 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1091 		}
1092 	}
1093 
1094 	/* Make all disks found till the moment take their places. */
1095 	do {
1096 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1097 			if (disk->d_state == G_RAID_DISK_S_NONE) {
1098 				g_raid_md_intel_start_disk(disk);
1099 				break;
1100 			}
1101 		}
1102 	} while (disk != NULL);
1103 
1104 	mdi->mdio_started = 1;
1105 	G_RAID_DEBUG1(0, sc, "Array started.");
1106 	g_raid_md_write_intel(md, NULL, NULL, NULL);
1107 
1108 	/* Pickup any STALE/SPARE disks to refill array if needed. */
1109 	g_raid_md_intel_refill(sc);
1110 
1111 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1112 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1113 		    G_RAID_EVENT_VOLUME);
1114 	}
1115 
1116 	callout_stop(&mdi->mdio_start_co);
1117 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
1118 	root_mount_rel(mdi->mdio_rootmount);
1119 	mdi->mdio_rootmount = NULL;
1120 }
1121 
1122 static void
1123 g_raid_md_intel_new_disk(struct g_raid_disk *disk)
1124 {
1125 	struct g_raid_softc *sc;
1126 	struct g_raid_md_object *md;
1127 	struct g_raid_md_intel_object *mdi;
1128 	struct intel_raid_conf *pdmeta;
1129 	struct g_raid_md_intel_perdisk *pd;
1130 
1131 	sc = disk->d_softc;
1132 	md = sc->sc_md;
1133 	mdi = (struct g_raid_md_intel_object *)md;
1134 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
1135 	pdmeta = pd->pd_meta;
1136 
1137 	if (mdi->mdio_started) {
1138 		if (g_raid_md_intel_start_disk(disk))
1139 			g_raid_md_write_intel(md, NULL, NULL, NULL);
1140 	} else {
1141 		/* If we haven't started yet - check metadata freshness. */
1142 		if (mdi->mdio_meta == NULL ||
1143 		    ((int32_t)(pdmeta->generation - mdi->mdio_generation)) > 0) {
1144 			G_RAID_DEBUG1(1, sc, "Newer disk");
1145 			if (mdi->mdio_meta != NULL)
1146 				free(mdi->mdio_meta, M_MD_INTEL);
1147 			mdi->mdio_meta = intel_meta_copy(pdmeta);
1148 			mdi->mdio_generation = mdi->mdio_meta->generation;
1149 			mdi->mdio_disks_present = 1;
1150 		} else if (pdmeta->generation == mdi->mdio_generation) {
1151 			mdi->mdio_disks_present++;
1152 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
1153 			    mdi->mdio_disks_present,
1154 			    mdi->mdio_meta->total_disks);
1155 		} else {
1156 			G_RAID_DEBUG1(1, sc, "Older disk");
1157 		}
1158 		/* If we collected all needed disks - start array. */
1159 		if (mdi->mdio_disks_present == mdi->mdio_meta->total_disks)
1160 			g_raid_md_intel_start(sc);
1161 	}
1162 }
1163 
1164 static void
1165 g_raid_intel_go(void *arg)
1166 {
1167 	struct g_raid_softc *sc;
1168 	struct g_raid_md_object *md;
1169 	struct g_raid_md_intel_object *mdi;
1170 
1171 	sc = arg;
1172 	md = sc->sc_md;
1173 	mdi = (struct g_raid_md_intel_object *)md;
1174 	if (!mdi->mdio_started) {
1175 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
1176 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
1177 	}
1178 }
1179 
1180 static int
1181 g_raid_md_create_intel(struct g_raid_md_object *md, struct g_class *mp,
1182     struct g_geom **gp)
1183 {
1184 	struct g_raid_softc *sc;
1185 	struct g_raid_md_intel_object *mdi;
1186 	char name[16];
1187 
1188 	mdi = (struct g_raid_md_intel_object *)md;
1189 	mdi->mdio_config_id = arc4random();
1190 	mdi->mdio_generation = 0;
1191 	snprintf(name, sizeof(name), "Intel-%08x", mdi->mdio_config_id);
1192 	sc = g_raid_create_node(mp, name, md);
1193 	if (sc == NULL)
1194 		return (G_RAID_MD_TASTE_FAIL);
1195 	md->mdo_softc = sc;
1196 	*gp = sc->sc_geom;
1197 	return (G_RAID_MD_TASTE_NEW);
1198 }
1199 
1200 /*
1201  * Return the last N characters of the serial label.  The Linux and
1202  * ataraid(7) code always uses the last 16 characters of the label to
1203  * store into the Intel meta format.  Generalize this to N characters
1204  * since that's easy.  Labels can be up to 20 characters for SATA drives
1205  * and up 251 characters for SAS drives.  Since intel controllers don't
1206  * support SAS drives, just stick with the SATA limits for stack friendliness.
1207  */
1208 static int
1209 g_raid_md_get_label(struct g_consumer *cp, char *serial, int serlen)
1210 {
1211 	char serial_buffer[24];
1212 	int len, error;
1213 
1214 	len = sizeof(serial_buffer);
1215 	error = g_io_getattr("GEOM::ident", cp, &len, serial_buffer);
1216 	if (error != 0)
1217 		return (error);
1218 	len = strlen(serial_buffer);
1219 	if (len > serlen)
1220 		len -= serlen;
1221 	else
1222 		len = 0;
1223 	strncpy(serial, serial_buffer + len, serlen);
1224 	return (0);
1225 }
1226 
1227 static int
1228 g_raid_md_taste_intel(struct g_raid_md_object *md, struct g_class *mp,
1229                               struct g_consumer *cp, struct g_geom **gp)
1230 {
1231 	struct g_consumer *rcp;
1232 	struct g_provider *pp;
1233 	struct g_raid_md_intel_object *mdi, *mdi1;
1234 	struct g_raid_softc *sc;
1235 	struct g_raid_disk *disk;
1236 	struct intel_raid_conf *meta;
1237 	struct g_raid_md_intel_perdisk *pd;
1238 	struct g_geom *geom;
1239 	int error, disk_pos, result, spare, len;
1240 	char serial[INTEL_SERIAL_LEN];
1241 	char name[16];
1242 	uint16_t vendor;
1243 
1244 	G_RAID_DEBUG(1, "Tasting Intel on %s", cp->provider->name);
1245 	mdi = (struct g_raid_md_intel_object *)md;
1246 	pp = cp->provider;
1247 
1248 	/* Read metadata from device. */
1249 	meta = NULL;
1250 	vendor = 0xffff;
1251 	disk_pos = 0;
1252 	if (g_access(cp, 1, 0, 0) != 0)
1253 		return (G_RAID_MD_TASTE_FAIL);
1254 	g_topology_unlock();
1255 	error = g_raid_md_get_label(cp, serial, sizeof(serial));
1256 	if (error != 0) {
1257 		G_RAID_DEBUG(1, "Cannot get serial number from %s (error=%d).",
1258 		    pp->name, error);
1259 		goto fail2;
1260 	}
1261 	len = 2;
1262 	if (pp->geom->rank == 1)
1263 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1264 	meta = intel_meta_read(cp);
1265 	g_topology_lock();
1266 	g_access(cp, -1, 0, 0);
1267 	if (meta == NULL) {
1268 		if (g_raid_aggressive_spare) {
1269 			if (vendor != 0x8086) {
1270 				G_RAID_DEBUG(1,
1271 				    "Intel vendor mismatch 0x%04x != 0x8086",
1272 				    vendor);
1273 			} else {
1274 				G_RAID_DEBUG(1,
1275 				    "No Intel metadata, forcing spare.");
1276 				spare = 2;
1277 				goto search;
1278 			}
1279 		}
1280 		return (G_RAID_MD_TASTE_FAIL);
1281 	}
1282 
1283 	/* Check this disk position in obtained metadata. */
1284 	disk_pos = intel_meta_find_disk(meta, serial);
1285 	if (disk_pos < 0) {
1286 		G_RAID_DEBUG(1, "Intel serial '%s' not found", serial);
1287 		goto fail1;
1288 	}
1289 	if (intel_get_disk_sectors(&meta->disk[disk_pos]) !=
1290 	    (pp->mediasize / pp->sectorsize)) {
1291 		G_RAID_DEBUG(1, "Intel size mismatch %ju != %ju",
1292 		    intel_get_disk_sectors(&meta->disk[disk_pos]),
1293 		    (off_t)(pp->mediasize / pp->sectorsize));
1294 		goto fail1;
1295 	}
1296 
1297 	/* Metadata valid. Print it. */
1298 	g_raid_md_intel_print(meta);
1299 	G_RAID_DEBUG(1, "Intel disk position %d", disk_pos);
1300 	spare = meta->disk[disk_pos].flags & INTEL_F_SPARE;
1301 
1302 search:
1303 	/* Search for matching node. */
1304 	sc = NULL;
1305 	mdi1 = NULL;
1306 	LIST_FOREACH(geom, &mp->geom, geom) {
1307 		sc = geom->softc;
1308 		if (sc == NULL)
1309 			continue;
1310 		if (sc->sc_stopping != 0)
1311 			continue;
1312 		if (sc->sc_md->mdo_class != md->mdo_class)
1313 			continue;
1314 		mdi1 = (struct g_raid_md_intel_object *)sc->sc_md;
1315 		if (spare) {
1316 			if (mdi1->mdio_incomplete)
1317 				break;
1318 		} else {
1319 			if (mdi1->mdio_config_id == meta->config_id)
1320 				break;
1321 		}
1322 	}
1323 
1324 	/* Found matching node. */
1325 	if (geom != NULL) {
1326 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1327 		result = G_RAID_MD_TASTE_EXISTING;
1328 
1329 	} else if (spare) { /* Not found needy node -- left for later. */
1330 		G_RAID_DEBUG(1, "Spare is not needed at this time");
1331 		goto fail1;
1332 
1333 	} else { /* Not found matching node -- create one. */
1334 		result = G_RAID_MD_TASTE_NEW;
1335 		mdi->mdio_config_id = meta->config_id;
1336 		snprintf(name, sizeof(name), "Intel-%08x", meta->config_id);
1337 		sc = g_raid_create_node(mp, name, md);
1338 		md->mdo_softc = sc;
1339 		geom = sc->sc_geom;
1340 		callout_init(&mdi->mdio_start_co, 1);
1341 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
1342 		    g_raid_intel_go, sc);
1343 		mdi->mdio_rootmount = root_mount_hold("GRAID-Intel");
1344 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
1345 	}
1346 
1347 	rcp = g_new_consumer(geom);
1348 	g_attach(rcp, pp);
1349 	if (g_access(rcp, 1, 1, 1) != 0)
1350 		; //goto fail1;
1351 
1352 	g_topology_unlock();
1353 	sx_xlock(&sc->sc_lock);
1354 
1355 	pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
1356 	pd->pd_meta = meta;
1357 	pd->pd_disk_pos = -1;
1358 	if (spare == 2) {
1359 		memcpy(&pd->pd_disk_meta.serial[0], serial, INTEL_SERIAL_LEN);
1360 		intel_set_disk_sectors(&pd->pd_disk_meta,
1361 		    pp->mediasize / pp->sectorsize);
1362 		pd->pd_disk_meta.id = 0;
1363 		pd->pd_disk_meta.flags = INTEL_F_SPARE;
1364 	} else {
1365 		pd->pd_disk_meta = meta->disk[disk_pos];
1366 	}
1367 	disk = g_raid_create_disk(sc);
1368 	disk->d_md_data = (void *)pd;
1369 	disk->d_consumer = rcp;
1370 	rcp->private = disk;
1371 
1372 	/* Read kernel dumping information. */
1373 	disk->d_kd.offset = 0;
1374 	disk->d_kd.length = OFF_MAX;
1375 	len = sizeof(disk->d_kd);
1376 	error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd);
1377 	if (disk->d_kd.di.dumper == NULL)
1378 		G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.",
1379 		    rcp->provider->name, error);
1380 
1381 	g_raid_md_intel_new_disk(disk);
1382 
1383 	sx_xunlock(&sc->sc_lock);
1384 	g_topology_lock();
1385 	*gp = geom;
1386 	return (result);
1387 fail2:
1388 	g_topology_lock();
1389 	g_access(cp, -1, 0, 0);
1390 fail1:
1391 	free(meta, M_MD_INTEL);
1392 	return (G_RAID_MD_TASTE_FAIL);
1393 }
1394 
1395 static int
1396 g_raid_md_event_intel(struct g_raid_md_object *md,
1397     struct g_raid_disk *disk, u_int event)
1398 {
1399 	struct g_raid_softc *sc;
1400 	struct g_raid_subdisk *sd;
1401 	struct g_raid_md_intel_object *mdi;
1402 	struct g_raid_md_intel_perdisk *pd;
1403 
1404 	sc = md->mdo_softc;
1405 	mdi = (struct g_raid_md_intel_object *)md;
1406 	if (disk == NULL) {
1407 		switch (event) {
1408 		case G_RAID_NODE_E_START:
1409 			if (!mdi->mdio_started)
1410 				g_raid_md_intel_start(sc);
1411 			return (0);
1412 		}
1413 		return (-1);
1414 	}
1415 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
1416 	switch (event) {
1417 	case G_RAID_DISK_E_DISCONNECTED:
1418 		/* If disk was assigned, just update statuses. */
1419 		if (pd->pd_disk_pos >= 0) {
1420 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1421 			if (disk->d_consumer) {
1422 				g_raid_kill_consumer(sc, disk->d_consumer);
1423 				disk->d_consumer = NULL;
1424 			}
1425 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1426 				g_raid_change_subdisk_state(sd,
1427 				    G_RAID_SUBDISK_S_NONE);
1428 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1429 				    G_RAID_EVENT_SUBDISK);
1430 			}
1431 		} else {
1432 			/* Otherwise -- delete. */
1433 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1434 			g_raid_destroy_disk(disk);
1435 		}
1436 
1437 		/* Write updated metadata to all disks. */
1438 		g_raid_md_write_intel(md, NULL, NULL, NULL);
1439 
1440 		/* Check if anything left except placeholders. */
1441 		if (g_raid_ndisks(sc, -1) ==
1442 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1443 			g_raid_destroy_node(sc, 0);
1444 		else
1445 			g_raid_md_intel_refill(sc);
1446 		return (0);
1447 	}
1448 	return (-2);
1449 }
1450 
1451 static int
1452 g_raid_md_ctl_intel(struct g_raid_md_object *md,
1453     struct gctl_req *req)
1454 {
1455 	struct g_raid_softc *sc;
1456 	struct g_raid_volume *vol, *vol1;
1457 	struct g_raid_subdisk *sd;
1458 	struct g_raid_disk *disk;
1459 	struct g_raid_md_intel_object *mdi;
1460 	struct g_raid_md_intel_perdisk *pd;
1461 	struct g_consumer *cp;
1462 	struct g_provider *pp;
1463 	char arg[16], serial[INTEL_SERIAL_LEN];
1464 	const char *verb, *volname, *levelname, *diskname;
1465 	char *tmp;
1466 	int *nargs, *force;
1467 	off_t off, size, sectorsize, strip, disk_sectors;
1468 	intmax_t *sizearg, *striparg;
1469 	int numdisks, i, len, level, qual, update;
1470 	int error;
1471 
1472 	sc = md->mdo_softc;
1473 	mdi = (struct g_raid_md_intel_object *)md;
1474 	verb = gctl_get_param(req, "verb", NULL);
1475 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1476 	error = 0;
1477 	if (strcmp(verb, "label") == 0) {
1478 
1479 		if (*nargs < 4) {
1480 			gctl_error(req, "Invalid number of arguments.");
1481 			return (-1);
1482 		}
1483 		volname = gctl_get_asciiparam(req, "arg1");
1484 		if (volname == NULL) {
1485 			gctl_error(req, "No volume name.");
1486 			return (-2);
1487 		}
1488 		levelname = gctl_get_asciiparam(req, "arg2");
1489 		if (levelname == NULL) {
1490 			gctl_error(req, "No RAID level.");
1491 			return (-3);
1492 		}
1493 		if (strcasecmp(levelname, "RAID5") == 0)
1494 			levelname = "RAID5-LA";
1495 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1496 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1497 			return (-4);
1498 		}
1499 		numdisks = *nargs - 3;
1500 		force = gctl_get_paraml(req, "force", sizeof(*force));
1501 		if (!g_raid_md_intel_supported(level, qual, numdisks,
1502 		    force ? *force : 0)) {
1503 			gctl_error(req, "Unsupported RAID level "
1504 			    "(0x%02x/0x%02x), or number of disks (%d).",
1505 			    level, qual, numdisks);
1506 			return (-5);
1507 		}
1508 
1509 		/* Search for disks, connect them and probe. */
1510 		size = 0x7fffffffffffffffllu;
1511 		sectorsize = 0;
1512 		for (i = 0; i < numdisks; i++) {
1513 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1514 			diskname = gctl_get_asciiparam(req, arg);
1515 			if (diskname == NULL) {
1516 				gctl_error(req, "No disk name (%s).", arg);
1517 				error = -6;
1518 				break;
1519 			}
1520 			if (strcmp(diskname, "NONE") == 0) {
1521 				cp = NULL;
1522 				pp = NULL;
1523 			} else {
1524 				g_topology_lock();
1525 				cp = g_raid_open_consumer(sc, diskname);
1526 				if (cp == NULL) {
1527 					gctl_error(req, "Can't open disk '%s'.",
1528 					    diskname);
1529 					g_topology_unlock();
1530 					error = -7;
1531 					break;
1532 				}
1533 				pp = cp->provider;
1534 			}
1535 			pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
1536 			pd->pd_disk_pos = i;
1537 			disk = g_raid_create_disk(sc);
1538 			disk->d_md_data = (void *)pd;
1539 			disk->d_consumer = cp;
1540 			if (cp == NULL) {
1541 				strcpy(&pd->pd_disk_meta.serial[0], "NONE");
1542 				pd->pd_disk_meta.id = 0xffffffff;
1543 				pd->pd_disk_meta.flags = INTEL_F_ASSIGNED;
1544 				continue;
1545 			}
1546 			cp->private = disk;
1547 			g_topology_unlock();
1548 
1549 			error = g_raid_md_get_label(cp,
1550 			    &pd->pd_disk_meta.serial[0], INTEL_SERIAL_LEN);
1551 			if (error != 0) {
1552 				gctl_error(req,
1553 				    "Can't get serial for provider '%s'.",
1554 				    diskname);
1555 				error = -8;
1556 				break;
1557 			}
1558 
1559 			/* Read kernel dumping information. */
1560 			disk->d_kd.offset = 0;
1561 			disk->d_kd.length = OFF_MAX;
1562 			len = sizeof(disk->d_kd);
1563 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1564 			if (disk->d_kd.di.dumper == NULL)
1565 				G_RAID_DEBUG1(2, sc,
1566 				    "Dumping not supported by %s.",
1567 				    cp->provider->name);
1568 
1569 			intel_set_disk_sectors(&pd->pd_disk_meta,
1570 			    pp->mediasize / pp->sectorsize);
1571 			if (size > pp->mediasize)
1572 				size = pp->mediasize;
1573 			if (sectorsize < pp->sectorsize)
1574 				sectorsize = pp->sectorsize;
1575 			pd->pd_disk_meta.id = 0;
1576 			pd->pd_disk_meta.flags = INTEL_F_ASSIGNED | INTEL_F_ONLINE;
1577 		}
1578 		if (error != 0)
1579 			return (error);
1580 
1581 		if (sectorsize <= 0) {
1582 			gctl_error(req, "Can't get sector size.");
1583 			return (-8);
1584 		}
1585 
1586 		/* Reserve some space for metadata. */
1587 		size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize;
1588 
1589 		/* Handle size argument. */
1590 		len = sizeof(*sizearg);
1591 		sizearg = gctl_get_param(req, "size", &len);
1592 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1593 		    *sizearg > 0) {
1594 			if (*sizearg > size) {
1595 				gctl_error(req, "Size too big %lld > %lld.",
1596 				    (long long)*sizearg, (long long)size);
1597 				return (-9);
1598 			}
1599 			size = *sizearg;
1600 		}
1601 
1602 		/* Handle strip argument. */
1603 		strip = 131072;
1604 		len = sizeof(*striparg);
1605 		striparg = gctl_get_param(req, "strip", &len);
1606 		if (striparg != NULL && len == sizeof(*striparg) &&
1607 		    *striparg > 0) {
1608 			if (*striparg < sectorsize) {
1609 				gctl_error(req, "Strip size too small.");
1610 				return (-10);
1611 			}
1612 			if (*striparg % sectorsize != 0) {
1613 				gctl_error(req, "Incorrect strip size.");
1614 				return (-11);
1615 			}
1616 			if (strip > 65535 * sectorsize) {
1617 				gctl_error(req, "Strip size too big.");
1618 				return (-12);
1619 			}
1620 			strip = *striparg;
1621 		}
1622 
1623 		/* Round size down to strip or sector. */
1624 		if (level == G_RAID_VOLUME_RL_RAID1)
1625 			size -= (size % sectorsize);
1626 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1627 		    (numdisks & 1) != 0)
1628 			size -= (size % (2 * strip));
1629 		else
1630 			size -= (size % strip);
1631 		if (size <= 0) {
1632 			gctl_error(req, "Size too small.");
1633 			return (-13);
1634 		}
1635 
1636 		/* We have all we need, create things: volume, ... */
1637 		mdi->mdio_started = 1;
1638 		vol = g_raid_create_volume(sc, volname, -1);
1639 		vol->v_md_data = (void *)(intptr_t)0;
1640 		vol->v_raid_level = level;
1641 		vol->v_raid_level_qualifier = qual;
1642 		vol->v_strip_size = strip;
1643 		vol->v_disks_count = numdisks;
1644 		if (level == G_RAID_VOLUME_RL_RAID0)
1645 			vol->v_mediasize = size * numdisks;
1646 		else if (level == G_RAID_VOLUME_RL_RAID1)
1647 			vol->v_mediasize = size;
1648 		else if (level == G_RAID_VOLUME_RL_RAID5)
1649 			vol->v_mediasize = size * (numdisks - 1);
1650 		else { /* RAID1E */
1651 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1652 			    strip;
1653 		}
1654 		vol->v_sectorsize = sectorsize;
1655 		g_raid_start_volume(vol);
1656 
1657 		/* , and subdisks. */
1658 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1659 			pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
1660 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1661 			sd->sd_disk = disk;
1662 			sd->sd_offset = 0;
1663 			sd->sd_size = size;
1664 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1665 			if (sd->sd_disk->d_consumer != NULL) {
1666 				g_raid_change_disk_state(disk,
1667 				    G_RAID_DISK_S_ACTIVE);
1668 				if (level == G_RAID_VOLUME_RL_RAID5)
1669 					g_raid_change_subdisk_state(sd,
1670 					    G_RAID_SUBDISK_S_UNINITIALIZED);
1671 				else
1672 					g_raid_change_subdisk_state(sd,
1673 					    G_RAID_SUBDISK_S_ACTIVE);
1674 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1675 				    G_RAID_EVENT_SUBDISK);
1676 			} else {
1677 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1678 			}
1679 		}
1680 
1681 		/* Write metadata based on created entities. */
1682 		G_RAID_DEBUG1(0, sc, "Array started.");
1683 		g_raid_md_write_intel(md, NULL, NULL, NULL);
1684 
1685 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1686 		g_raid_md_intel_refill(sc);
1687 
1688 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1689 		    G_RAID_EVENT_VOLUME);
1690 		return (0);
1691 	}
1692 	if (strcmp(verb, "add") == 0) {
1693 
1694 		if (*nargs != 3) {
1695 			gctl_error(req, "Invalid number of arguments.");
1696 			return (-1);
1697 		}
1698 		volname = gctl_get_asciiparam(req, "arg1");
1699 		if (volname == NULL) {
1700 			gctl_error(req, "No volume name.");
1701 			return (-2);
1702 		}
1703 		levelname = gctl_get_asciiparam(req, "arg2");
1704 		if (levelname == NULL) {
1705 			gctl_error(req, "No RAID level.");
1706 			return (-3);
1707 		}
1708 		if (strcasecmp(levelname, "RAID5") == 0)
1709 			levelname = "RAID5-LA";
1710 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1711 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1712 			return (-4);
1713 		}
1714 
1715 		/* Look for existing volumes. */
1716 		i = 0;
1717 		vol1 = NULL;
1718 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1719 			vol1 = vol;
1720 			i++;
1721 		}
1722 		if (i > 1) {
1723 			gctl_error(req, "Maximum two volumes supported.");
1724 			return (-6);
1725 		}
1726 		if (vol1 == NULL) {
1727 			gctl_error(req, "At least one volume must exist.");
1728 			return (-7);
1729 		}
1730 
1731 		numdisks = vol1->v_disks_count;
1732 		force = gctl_get_paraml(req, "force", sizeof(*force));
1733 		if (!g_raid_md_intel_supported(level, qual, numdisks,
1734 		    force ? *force : 0)) {
1735 			gctl_error(req, "Unsupported RAID level "
1736 			    "(0x%02x/0x%02x), or number of disks (%d).",
1737 			    level, qual, numdisks);
1738 			return (-5);
1739 		}
1740 
1741 		/* Collect info about present disks. */
1742 		size = 0x7fffffffffffffffllu;
1743 		sectorsize = 512;
1744 		for (i = 0; i < numdisks; i++) {
1745 			disk = vol1->v_subdisks[i].sd_disk;
1746 			pd = (struct g_raid_md_intel_perdisk *)
1747 			    disk->d_md_data;
1748 			disk_sectors =
1749 			    intel_get_disk_sectors(&pd->pd_disk_meta);
1750 
1751 			if (disk_sectors * 512 < size)
1752 				size = disk_sectors * 512;
1753 			if (disk->d_consumer != NULL &&
1754 			    disk->d_consumer->provider != NULL &&
1755 			    disk->d_consumer->provider->sectorsize >
1756 			     sectorsize) {
1757 				sectorsize =
1758 				    disk->d_consumer->provider->sectorsize;
1759 			}
1760 		}
1761 
1762 		/* Reserve some space for metadata. */
1763 		size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize;
1764 
1765 		/* Decide insert before or after. */
1766 		sd = &vol1->v_subdisks[0];
1767 		if (sd->sd_offset >
1768 		    size - (sd->sd_offset + sd->sd_size)) {
1769 			off = 0;
1770 			size = sd->sd_offset;
1771 		} else {
1772 			off = sd->sd_offset + sd->sd_size;
1773 			size = size - (sd->sd_offset + sd->sd_size);
1774 		}
1775 
1776 		/* Handle strip argument. */
1777 		strip = 131072;
1778 		len = sizeof(*striparg);
1779 		striparg = gctl_get_param(req, "strip", &len);
1780 		if (striparg != NULL && len == sizeof(*striparg) &&
1781 		    *striparg > 0) {
1782 			if (*striparg < sectorsize) {
1783 				gctl_error(req, "Strip size too small.");
1784 				return (-10);
1785 			}
1786 			if (*striparg % sectorsize != 0) {
1787 				gctl_error(req, "Incorrect strip size.");
1788 				return (-11);
1789 			}
1790 			if (strip > 65535 * sectorsize) {
1791 				gctl_error(req, "Strip size too big.");
1792 				return (-12);
1793 			}
1794 			strip = *striparg;
1795 		}
1796 
1797 		/* Round offset up to strip. */
1798 		if (off % strip != 0) {
1799 			size -= strip - off % strip;
1800 			off += strip - off % strip;
1801 		}
1802 
1803 		/* Handle size argument. */
1804 		len = sizeof(*sizearg);
1805 		sizearg = gctl_get_param(req, "size", &len);
1806 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1807 		    *sizearg > 0) {
1808 			if (*sizearg > size) {
1809 				gctl_error(req, "Size too big %lld > %lld.",
1810 				    (long long)*sizearg, (long long)size);
1811 				return (-9);
1812 			}
1813 			size = *sizearg;
1814 		}
1815 
1816 		/* Round size down to strip or sector. */
1817 		if (level == G_RAID_VOLUME_RL_RAID1)
1818 			size -= (size % sectorsize);
1819 		else
1820 			size -= (size % strip);
1821 		if (size <= 0) {
1822 			gctl_error(req, "Size too small.");
1823 			return (-13);
1824 		}
1825 		if (size > 0xffffffffllu * sectorsize) {
1826 			gctl_error(req, "Size too big.");
1827 			return (-14);
1828 		}
1829 
1830 		/* We have all we need, create things: volume, ... */
1831 		vol = g_raid_create_volume(sc, volname, -1);
1832 		vol->v_md_data = (void *)(intptr_t)i;
1833 		vol->v_raid_level = level;
1834 		vol->v_raid_level_qualifier = qual;
1835 		vol->v_strip_size = strip;
1836 		vol->v_disks_count = numdisks;
1837 		if (level == G_RAID_VOLUME_RL_RAID0)
1838 			vol->v_mediasize = size * numdisks;
1839 		else if (level == G_RAID_VOLUME_RL_RAID1)
1840 			vol->v_mediasize = size;
1841 		else if (level == G_RAID_VOLUME_RL_RAID5)
1842 			vol->v_mediasize = size * (numdisks - 1);
1843 		else { /* RAID1E */
1844 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1845 			    strip;
1846 		}
1847 		vol->v_sectorsize = sectorsize;
1848 		g_raid_start_volume(vol);
1849 
1850 		/* , and subdisks. */
1851 		for (i = 0; i < numdisks; i++) {
1852 			disk = vol1->v_subdisks[i].sd_disk;
1853 			sd = &vol->v_subdisks[i];
1854 			sd->sd_disk = disk;
1855 			sd->sd_offset = off;
1856 			sd->sd_size = size;
1857 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1858 			if (disk->d_state == G_RAID_DISK_S_ACTIVE) {
1859 				if (level == G_RAID_VOLUME_RL_RAID5)
1860 					g_raid_change_subdisk_state(sd,
1861 					    G_RAID_SUBDISK_S_UNINITIALIZED);
1862 				else
1863 					g_raid_change_subdisk_state(sd,
1864 					    G_RAID_SUBDISK_S_ACTIVE);
1865 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1866 				    G_RAID_EVENT_SUBDISK);
1867 			}
1868 		}
1869 
1870 		/* Write metadata based on created entities. */
1871 		g_raid_md_write_intel(md, NULL, NULL, NULL);
1872 
1873 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1874 		    G_RAID_EVENT_VOLUME);
1875 		return (0);
1876 	}
1877 	if (strcmp(verb, "delete") == 0) {
1878 
1879 		/* Full node destruction. */
1880 		if (*nargs == 1) {
1881 			/* Check if some volume is still open. */
1882 			force = gctl_get_paraml(req, "force", sizeof(*force));
1883 			if (force != NULL && *force == 0 &&
1884 			    g_raid_nopens(sc) != 0) {
1885 				gctl_error(req, "Some volume is still open.");
1886 				return (-4);
1887 			}
1888 
1889 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1890 				if (disk->d_consumer)
1891 					intel_meta_erase(disk->d_consumer);
1892 			}
1893 			g_raid_destroy_node(sc, 0);
1894 			return (0);
1895 		}
1896 
1897 		/* Destroy specified volume. If it was last - all node. */
1898 		if (*nargs != 2) {
1899 			gctl_error(req, "Invalid number of arguments.");
1900 			return (-1);
1901 		}
1902 		volname = gctl_get_asciiparam(req, "arg1");
1903 		if (volname == NULL) {
1904 			gctl_error(req, "No volume name.");
1905 			return (-2);
1906 		}
1907 
1908 		/* Search for volume. */
1909 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1910 			if (strcmp(vol->v_name, volname) == 0)
1911 				break;
1912 		}
1913 		if (vol == NULL) {
1914 			i = strtol(volname, &tmp, 10);
1915 			if (verb != volname && tmp[0] == 0) {
1916 				TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1917 					if (vol->v_global_id == i)
1918 						break;
1919 				}
1920 			}
1921 		}
1922 		if (vol == NULL) {
1923 			gctl_error(req, "Volume '%s' not found.", volname);
1924 			return (-3);
1925 		}
1926 
1927 		/* Check if volume is still open. */
1928 		force = gctl_get_paraml(req, "force", sizeof(*force));
1929 		if (force != NULL && *force == 0 &&
1930 		    vol->v_provider_open != 0) {
1931 			gctl_error(req, "Volume is still open.");
1932 			return (-4);
1933 		}
1934 
1935 		/* Destroy volume and potentially node. */
1936 		i = 0;
1937 		TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
1938 			i++;
1939 		if (i >= 2) {
1940 			g_raid_destroy_volume(vol);
1941 			g_raid_md_write_intel(md, NULL, NULL, NULL);
1942 		} else {
1943 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1944 				if (disk->d_consumer)
1945 					intel_meta_erase(disk->d_consumer);
1946 			}
1947 			g_raid_destroy_node(sc, 0);
1948 		}
1949 		return (0);
1950 	}
1951 	if (strcmp(verb, "remove") == 0 ||
1952 	    strcmp(verb, "fail") == 0) {
1953 		if (*nargs < 2) {
1954 			gctl_error(req, "Invalid number of arguments.");
1955 			return (-1);
1956 		}
1957 		for (i = 1; i < *nargs; i++) {
1958 			snprintf(arg, sizeof(arg), "arg%d", i);
1959 			diskname = gctl_get_asciiparam(req, arg);
1960 			if (diskname == NULL) {
1961 				gctl_error(req, "No disk name (%s).", arg);
1962 				error = -2;
1963 				break;
1964 			}
1965 			if (strncmp(diskname, "/dev/", 5) == 0)
1966 				diskname += 5;
1967 
1968 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1969 				if (disk->d_consumer != NULL &&
1970 				    disk->d_consumer->provider != NULL &&
1971 				    strcmp(disk->d_consumer->provider->name,
1972 				     diskname) == 0)
1973 					break;
1974 			}
1975 			if (disk == NULL) {
1976 				gctl_error(req, "Disk '%s' not found.",
1977 				    diskname);
1978 				error = -3;
1979 				break;
1980 			}
1981 
1982 			if (strcmp(verb, "fail") == 0) {
1983 				g_raid_md_fail_disk_intel(md, NULL, disk);
1984 				continue;
1985 			}
1986 
1987 			pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
1988 
1989 			/* Erase metadata on deleting disk. */
1990 			intel_meta_erase(disk->d_consumer);
1991 
1992 			/* If disk was assigned, just update statuses. */
1993 			if (pd->pd_disk_pos >= 0) {
1994 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1995 				g_raid_kill_consumer(sc, disk->d_consumer);
1996 				disk->d_consumer = NULL;
1997 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1998 					g_raid_change_subdisk_state(sd,
1999 					    G_RAID_SUBDISK_S_NONE);
2000 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2001 					    G_RAID_EVENT_SUBDISK);
2002 				}
2003 			} else {
2004 				/* Otherwise -- delete. */
2005 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
2006 				g_raid_destroy_disk(disk);
2007 			}
2008 		}
2009 
2010 		/* Write updated metadata to remaining disks. */
2011 		g_raid_md_write_intel(md, NULL, NULL, NULL);
2012 
2013 		/* Check if anything left except placeholders. */
2014 		if (g_raid_ndisks(sc, -1) ==
2015 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
2016 			g_raid_destroy_node(sc, 0);
2017 		else
2018 			g_raid_md_intel_refill(sc);
2019 		return (error);
2020 	}
2021 	if (strcmp(verb, "insert") == 0) {
2022 		if (*nargs < 2) {
2023 			gctl_error(req, "Invalid number of arguments.");
2024 			return (-1);
2025 		}
2026 		update = 0;
2027 		for (i = 1; i < *nargs; i++) {
2028 			/* Get disk name. */
2029 			snprintf(arg, sizeof(arg), "arg%d", i);
2030 			diskname = gctl_get_asciiparam(req, arg);
2031 			if (diskname == NULL) {
2032 				gctl_error(req, "No disk name (%s).", arg);
2033 				error = -3;
2034 				break;
2035 			}
2036 
2037 			/* Try to find provider with specified name. */
2038 			g_topology_lock();
2039 			cp = g_raid_open_consumer(sc, diskname);
2040 			if (cp == NULL) {
2041 				gctl_error(req, "Can't open disk '%s'.",
2042 				    diskname);
2043 				g_topology_unlock();
2044 				error = -4;
2045 				break;
2046 			}
2047 			pp = cp->provider;
2048 			g_topology_unlock();
2049 
2050 			/* Read disk serial. */
2051 			error = g_raid_md_get_label(cp,
2052 			    &serial[0], INTEL_SERIAL_LEN);
2053 			if (error != 0) {
2054 				gctl_error(req,
2055 				    "Can't get serial for provider '%s'.",
2056 				    diskname);
2057 				g_raid_kill_consumer(sc, cp);
2058 				error = -7;
2059 				break;
2060 			}
2061 
2062 			pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
2063 			pd->pd_disk_pos = -1;
2064 
2065 			disk = g_raid_create_disk(sc);
2066 			disk->d_consumer = cp;
2067 			disk->d_md_data = (void *)pd;
2068 			cp->private = disk;
2069 
2070 			/* Read kernel dumping information. */
2071 			disk->d_kd.offset = 0;
2072 			disk->d_kd.length = OFF_MAX;
2073 			len = sizeof(disk->d_kd);
2074 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
2075 			if (disk->d_kd.di.dumper == NULL)
2076 				G_RAID_DEBUG1(2, sc,
2077 				    "Dumping not supported by %s.",
2078 				    cp->provider->name);
2079 
2080 			memcpy(&pd->pd_disk_meta.serial[0], &serial[0],
2081 			    INTEL_SERIAL_LEN);
2082 			intel_set_disk_sectors(&pd->pd_disk_meta,
2083 			    pp->mediasize / pp->sectorsize);
2084 			pd->pd_disk_meta.id = 0;
2085 			pd->pd_disk_meta.flags = INTEL_F_SPARE;
2086 
2087 			/* Welcome the "new" disk. */
2088 			update += g_raid_md_intel_start_disk(disk);
2089 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
2090 				intel_meta_write_spare(cp, &pd->pd_disk_meta);
2091 				g_raid_destroy_disk(disk);
2092 			} else if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2093 				gctl_error(req, "Disk '%s' doesn't fit.",
2094 				    diskname);
2095 				g_raid_destroy_disk(disk);
2096 				error = -8;
2097 				break;
2098 			}
2099 		}
2100 
2101 		/* Write new metadata if we changed something. */
2102 		if (update)
2103 			g_raid_md_write_intel(md, NULL, NULL, NULL);
2104 		return (error);
2105 	}
2106 	return (-100);
2107 }
2108 
2109 static int
2110 g_raid_md_write_intel(struct g_raid_md_object *md, struct g_raid_volume *tvol,
2111     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
2112 {
2113 	struct g_raid_softc *sc;
2114 	struct g_raid_volume *vol;
2115 	struct g_raid_subdisk *sd;
2116 	struct g_raid_disk *disk;
2117 	struct g_raid_md_intel_object *mdi;
2118 	struct g_raid_md_intel_perdisk *pd;
2119 	struct intel_raid_conf *meta;
2120 	struct intel_raid_vol *mvol;
2121 	struct intel_raid_map *mmap0, *mmap1;
2122 	off_t sectorsize = 512, pos;
2123 	const char *version, *cv;
2124 	int vi, sdi, numdisks, len, state, stale;
2125 
2126 	sc = md->mdo_softc;
2127 	mdi = (struct g_raid_md_intel_object *)md;
2128 
2129 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2130 		return (0);
2131 
2132 	/* Bump generation. Newly written metadata may differ from previous. */
2133 	mdi->mdio_generation++;
2134 
2135 	/* Count number of disks. */
2136 	numdisks = 0;
2137 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2138 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2139 		if (pd->pd_disk_pos < 0)
2140 			continue;
2141 		numdisks++;
2142 		if (disk->d_state == G_RAID_DISK_S_ACTIVE) {
2143 			pd->pd_disk_meta.flags =
2144 			    INTEL_F_ONLINE | INTEL_F_ASSIGNED;
2145 		} else if (disk->d_state == G_RAID_DISK_S_FAILED) {
2146 			pd->pd_disk_meta.flags = INTEL_F_FAILED | INTEL_F_ASSIGNED;
2147 		} else {
2148 			pd->pd_disk_meta.flags = INTEL_F_ASSIGNED;
2149 			if (pd->pd_disk_meta.id != 0xffffffff) {
2150 				pd->pd_disk_meta.id = 0xffffffff;
2151 				len = strlen(pd->pd_disk_meta.serial);
2152 				len = min(len, INTEL_SERIAL_LEN - 3);
2153 				strcpy(pd->pd_disk_meta.serial + len, ":0");
2154 			}
2155 		}
2156 	}
2157 
2158 	/* Fill anchor and disks. */
2159 	meta = malloc(INTEL_MAX_MD_SIZE(numdisks),
2160 	    M_MD_INTEL, M_WAITOK | M_ZERO);
2161 	memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1);
2162 	meta->config_size = INTEL_MAX_MD_SIZE(numdisks);
2163 	meta->config_id = mdi->mdio_config_id;
2164 	meta->generation = mdi->mdio_generation;
2165 	meta->attributes = INTEL_ATTR_CHECKSUM;
2166 	meta->total_disks = numdisks;
2167 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2168 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2169 		if (pd->pd_disk_pos < 0)
2170 			continue;
2171 		meta->disk[pd->pd_disk_pos] = pd->pd_disk_meta;
2172 	}
2173 
2174 	/* Fill volumes and maps. */
2175 	vi = 0;
2176 	version = INTEL_VERSION_1000;
2177 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2178 		if (vol->v_stopping)
2179 			continue;
2180 		mvol = intel_get_volume(meta, vi);
2181 
2182 		/* New metadata may have different volumes order. */
2183 		vol->v_md_data = (void *)(intptr_t)vi;
2184 
2185 		for (sdi = 0; sdi < vol->v_disks_count; sdi++) {
2186 			sd = &vol->v_subdisks[sdi];
2187 			if (sd->sd_disk != NULL)
2188 				break;
2189 		}
2190 		if (sdi >= vol->v_disks_count)
2191 			panic("No any filled subdisk in volume");
2192 		if (vol->v_mediasize >= 0x20000000000llu)
2193 			meta->attributes |= INTEL_ATTR_2TB;
2194 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
2195 			meta->attributes |= INTEL_ATTR_RAID0;
2196 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
2197 			meta->attributes |= INTEL_ATTR_RAID1;
2198 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
2199 			meta->attributes |= INTEL_ATTR_RAID5;
2200 		else
2201 			meta->attributes |= INTEL_ATTR_RAID10;
2202 
2203 		if (meta->attributes & INTEL_ATTR_2TB)
2204 			cv = INTEL_VERSION_1300;
2205 //		else if (dev->status == DEV_CLONE_N_GO)
2206 //			cv = INTEL_VERSION_1206;
2207 		else if (vol->v_disks_count > 4)
2208 			cv = INTEL_VERSION_1204;
2209 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
2210 			cv = INTEL_VERSION_1202;
2211 		else if (vol->v_disks_count > 2)
2212 			cv = INTEL_VERSION_1201;
2213 		else if (vi > 0)
2214 			cv = INTEL_VERSION_1200;
2215 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
2216 			cv = INTEL_VERSION_1100;
2217 		else
2218 			cv = INTEL_VERSION_1000;
2219 		if (strcmp(cv, version) > 0)
2220 			version = cv;
2221 
2222 		strlcpy(&mvol->name[0], vol->v_name, sizeof(mvol->name));
2223 		mvol->total_sectors = vol->v_mediasize / sectorsize;
2224 
2225 		/* Check for any recovery in progress. */
2226 		state = G_RAID_SUBDISK_S_ACTIVE;
2227 		pos = 0x7fffffffffffffffllu;
2228 		stale = 0;
2229 		for (sdi = 0; sdi < vol->v_disks_count; sdi++) {
2230 			sd = &vol->v_subdisks[sdi];
2231 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD)
2232 				state = G_RAID_SUBDISK_S_REBUILD;
2233 			else if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC &&
2234 			    state != G_RAID_SUBDISK_S_REBUILD)
2235 				state = G_RAID_SUBDISK_S_RESYNC;
2236 			else if (sd->sd_state == G_RAID_SUBDISK_S_STALE)
2237 				stale = 1;
2238 			if ((sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2239 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) &&
2240 			     sd->sd_rebuild_pos < pos)
2241 			        pos = sd->sd_rebuild_pos;
2242 		}
2243 		if (state == G_RAID_SUBDISK_S_REBUILD) {
2244 			mvol->migr_state = 1;
2245 			mvol->migr_type = INTEL_MT_REBUILD;
2246 		} else if (state == G_RAID_SUBDISK_S_RESYNC) {
2247 			mvol->migr_state = 1;
2248 			/* mvol->migr_type = INTEL_MT_REPAIR; */
2249 			mvol->migr_type = INTEL_MT_VERIFY;
2250 			mvol->state |= INTEL_ST_VERIFY_AND_FIX;
2251 		} else
2252 			mvol->migr_state = 0;
2253 		mvol->dirty = (vol->v_dirty || stale);
2254 
2255 		mmap0 = intel_get_map(mvol, 0);
2256 
2257 		/* Write map / common part of two maps. */
2258 		intel_set_map_offset(mmap0, sd->sd_offset / sectorsize);
2259 		intel_set_map_disk_sectors(mmap0, sd->sd_size / sectorsize);
2260 		mmap0->strip_sectors = vol->v_strip_size / sectorsize;
2261 		if (vol->v_state == G_RAID_VOLUME_S_BROKEN)
2262 			mmap0->status = INTEL_S_FAILURE;
2263 		else if (vol->v_state == G_RAID_VOLUME_S_DEGRADED)
2264 			mmap0->status = INTEL_S_DEGRADED;
2265 		else if (g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED)
2266 		    == g_raid_nsubdisks(vol, -1))
2267 			mmap0->status = INTEL_S_UNINITIALIZED;
2268 		else
2269 			mmap0->status = INTEL_S_READY;
2270 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
2271 			mmap0->type = INTEL_T_RAID0;
2272 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
2273 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
2274 			mmap0->type = INTEL_T_RAID1;
2275 		else
2276 			mmap0->type = INTEL_T_RAID5;
2277 		mmap0->total_disks = vol->v_disks_count;
2278 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
2279 			mmap0->total_domains = vol->v_disks_count;
2280 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
2281 			mmap0->total_domains = 2;
2282 		else
2283 			mmap0->total_domains = 1;
2284 		intel_set_map_stripe_count(mmap0,
2285 		    sd->sd_size / vol->v_strip_size / mmap0->total_domains);
2286 		mmap0->failed_disk_num = 0xff;
2287 		mmap0->ddf = 1;
2288 
2289 		/* If there are two maps - copy common and update. */
2290 		if (mvol->migr_state) {
2291 			intel_set_vol_curr_migr_unit(mvol,
2292 			    pos / vol->v_strip_size / mmap0->total_domains);
2293 			mmap1 = intel_get_map(mvol, 1);
2294 			memcpy(mmap1, mmap0, sizeof(struct intel_raid_map));
2295 			mmap0->status = INTEL_S_READY;
2296 		} else
2297 			mmap1 = NULL;
2298 
2299 		/* Write disk indexes and put rebuild flags. */
2300 		for (sdi = 0; sdi < vol->v_disks_count; sdi++) {
2301 			sd = &vol->v_subdisks[sdi];
2302 			pd = (struct g_raid_md_intel_perdisk *)
2303 			    sd->sd_disk->d_md_data;
2304 			mmap0->disk_idx[sdi] = pd->pd_disk_pos;
2305 			if (mvol->migr_state)
2306 				mmap1->disk_idx[sdi] = pd->pd_disk_pos;
2307 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2308 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2309 				mmap1->disk_idx[sdi] |= INTEL_DI_RBLD;
2310 			} else if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE &&
2311 			    sd->sd_state != G_RAID_SUBDISK_S_STALE &&
2312 			    sd->sd_state != G_RAID_SUBDISK_S_UNINITIALIZED) {
2313 				mmap0->disk_idx[sdi] |= INTEL_DI_RBLD;
2314 				if (mvol->migr_state)
2315 					mmap1->disk_idx[sdi] |= INTEL_DI_RBLD;
2316 			}
2317 			if ((sd->sd_state == G_RAID_SUBDISK_S_NONE ||
2318 			     sd->sd_state == G_RAID_SUBDISK_S_FAILED) &&
2319 			    mmap0->failed_disk_num == 0xff) {
2320 				mmap0->failed_disk_num = sdi;
2321 				if (mvol->migr_state)
2322 					mmap1->failed_disk_num = sdi;
2323 			}
2324 		}
2325 		vi++;
2326 	}
2327 	meta->total_volumes = vi;
2328 	if (strcmp(version, INTEL_VERSION_1300) != 0)
2329 		meta->attributes &= INTEL_ATTR_CHECKSUM;
2330 	memcpy(&meta->version[0], version, sizeof(INTEL_VERSION_1000) - 1);
2331 
2332 	/* We are done. Print meta data and store them to disks. */
2333 	g_raid_md_intel_print(meta);
2334 	if (mdi->mdio_meta != NULL)
2335 		free(mdi->mdio_meta, M_MD_INTEL);
2336 	mdi->mdio_meta = meta;
2337 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2338 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2339 		if (disk->d_state != G_RAID_DISK_S_ACTIVE)
2340 			continue;
2341 		if (pd->pd_meta != NULL) {
2342 			free(pd->pd_meta, M_MD_INTEL);
2343 			pd->pd_meta = NULL;
2344 		}
2345 		pd->pd_meta = intel_meta_copy(meta);
2346 		intel_meta_write(disk->d_consumer, meta);
2347 	}
2348 	return (0);
2349 }
2350 
2351 static int
2352 g_raid_md_fail_disk_intel(struct g_raid_md_object *md,
2353     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
2354 {
2355 	struct g_raid_softc *sc;
2356 	struct g_raid_md_intel_object *mdi;
2357 	struct g_raid_md_intel_perdisk *pd;
2358 	struct g_raid_subdisk *sd;
2359 
2360 	sc = md->mdo_softc;
2361 	mdi = (struct g_raid_md_intel_object *)md;
2362 	pd = (struct g_raid_md_intel_perdisk *)tdisk->d_md_data;
2363 
2364 	/* We can't fail disk that is not a part of array now. */
2365 	if (pd->pd_disk_pos < 0)
2366 		return (-1);
2367 
2368 	/*
2369 	 * Mark disk as failed in metadata and try to write that metadata
2370 	 * to the disk itself to prevent it's later resurrection as STALE.
2371 	 */
2372 	mdi->mdio_meta->disk[pd->pd_disk_pos].flags = INTEL_F_FAILED;
2373 	pd->pd_disk_meta.flags = INTEL_F_FAILED;
2374 	g_raid_md_intel_print(mdi->mdio_meta);
2375 	if (tdisk->d_consumer)
2376 		intel_meta_write(tdisk->d_consumer, mdi->mdio_meta);
2377 
2378 	/* Change states. */
2379 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
2380 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
2381 		g_raid_change_subdisk_state(sd,
2382 		    G_RAID_SUBDISK_S_FAILED);
2383 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
2384 		    G_RAID_EVENT_SUBDISK);
2385 	}
2386 
2387 	/* Write updated metadata to remaining disks. */
2388 	g_raid_md_write_intel(md, NULL, NULL, tdisk);
2389 
2390 	/* Check if anything left except placeholders. */
2391 	if (g_raid_ndisks(sc, -1) ==
2392 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
2393 		g_raid_destroy_node(sc, 0);
2394 	else
2395 		g_raid_md_intel_refill(sc);
2396 	return (0);
2397 }
2398 
2399 static int
2400 g_raid_md_free_disk_intel(struct g_raid_md_object *md,
2401     struct g_raid_disk *disk)
2402 {
2403 	struct g_raid_md_intel_perdisk *pd;
2404 
2405 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2406 	if (pd->pd_meta != NULL) {
2407 		free(pd->pd_meta, M_MD_INTEL);
2408 		pd->pd_meta = NULL;
2409 	}
2410 	free(pd, M_MD_INTEL);
2411 	disk->d_md_data = NULL;
2412 	return (0);
2413 }
2414 
2415 static int
2416 g_raid_md_free_intel(struct g_raid_md_object *md)
2417 {
2418 	struct g_raid_md_intel_object *mdi;
2419 
2420 	mdi = (struct g_raid_md_intel_object *)md;
2421 	if (!mdi->mdio_started) {
2422 		mdi->mdio_started = 0;
2423 		callout_stop(&mdi->mdio_start_co);
2424 		G_RAID_DEBUG1(1, md->mdo_softc,
2425 		    "root_mount_rel %p", mdi->mdio_rootmount);
2426 		root_mount_rel(mdi->mdio_rootmount);
2427 		mdi->mdio_rootmount = NULL;
2428 	}
2429 	if (mdi->mdio_meta != NULL) {
2430 		free(mdi->mdio_meta, M_MD_INTEL);
2431 		mdi->mdio_meta = NULL;
2432 	}
2433 	return (0);
2434 }
2435 
2436 G_RAID_MD_DECLARE(intel, "Intel");
2437