xref: /freebsd/sys/geom/raid/md_intel.c (revision 8ef24a0d4b28fe230e20637f56869cc4148cd2ca)
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 #define INTEL_CNGST_UPDATED		0
102 #define INTEL_CNGST_NEEDS_UPDATE	1
103 #define INTEL_CNGST_MASTER_MISSING	2
104 	uint8_t		cng_sub_state;
105 	uint32_t	filler_0[10];
106 
107 	uint32_t	curr_migr_unit;
108 	uint32_t	checkpoint_id;
109 	uint8_t		migr_state;
110 	uint8_t		migr_type;
111 #define INTEL_MT_INIT		0
112 #define INTEL_MT_REBUILD	1
113 #define INTEL_MT_VERIFY		2
114 #define INTEL_MT_GEN_MIGR	3
115 #define INTEL_MT_STATE_CHANGE	4
116 #define INTEL_MT_REPAIR		5
117 	uint8_t		dirty;
118 	uint8_t		fs_state;
119 	uint16_t	verify_errors;
120 	uint16_t	bad_blocks;
121 	uint32_t	curr_migr_unit_hi;
122 	uint32_t	filler_1[3];
123 	struct intel_raid_map map[1];	/* 2 entries if migr_state != 0. */
124 } __packed;
125 
126 struct intel_raid_disk {
127 #define INTEL_SERIAL_LEN	16
128 	uint8_t		serial[INTEL_SERIAL_LEN];
129 	uint32_t	sectors;
130 	uint32_t	id;
131 	uint32_t	flags;
132 #define INTEL_F_SPARE		0x01
133 #define INTEL_F_ASSIGNED	0x02
134 #define INTEL_F_FAILED		0x04
135 #define INTEL_F_ONLINE		0x08
136 #define INTEL_F_DISABLED	0x80
137 	uint32_t	owner_cfg_num;
138 	uint32_t	sectors_hi;
139 	uint32_t	filler[3];
140 } __packed;
141 
142 struct intel_raid_conf {
143 	uint8_t		intel_id[24];
144 #define INTEL_MAGIC             "Intel Raid ISM Cfg Sig. "
145 
146 	uint8_t		version[6];
147 #define INTEL_VERSION_1000	"1.0.00"	/* RAID0 */
148 #define INTEL_VERSION_1100	"1.1.00"	/* RAID1 */
149 #define INTEL_VERSION_1200	"1.2.00"	/* Many volumes */
150 #define INTEL_VERSION_1201	"1.2.01"	/* 3 or 4 disks */
151 #define INTEL_VERSION_1202	"1.2.02"	/* RAID5 */
152 #define INTEL_VERSION_1204	"1.2.04"	/* 5 or 6 disks */
153 #define INTEL_VERSION_1206	"1.2.06"	/* CNG */
154 #define INTEL_VERSION_1300	"1.3.00"	/* Attributes */
155 
156 	uint8_t		dummy_0[2];
157 	uint32_t	checksum;
158 	uint32_t	config_size;
159 	uint32_t	config_id;
160 	uint32_t	generation;
161 	uint32_t	error_log_size;
162 	uint32_t	attributes;
163 #define INTEL_ATTR_RAID0	0x00000001
164 #define INTEL_ATTR_RAID1	0x00000002
165 #define INTEL_ATTR_RAID10	0x00000004
166 #define INTEL_ATTR_RAID1E	0x00000008
167 #define INTEL_ATTR_RAID5	0x00000010
168 #define INTEL_ATTR_RAIDCNG	0x00000020
169 #define INTEL_ATTR_EXT_STRIP	0x00000040
170 #define INTEL_ATTR_NVM_CACHE	0x02000000
171 #define INTEL_ATTR_2TB_DISK	0x04000000
172 #define INTEL_ATTR_BBM		0x08000000
173 #define INTEL_ATTR_NVM_CACHE2	0x10000000
174 #define INTEL_ATTR_2TB		0x20000000
175 #define INTEL_ATTR_PM		0x40000000
176 #define INTEL_ATTR_CHECKSUM	0x80000000
177 
178 	uint8_t		total_disks;
179 	uint8_t		total_volumes;
180 	uint8_t		error_log_pos;
181 	uint8_t		dummy_2[1];
182 	uint32_t	cache_size;
183 	uint32_t	orig_config_id;
184 	uint32_t	pwr_cycle_count;
185 	uint32_t	bbm_log_size;
186 	uint32_t	filler_0[35];
187 	struct intel_raid_disk	disk[1];	/* total_disks entries. */
188 	/* Here goes total_volumes of struct intel_raid_vol. */
189 } __packed;
190 
191 #define INTEL_ATTR_SUPPORTED	( INTEL_ATTR_RAID0 | INTEL_ATTR_RAID1 |	\
192     INTEL_ATTR_RAID10 | INTEL_ATTR_RAID1E | INTEL_ATTR_RAID5 |		\
193     INTEL_ATTR_RAIDCNG | INTEL_ATTR_EXT_STRIP | INTEL_ATTR_2TB_DISK |	\
194     INTEL_ATTR_2TB | INTEL_ATTR_PM | INTEL_ATTR_CHECKSUM )
195 
196 #define INTEL_MAX_MD_SIZE(ndisks)				\
197     (sizeof(struct intel_raid_conf) +				\
198      sizeof(struct intel_raid_disk) * (ndisks - 1) +		\
199      sizeof(struct intel_raid_vol) * 2 +			\
200      sizeof(struct intel_raid_map) * 2 +			\
201      sizeof(uint32_t) * (ndisks - 1) * 4)
202 
203 struct g_raid_md_intel_perdisk {
204 	struct intel_raid_conf	*pd_meta;
205 	int			 pd_disk_pos;
206 	struct intel_raid_disk	 pd_disk_meta;
207 };
208 
209 struct g_raid_md_intel_pervolume {
210 	int			 pv_volume_pos;
211 	int			 pv_cng;
212 	int			 pv_cng_man_sync;
213 	int			 pv_cng_master_disk;
214 };
215 
216 struct g_raid_md_intel_object {
217 	struct g_raid_md_object	 mdio_base;
218 	uint32_t		 mdio_config_id;
219 	uint32_t		 mdio_orig_config_id;
220 	uint32_t		 mdio_generation;
221 	struct intel_raid_conf	*mdio_meta;
222 	struct callout		 mdio_start_co;	/* STARTING state timer. */
223 	int			 mdio_disks_present;
224 	int			 mdio_started;
225 	int			 mdio_incomplete;
226 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
227 };
228 
229 static g_raid_md_create_t g_raid_md_create_intel;
230 static g_raid_md_taste_t g_raid_md_taste_intel;
231 static g_raid_md_event_t g_raid_md_event_intel;
232 static g_raid_md_ctl_t g_raid_md_ctl_intel;
233 static g_raid_md_write_t g_raid_md_write_intel;
234 static g_raid_md_fail_disk_t g_raid_md_fail_disk_intel;
235 static g_raid_md_free_disk_t g_raid_md_free_disk_intel;
236 static g_raid_md_free_volume_t g_raid_md_free_volume_intel;
237 static g_raid_md_free_t g_raid_md_free_intel;
238 
239 static kobj_method_t g_raid_md_intel_methods[] = {
240 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_intel),
241 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_intel),
242 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_intel),
243 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_intel),
244 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_intel),
245 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_intel),
246 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_intel),
247 	KOBJMETHOD(g_raid_md_free_volume,	g_raid_md_free_volume_intel),
248 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_intel),
249 	{ 0, 0 }
250 };
251 
252 static struct g_raid_md_class g_raid_md_intel_class = {
253 	"Intel",
254 	g_raid_md_intel_methods,
255 	sizeof(struct g_raid_md_intel_object),
256 	.mdc_enable = 1,
257 	.mdc_priority = 100
258 };
259 
260 
261 static struct intel_raid_map *
262 intel_get_map(struct intel_raid_vol *mvol, int i)
263 {
264 	struct intel_raid_map *mmap;
265 
266 	if (i > (mvol->migr_state ? 1 : 0))
267 		return (NULL);
268 	mmap = &mvol->map[0];
269 	for (; i > 0; i--) {
270 		mmap = (struct intel_raid_map *)
271 		    &mmap->disk_idx[mmap->total_disks];
272 	}
273 	return ((struct intel_raid_map *)mmap);
274 }
275 
276 static struct intel_raid_vol *
277 intel_get_volume(struct intel_raid_conf *meta, int i)
278 {
279 	struct intel_raid_vol *mvol;
280 	struct intel_raid_map *mmap;
281 
282 	if (i > 1)
283 		return (NULL);
284 	mvol = (struct intel_raid_vol *)&meta->disk[meta->total_disks];
285 	for (; i > 0; i--) {
286 		mmap = intel_get_map(mvol, mvol->migr_state ? 1 : 0);
287 		mvol = (struct intel_raid_vol *)
288 		    &mmap->disk_idx[mmap->total_disks];
289 	}
290 	return (mvol);
291 }
292 
293 static off_t
294 intel_get_map_offset(struct intel_raid_map *mmap)
295 {
296 	off_t offset = (off_t)mmap->offset_hi << 32;
297 
298 	offset += mmap->offset;
299 	return (offset);
300 }
301 
302 static void
303 intel_set_map_offset(struct intel_raid_map *mmap, off_t offset)
304 {
305 
306 	mmap->offset = offset & 0xffffffff;
307 	mmap->offset_hi = offset >> 32;
308 }
309 
310 static off_t
311 intel_get_map_disk_sectors(struct intel_raid_map *mmap)
312 {
313 	off_t disk_sectors = (off_t)mmap->disk_sectors_hi << 32;
314 
315 	disk_sectors += mmap->disk_sectors;
316 	return (disk_sectors);
317 }
318 
319 static void
320 intel_set_map_disk_sectors(struct intel_raid_map *mmap, off_t disk_sectors)
321 {
322 
323 	mmap->disk_sectors = disk_sectors & 0xffffffff;
324 	mmap->disk_sectors_hi = disk_sectors >> 32;
325 }
326 
327 static void
328 intel_set_map_stripe_count(struct intel_raid_map *mmap, off_t stripe_count)
329 {
330 
331 	mmap->stripe_count = stripe_count & 0xffffffff;
332 	mmap->stripe_count_hi = stripe_count >> 32;
333 }
334 
335 static off_t
336 intel_get_disk_sectors(struct intel_raid_disk *disk)
337 {
338 	off_t sectors = (off_t)disk->sectors_hi << 32;
339 
340 	sectors += disk->sectors;
341 	return (sectors);
342 }
343 
344 static void
345 intel_set_disk_sectors(struct intel_raid_disk *disk, off_t sectors)
346 {
347 
348 	disk->sectors = sectors & 0xffffffff;
349 	disk->sectors_hi = sectors >> 32;
350 }
351 
352 static off_t
353 intel_get_vol_curr_migr_unit(struct intel_raid_vol *vol)
354 {
355 	off_t curr_migr_unit = (off_t)vol->curr_migr_unit_hi << 32;
356 
357 	curr_migr_unit += vol->curr_migr_unit;
358 	return (curr_migr_unit);
359 }
360 
361 static void
362 intel_set_vol_curr_migr_unit(struct intel_raid_vol *vol, off_t curr_migr_unit)
363 {
364 
365 	vol->curr_migr_unit = curr_migr_unit & 0xffffffff;
366 	vol->curr_migr_unit_hi = curr_migr_unit >> 32;
367 }
368 
369 static char *
370 intel_status2str(int status)
371 {
372 
373 	switch (status) {
374 	case INTEL_S_READY:
375 		return ("READY");
376 	case INTEL_S_UNINITIALIZED:
377 		return ("UNINITIALIZED");
378 	case INTEL_S_DEGRADED:
379 		return ("DEGRADED");
380 	case INTEL_S_FAILURE:
381 		return ("FAILURE");
382 	default:
383 		return ("UNKNOWN");
384 	}
385 }
386 
387 static char *
388 intel_type2str(int type)
389 {
390 
391 	switch (type) {
392 	case INTEL_T_RAID0:
393 		return ("RAID0");
394 	case INTEL_T_RAID1:
395 		return ("RAID1");
396 	case INTEL_T_RAID5:
397 		return ("RAID5");
398 	default:
399 		return ("UNKNOWN");
400 	}
401 }
402 
403 static char *
404 intel_cngst2str(int cng_state)
405 {
406 
407 	switch (cng_state) {
408 	case INTEL_CNGST_UPDATED:
409 		return ("UPDATED");
410 	case INTEL_CNGST_NEEDS_UPDATE:
411 		return ("NEEDS_UPDATE");
412 	case INTEL_CNGST_MASTER_MISSING:
413 		return ("MASTER_MISSING");
414 	default:
415 		return ("UNKNOWN");
416 	}
417 }
418 
419 static char *
420 intel_mt2str(int type)
421 {
422 
423 	switch (type) {
424 	case INTEL_MT_INIT:
425 		return ("INIT");
426 	case INTEL_MT_REBUILD:
427 		return ("REBUILD");
428 	case INTEL_MT_VERIFY:
429 		return ("VERIFY");
430 	case INTEL_MT_GEN_MIGR:
431 		return ("GEN_MIGR");
432 	case INTEL_MT_STATE_CHANGE:
433 		return ("STATE_CHANGE");
434 	case INTEL_MT_REPAIR:
435 		return ("REPAIR");
436 	default:
437 		return ("UNKNOWN");
438 	}
439 }
440 
441 static void
442 g_raid_md_intel_print(struct intel_raid_conf *meta)
443 {
444 	struct intel_raid_vol *mvol;
445 	struct intel_raid_map *mmap;
446 	int i, j, k;
447 
448 	if (g_raid_debug < 1)
449 		return;
450 
451 	printf("********* ATA Intel MatrixRAID Metadata *********\n");
452 	printf("intel_id            <%.24s>\n", meta->intel_id);
453 	printf("version             <%.6s>\n", meta->version);
454 	printf("checksum            0x%08x\n", meta->checksum);
455 	printf("config_size         0x%08x\n", meta->config_size);
456 	printf("config_id           0x%08x\n", meta->config_id);
457 	printf("generation          0x%08x\n", meta->generation);
458 	printf("error_log_size      %d\n", meta->error_log_size);
459 	printf("attributes          0x%b\n", meta->attributes,
460 		"\020"
461 		"\001RAID0"
462 		"\002RAID1"
463 		"\003RAID10"
464 		"\004RAID1E"
465 		"\005RAID15"
466 		"\006RAIDCNG"
467 		"\007EXT_STRIP"
468 		"\032NVM_CACHE"
469 		"\0332TB_DISK"
470 		"\034BBM"
471 		"\035NVM_CACHE"
472 		"\0362TB"
473 		"\037PM"
474 		"\040CHECKSUM");
475 	printf("total_disks         %u\n", meta->total_disks);
476 	printf("total_volumes       %u\n", meta->total_volumes);
477 	printf("error_log_pos       %u\n", meta->error_log_pos);
478 	printf("cache_size          %u\n", meta->cache_size);
479 	printf("orig_config_id      0x%08x\n", meta->orig_config_id);
480 	printf("pwr_cycle_count     %u\n", meta->pwr_cycle_count);
481 	printf("bbm_log_size        %u\n", meta->bbm_log_size);
482 	printf("Flags: S - Spare, A - Assigned, F - Failed, O - Online, D - Disabled\n");
483 	printf("DISK#   serial disk_sectors disk_sectors_hi disk_id flags owner\n");
484 	for (i = 0; i < meta->total_disks; i++ ) {
485 		printf("    %d   <%.16s> %u %u 0x%08x 0x%b %08x\n", i,
486 		    meta->disk[i].serial, meta->disk[i].sectors,
487 		    meta->disk[i].sectors_hi, meta->disk[i].id,
488 		    meta->disk[i].flags, "\20\01S\02A\03F\04O\05D",
489 		    meta->disk[i].owner_cfg_num);
490 	}
491 	for (i = 0; i < meta->total_volumes; i++) {
492 		mvol = intel_get_volume(meta, i);
493 		printf(" ****** Volume %d ******\n", i);
494 		printf(" name               %.16s\n", mvol->name);
495 		printf(" total_sectors      %ju\n", mvol->total_sectors);
496 		printf(" state              0x%b\n", mvol->state,
497 			"\020"
498 			"\001BOOTABLE"
499 			"\002BOOT_DEVICE"
500 			"\003READ_COALESCING"
501 			"\004WRITE_COALESCING"
502 			"\005LAST_SHUTDOWN_DIRTY"
503 			"\006HIDDEN_AT_BOOT"
504 			"\007CURRENTLY_HIDDEN"
505 			"\010VERIFY_AND_FIX"
506 			"\011MAP_STATE_UNINIT"
507 			"\012NO_AUTO_RECOVERY"
508 			"\013CLONE_N_GO"
509 			"\014CLONE_MAN_SYNC"
510 			"\015CNG_MASTER_DISK_NUM");
511 		printf(" reserved           %u\n", mvol->reserved);
512 		printf(" migr_priority      %u\n", mvol->migr_priority);
513 		printf(" num_sub_vols       %u\n", mvol->num_sub_vols);
514 		printf(" tid                %u\n", mvol->tid);
515 		printf(" cng_master_disk    %u\n", mvol->cng_master_disk);
516 		printf(" cache_policy       %u\n", mvol->cache_policy);
517 		printf(" cng_state          %u (%s)\n", mvol->cng_state,
518 			intel_cngst2str(mvol->cng_state));
519 		printf(" cng_sub_state      %u\n", mvol->cng_sub_state);
520 		printf(" curr_migr_unit     %u\n", mvol->curr_migr_unit);
521 		printf(" curr_migr_unit_hi  %u\n", mvol->curr_migr_unit_hi);
522 		printf(" checkpoint_id      %u\n", mvol->checkpoint_id);
523 		printf(" migr_state         %u\n", mvol->migr_state);
524 		printf(" migr_type          %u (%s)\n", mvol->migr_type,
525 			intel_mt2str(mvol->migr_type));
526 		printf(" dirty              %u\n", mvol->dirty);
527 		printf(" fs_state           %u\n", mvol->fs_state);
528 		printf(" verify_errors      %u\n", mvol->verify_errors);
529 		printf(" bad_blocks         %u\n", mvol->bad_blocks);
530 
531 		for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) {
532 			printf("  *** Map %d ***\n", j);
533 			mmap = intel_get_map(mvol, j);
534 			printf("  offset            %u\n", mmap->offset);
535 			printf("  offset_hi         %u\n", mmap->offset_hi);
536 			printf("  disk_sectors      %u\n", mmap->disk_sectors);
537 			printf("  disk_sectors_hi   %u\n", mmap->disk_sectors_hi);
538 			printf("  stripe_count      %u\n", mmap->stripe_count);
539 			printf("  stripe_count_hi   %u\n", mmap->stripe_count_hi);
540 			printf("  strip_sectors     %u\n", mmap->strip_sectors);
541 			printf("  status            %u (%s)\n", mmap->status,
542 				intel_status2str(mmap->status));
543 			printf("  type              %u (%s)\n", mmap->type,
544 				intel_type2str(mmap->type));
545 			printf("  total_disks       %u\n", mmap->total_disks);
546 			printf("  total_domains     %u\n", mmap->total_domains);
547 			printf("  failed_disk_num   %u\n", mmap->failed_disk_num);
548 			printf("  ddf               %u\n", mmap->ddf);
549 			printf("  disk_idx         ");
550 			for (k = 0; k < mmap->total_disks; k++)
551 				printf(" 0x%08x", mmap->disk_idx[k]);
552 			printf("\n");
553 		}
554 	}
555 	printf("=================================================\n");
556 }
557 
558 static struct intel_raid_conf *
559 intel_meta_copy(struct intel_raid_conf *meta)
560 {
561 	struct intel_raid_conf *nmeta;
562 
563 	nmeta = malloc(meta->config_size, M_MD_INTEL, M_WAITOK);
564 	memcpy(nmeta, meta, meta->config_size);
565 	return (nmeta);
566 }
567 
568 static int
569 intel_meta_find_disk(struct intel_raid_conf *meta, char *serial)
570 {
571 	int pos;
572 
573 	for (pos = 0; pos < meta->total_disks; pos++) {
574 		if (strncmp(meta->disk[pos].serial,
575 		    serial, INTEL_SERIAL_LEN) == 0)
576 			return (pos);
577 	}
578 	return (-1);
579 }
580 
581 static struct intel_raid_conf *
582 intel_meta_read(struct g_consumer *cp)
583 {
584 	struct g_provider *pp;
585 	struct intel_raid_conf *meta;
586 	struct intel_raid_vol *mvol;
587 	struct intel_raid_map *mmap, *mmap1;
588 	char *buf;
589 	int error, i, j, k, left, size;
590 	uint32_t checksum, *ptr;
591 
592 	pp = cp->provider;
593 
594 	/* Read the anchor sector. */
595 	buf = g_read_data(cp,
596 	    pp->mediasize - pp->sectorsize * 2, pp->sectorsize, &error);
597 	if (buf == NULL) {
598 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
599 		    pp->name, error);
600 		return (NULL);
601 	}
602 	meta = (struct intel_raid_conf *)buf;
603 
604 	/* Check if this is an Intel RAID struct */
605 	if (strncmp(meta->intel_id, INTEL_MAGIC, strlen(INTEL_MAGIC))) {
606 		G_RAID_DEBUG(1, "Intel signature check failed on %s", pp->name);
607 		g_free(buf);
608 		return (NULL);
609 	}
610 	if (meta->config_size > 65536 ||
611 	    meta->config_size < sizeof(struct intel_raid_conf)) {
612 		G_RAID_DEBUG(1, "Intel metadata size looks wrong: %d",
613 		    meta->config_size);
614 		g_free(buf);
615 		return (NULL);
616 	}
617 	size = meta->config_size;
618 	meta = malloc(size, M_MD_INTEL, M_WAITOK);
619 	memcpy(meta, buf, min(size, pp->sectorsize));
620 	g_free(buf);
621 
622 	/* Read all the rest, if needed. */
623 	if (meta->config_size > pp->sectorsize) {
624 		left = (meta->config_size - 1) / pp->sectorsize;
625 		buf = g_read_data(cp,
626 		    pp->mediasize - pp->sectorsize * (2 + left),
627 		    pp->sectorsize * left, &error);
628 		if (buf == NULL) {
629 			G_RAID_DEBUG(1, "Cannot read remaining metadata"
630 			    " part from %s (error=%d).",
631 			    pp->name, error);
632 			free(meta, M_MD_INTEL);
633 			return (NULL);
634 		}
635 		memcpy(((char *)meta) + pp->sectorsize, buf,
636 		    pp->sectorsize * left);
637 		g_free(buf);
638 	}
639 
640 	/* Check metadata checksum. */
641 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0;
642 	    i < (meta->config_size / sizeof(uint32_t)); i++) {
643 		checksum += *ptr++;
644 	}
645 	checksum -= meta->checksum;
646 	if (checksum != meta->checksum) {
647 		G_RAID_DEBUG(1, "Intel checksum check failed on %s", pp->name);
648 		free(meta, M_MD_INTEL);
649 		return (NULL);
650 	}
651 
652 	/* Validate metadata size. */
653 	size = sizeof(struct intel_raid_conf) +
654 	    sizeof(struct intel_raid_disk) * (meta->total_disks - 1) +
655 	    sizeof(struct intel_raid_vol) * meta->total_volumes;
656 	if (size > meta->config_size) {
657 badsize:
658 		G_RAID_DEBUG(1, "Intel metadata size incorrect %d < %d",
659 		    meta->config_size, size);
660 		free(meta, M_MD_INTEL);
661 		return (NULL);
662 	}
663 	for (i = 0; i < meta->total_volumes; i++) {
664 		mvol = intel_get_volume(meta, i);
665 		mmap = intel_get_map(mvol, 0);
666 		size += 4 * (mmap->total_disks - 1);
667 		if (size > meta->config_size)
668 			goto badsize;
669 		if (mvol->migr_state) {
670 			size += sizeof(struct intel_raid_map);
671 			if (size > meta->config_size)
672 				goto badsize;
673 			mmap = intel_get_map(mvol, 1);
674 			size += 4 * (mmap->total_disks - 1);
675 			if (size > meta->config_size)
676 				goto badsize;
677 		}
678 	}
679 
680 	g_raid_md_intel_print(meta);
681 
682 	if (strncmp(meta->version, INTEL_VERSION_1300, 6) > 0) {
683 		G_RAID_DEBUG(1, "Intel unsupported version: '%.6s'",
684 		    meta->version);
685 		free(meta, M_MD_INTEL);
686 		return (NULL);
687 	}
688 
689 	if (strncmp(meta->version, INTEL_VERSION_1300, 6) >= 0 &&
690 	    (meta->attributes & ~INTEL_ATTR_SUPPORTED) != 0) {
691 		G_RAID_DEBUG(1, "Intel unsupported attributes: 0x%08x",
692 		    meta->attributes & ~INTEL_ATTR_SUPPORTED);
693 		free(meta, M_MD_INTEL);
694 		return (NULL);
695 	}
696 
697 	/* Validate disk indexes. */
698 	for (i = 0; i < meta->total_volumes; i++) {
699 		mvol = intel_get_volume(meta, i);
700 		for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) {
701 			mmap = intel_get_map(mvol, j);
702 			for (k = 0; k < mmap->total_disks; k++) {
703 				if ((mmap->disk_idx[k] & INTEL_DI_IDX) >
704 				    meta->total_disks) {
705 					G_RAID_DEBUG(1, "Intel metadata disk"
706 					    " index %d too big (>%d)",
707 					    mmap->disk_idx[k] & INTEL_DI_IDX,
708 					    meta->total_disks);
709 					free(meta, M_MD_INTEL);
710 					return (NULL);
711 				}
712 			}
713 		}
714 	}
715 
716 	/* Validate migration types. */
717 	for (i = 0; i < meta->total_volumes; i++) {
718 		mvol = intel_get_volume(meta, i);
719 		/* Deny unknown migration types. */
720 		if (mvol->migr_state &&
721 		    mvol->migr_type != INTEL_MT_INIT &&
722 		    mvol->migr_type != INTEL_MT_REBUILD &&
723 		    mvol->migr_type != INTEL_MT_VERIFY &&
724 		    mvol->migr_type != INTEL_MT_GEN_MIGR &&
725 		    mvol->migr_type != INTEL_MT_REPAIR) {
726 			G_RAID_DEBUG(1, "Intel metadata has unsupported"
727 			    " migration type %d", mvol->migr_type);
728 			free(meta, M_MD_INTEL);
729 			return (NULL);
730 		}
731 		/* Deny general migrations except SINGLE->RAID1. */
732 		if (mvol->migr_state &&
733 		    mvol->migr_type == INTEL_MT_GEN_MIGR) {
734 			mmap = intel_get_map(mvol, 0);
735 			mmap1 = intel_get_map(mvol, 1);
736 			if (mmap1->total_disks != 1 ||
737 			    mmap->type != INTEL_T_RAID1 ||
738 			    mmap->total_disks != 2 ||
739 			    mmap->offset != mmap1->offset ||
740 			    mmap->disk_sectors != mmap1->disk_sectors ||
741 			    mmap->total_domains != mmap->total_disks ||
742 			    mmap->offset_hi != mmap1->offset_hi ||
743 			    mmap->disk_sectors_hi != mmap1->disk_sectors_hi ||
744 			    (mmap->disk_idx[0] != mmap1->disk_idx[0] &&
745 			     mmap->disk_idx[0] != mmap1->disk_idx[1])) {
746 				G_RAID_DEBUG(1, "Intel metadata has unsupported"
747 				    " variant of general migration");
748 				free(meta, M_MD_INTEL);
749 				return (NULL);
750 			}
751 		}
752 	}
753 
754 	return (meta);
755 }
756 
757 static int
758 intel_meta_write(struct g_consumer *cp, struct intel_raid_conf *meta)
759 {
760 	struct g_provider *pp;
761 	char *buf;
762 	int error, i, sectors;
763 	uint32_t checksum, *ptr;
764 
765 	pp = cp->provider;
766 
767 	/* Recalculate checksum for case if metadata were changed. */
768 	meta->checksum = 0;
769 	for (checksum = 0, ptr = (uint32_t *)meta, i = 0;
770 	    i < (meta->config_size / sizeof(uint32_t)); i++) {
771 		checksum += *ptr++;
772 	}
773 	meta->checksum = checksum;
774 
775 	/* Create and fill buffer. */
776 	sectors = howmany(meta->config_size, pp->sectorsize);
777 	buf = malloc(sectors * pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO);
778 	if (sectors > 1) {
779 		memcpy(buf, ((char *)meta) + pp->sectorsize,
780 		    (sectors - 1) * pp->sectorsize);
781 	}
782 	memcpy(buf + (sectors - 1) * pp->sectorsize, meta, pp->sectorsize);
783 
784 	error = g_write_data(cp,
785 	    pp->mediasize - pp->sectorsize * (1 + sectors),
786 	    buf, pp->sectorsize * sectors);
787 	if (error != 0) {
788 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
789 		    pp->name, error);
790 	}
791 
792 	free(buf, M_MD_INTEL);
793 	return (error);
794 }
795 
796 static int
797 intel_meta_erase(struct g_consumer *cp)
798 {
799 	struct g_provider *pp;
800 	char *buf;
801 	int error;
802 
803 	pp = cp->provider;
804 	buf = malloc(pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO);
805 	error = g_write_data(cp,
806 	    pp->mediasize - 2 * pp->sectorsize,
807 	    buf, pp->sectorsize);
808 	if (error != 0) {
809 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
810 		    pp->name, error);
811 	}
812 	free(buf, M_MD_INTEL);
813 	return (error);
814 }
815 
816 static int
817 intel_meta_write_spare(struct g_consumer *cp, struct intel_raid_disk *d)
818 {
819 	struct intel_raid_conf *meta;
820 	int error;
821 
822 	/* Fill anchor and single disk. */
823 	meta = malloc(INTEL_MAX_MD_SIZE(1), M_MD_INTEL, M_WAITOK | M_ZERO);
824 	memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1);
825 	memcpy(&meta->version[0], INTEL_VERSION_1000,
826 	    sizeof(INTEL_VERSION_1000) - 1);
827 	meta->config_size = INTEL_MAX_MD_SIZE(1);
828 	meta->config_id = meta->orig_config_id = arc4random();
829 	meta->generation = 1;
830 	meta->total_disks = 1;
831 	meta->disk[0] = *d;
832 	error = intel_meta_write(cp, meta);
833 	free(meta, M_MD_INTEL);
834 	return (error);
835 }
836 
837 static struct g_raid_disk *
838 g_raid_md_intel_get_disk(struct g_raid_softc *sc, int id)
839 {
840 	struct g_raid_disk	*disk;
841 	struct g_raid_md_intel_perdisk *pd;
842 
843 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
844 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
845 		if (pd->pd_disk_pos == id)
846 			break;
847 	}
848 	return (disk);
849 }
850 
851 static int
852 g_raid_md_intel_supported(int level, int qual, int disks, int force)
853 {
854 
855 	switch (level) {
856 	case G_RAID_VOLUME_RL_RAID0:
857 		if (disks < 1)
858 			return (0);
859 		if (!force && (disks < 2 || disks > 6))
860 			return (0);
861 		break;
862 	case G_RAID_VOLUME_RL_RAID1:
863 		if (disks < 1)
864 			return (0);
865 		if (!force && (disks != 2))
866 			return (0);
867 		break;
868 	case G_RAID_VOLUME_RL_RAID1E:
869 		if (disks < 2)
870 			return (0);
871 		if (!force && (disks != 4))
872 			return (0);
873 		break;
874 	case G_RAID_VOLUME_RL_RAID5:
875 		if (disks < 3)
876 			return (0);
877 		if (!force && disks > 6)
878 			return (0);
879 		if (qual != G_RAID_VOLUME_RLQ_R5LA)
880 			return (0);
881 		break;
882 	default:
883 		return (0);
884 	}
885 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
886 		return (0);
887 	return (1);
888 }
889 
890 static struct g_raid_volume *
891 g_raid_md_intel_get_volume(struct g_raid_softc *sc, int id)
892 {
893 	struct g_raid_volume	*mvol;
894 	struct g_raid_md_intel_pervolume *pv;
895 
896 	TAILQ_FOREACH(mvol, &sc->sc_volumes, v_next) {
897 		pv = mvol->v_md_data;
898 		if (pv->pv_volume_pos == id)
899 			break;
900 	}
901 	return (mvol);
902 }
903 
904 static int
905 g_raid_md_intel_start_disk(struct g_raid_disk *disk)
906 {
907 	struct g_raid_softc *sc;
908 	struct g_raid_subdisk *sd, *tmpsd;
909 	struct g_raid_disk *olddisk, *tmpdisk;
910 	struct g_raid_md_object *md;
911 	struct g_raid_md_intel_object *mdi;
912 	struct g_raid_md_intel_pervolume *pv;
913 	struct g_raid_md_intel_perdisk *pd, *oldpd;
914 	struct intel_raid_conf *meta;
915 	struct intel_raid_vol *mvol;
916 	struct intel_raid_map *mmap0, *mmap1;
917 	int disk_pos, resurrection = 0, migr_global, i;
918 
919 	sc = disk->d_softc;
920 	md = sc->sc_md;
921 	mdi = (struct g_raid_md_intel_object *)md;
922 	meta = mdi->mdio_meta;
923 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
924 	olddisk = NULL;
925 
926 	/* Find disk position in metadata by it's serial. */
927 	disk_pos = intel_meta_find_disk(meta, pd->pd_disk_meta.serial);
928 	if (disk_pos < 0) {
929 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
930 		/* Failed stale disk is useless for us. */
931 		if ((pd->pd_disk_meta.flags & INTEL_F_FAILED) &&
932 		    !(pd->pd_disk_meta.flags & INTEL_F_DISABLED)) {
933 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
934 			return (0);
935 		}
936 		/* If we are in the start process, that's all for now. */
937 		if (!mdi->mdio_started)
938 			goto nofit;
939 		/*
940 		 * If we have already started - try to get use of the disk.
941 		 * Try to replace OFFLINE disks first, then FAILED.
942 		 */
943 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
944 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
945 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
946 				continue;
947 			/* Make sure this disk is big enough. */
948 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
949 				off_t disk_sectors =
950 				    intel_get_disk_sectors(&pd->pd_disk_meta);
951 
952 				if (sd->sd_offset + sd->sd_size + 4096 >
953 				    disk_sectors * 512) {
954 					G_RAID_DEBUG1(1, sc,
955 					    "Disk too small (%llu < %llu)",
956 					    (unsigned long long)
957 					    disk_sectors * 512,
958 					    (unsigned long long)
959 					    sd->sd_offset + sd->sd_size + 4096);
960 					break;
961 				}
962 			}
963 			if (sd != NULL)
964 				continue;
965 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
966 				olddisk = tmpdisk;
967 				break;
968 			} else if (olddisk == NULL)
969 				olddisk = tmpdisk;
970 		}
971 		if (olddisk == NULL) {
972 nofit:
973 			if (pd->pd_disk_meta.flags & INTEL_F_SPARE) {
974 				g_raid_change_disk_state(disk,
975 				    G_RAID_DISK_S_SPARE);
976 				return (1);
977 			} else {
978 				g_raid_change_disk_state(disk,
979 				    G_RAID_DISK_S_STALE);
980 				return (0);
981 			}
982 		}
983 		oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data;
984 		disk_pos = oldpd->pd_disk_pos;
985 		resurrection = 1;
986 	}
987 
988 	if (olddisk == NULL) {
989 		/* Find placeholder by position. */
990 		olddisk = g_raid_md_intel_get_disk(sc, disk_pos);
991 		if (olddisk == NULL)
992 			panic("No disk at position %d!", disk_pos);
993 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
994 			G_RAID_DEBUG1(1, sc, "More than one disk for pos %d",
995 			    disk_pos);
996 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
997 			return (0);
998 		}
999 		oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data;
1000 	}
1001 
1002 	/* Replace failed disk or placeholder with new disk. */
1003 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
1004 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
1005 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1006 		sd->sd_disk = disk;
1007 	}
1008 	oldpd->pd_disk_pos = -2;
1009 	pd->pd_disk_pos = disk_pos;
1010 
1011 	/* If it was placeholder -- destroy it. */
1012 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
1013 		g_raid_destroy_disk(olddisk);
1014 	} else {
1015 		/* Otherwise, make it STALE_FAILED. */
1016 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
1017 		/* Update global metadata just in case. */
1018 		memcpy(&meta->disk[disk_pos], &pd->pd_disk_meta,
1019 		    sizeof(struct intel_raid_disk));
1020 	}
1021 
1022 	/* Welcome the new disk. */
1023 	if ((meta->disk[disk_pos].flags & INTEL_F_DISABLED) &&
1024 	    !(pd->pd_disk_meta.flags & INTEL_F_SPARE))
1025 		g_raid_change_disk_state(disk, G_RAID_DISK_S_DISABLED);
1026 	else if (resurrection)
1027 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
1028 	else if (meta->disk[disk_pos].flags & INTEL_F_FAILED)
1029 		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
1030 	else if (meta->disk[disk_pos].flags & INTEL_F_SPARE)
1031 		g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1032 	else
1033 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
1034 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1035 		pv = sd->sd_volume->v_md_data;
1036 		mvol = intel_get_volume(meta, pv->pv_volume_pos);
1037 		mmap0 = intel_get_map(mvol, 0);
1038 		if (mvol->migr_state)
1039 			mmap1 = intel_get_map(mvol, 1);
1040 		else
1041 			mmap1 = mmap0;
1042 
1043 		migr_global = 1;
1044 		for (i = 0; i < mmap0->total_disks; i++) {
1045 			if ((mmap0->disk_idx[i] & INTEL_DI_RBLD) == 0 &&
1046 			    (mmap1->disk_idx[i] & INTEL_DI_RBLD) != 0)
1047 				migr_global = 0;
1048 		}
1049 
1050 		if ((meta->disk[disk_pos].flags & INTEL_F_DISABLED) &&
1051 		    !(pd->pd_disk_meta.flags & INTEL_F_SPARE)) {
1052 			/* Disabled disk, useless. */
1053 			g_raid_change_subdisk_state(sd,
1054 			    G_RAID_SUBDISK_S_NONE);
1055 		} else if (resurrection) {
1056 			/* Stale disk, almost same as new. */
1057 			g_raid_change_subdisk_state(sd,
1058 			    G_RAID_SUBDISK_S_NEW);
1059 		} else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) {
1060 			/* Failed disk, almost useless. */
1061 			g_raid_change_subdisk_state(sd,
1062 			    G_RAID_SUBDISK_S_FAILED);
1063 		} else if (mvol->migr_state == 0) {
1064 			if (mmap0->status == INTEL_S_UNINITIALIZED &&
1065 			    (!pv->pv_cng || pv->pv_cng_master_disk != disk_pos)) {
1066 				/* Freshly created uninitialized volume. */
1067 				g_raid_change_subdisk_state(sd,
1068 				    G_RAID_SUBDISK_S_UNINITIALIZED);
1069 			} else if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
1070 				/* Freshly inserted disk. */
1071 				g_raid_change_subdisk_state(sd,
1072 				    G_RAID_SUBDISK_S_NEW);
1073 			} else if (mvol->dirty && (!pv->pv_cng ||
1074 			    pv->pv_cng_master_disk != disk_pos)) {
1075 				/* Dirty volume (unclean shutdown). */
1076 				g_raid_change_subdisk_state(sd,
1077 				    G_RAID_SUBDISK_S_STALE);
1078 			} else {
1079 				/* Up to date disk. */
1080 				g_raid_change_subdisk_state(sd,
1081 				    G_RAID_SUBDISK_S_ACTIVE);
1082 			}
1083 		} else if (mvol->migr_type == INTEL_MT_INIT ||
1084 			   mvol->migr_type == INTEL_MT_REBUILD) {
1085 			if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
1086 				/* Freshly inserted disk. */
1087 				g_raid_change_subdisk_state(sd,
1088 				    G_RAID_SUBDISK_S_NEW);
1089 			} else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
1090 				/* Rebuilding disk. */
1091 				g_raid_change_subdisk_state(sd,
1092 				    G_RAID_SUBDISK_S_REBUILD);
1093 				if (mvol->dirty) {
1094 					sd->sd_rebuild_pos = 0;
1095 				} else {
1096 					sd->sd_rebuild_pos =
1097 					    intel_get_vol_curr_migr_unit(mvol) *
1098 					    sd->sd_volume->v_strip_size *
1099 					    mmap0->total_domains;
1100 				}
1101 			} else if (mvol->migr_type == INTEL_MT_INIT &&
1102 			    migr_global) {
1103 				/* Freshly created uninitialized volume. */
1104 				g_raid_change_subdisk_state(sd,
1105 				    G_RAID_SUBDISK_S_UNINITIALIZED);
1106 			} else if (mvol->dirty && (!pv->pv_cng ||
1107 			    pv->pv_cng_master_disk != disk_pos)) {
1108 				/* Dirty volume (unclean shutdown). */
1109 				g_raid_change_subdisk_state(sd,
1110 				    G_RAID_SUBDISK_S_STALE);
1111 			} else {
1112 				/* Up to date disk. */
1113 				g_raid_change_subdisk_state(sd,
1114 				    G_RAID_SUBDISK_S_ACTIVE);
1115 			}
1116 		} else if (mvol->migr_type == INTEL_MT_VERIFY ||
1117 			   mvol->migr_type == INTEL_MT_REPAIR) {
1118 			if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) {
1119 				/* Freshly inserted disk. */
1120 				g_raid_change_subdisk_state(sd,
1121 				    G_RAID_SUBDISK_S_NEW);
1122 			} else if ((mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) ||
1123 			    migr_global) {
1124 				/* Resyncing disk. */
1125 				g_raid_change_subdisk_state(sd,
1126 				    G_RAID_SUBDISK_S_RESYNC);
1127 				if (mvol->dirty) {
1128 					sd->sd_rebuild_pos = 0;
1129 				} else {
1130 					sd->sd_rebuild_pos =
1131 					    intel_get_vol_curr_migr_unit(mvol) *
1132 					    sd->sd_volume->v_strip_size *
1133 					    mmap0->total_domains;
1134 				}
1135 			} else if (mvol->dirty) {
1136 				/* Dirty volume (unclean shutdown). */
1137 				g_raid_change_subdisk_state(sd,
1138 				    G_RAID_SUBDISK_S_STALE);
1139 			} else {
1140 				/* Up to date disk. */
1141 				g_raid_change_subdisk_state(sd,
1142 				    G_RAID_SUBDISK_S_ACTIVE);
1143 			}
1144 		} else if (mvol->migr_type == INTEL_MT_GEN_MIGR) {
1145 			if ((mmap1->disk_idx[0] & INTEL_DI_IDX) != disk_pos) {
1146 				/* Freshly inserted disk. */
1147 				g_raid_change_subdisk_state(sd,
1148 				    G_RAID_SUBDISK_S_NEW);
1149 			} else {
1150 				/* Up to date disk. */
1151 				g_raid_change_subdisk_state(sd,
1152 				    G_RAID_SUBDISK_S_ACTIVE);
1153 			}
1154 		}
1155 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1156 		    G_RAID_EVENT_SUBDISK);
1157 	}
1158 
1159 	/* Update status of our need for spare. */
1160 	if (mdi->mdio_started) {
1161 		mdi->mdio_incomplete =
1162 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) +
1163 		     g_raid_ndisks(sc, G_RAID_DISK_S_DISABLED) <
1164 		     meta->total_disks);
1165 	}
1166 
1167 	return (resurrection);
1168 }
1169 
1170 static void
1171 g_disk_md_intel_retaste(void *arg, int pending)
1172 {
1173 
1174 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
1175 	g_retaste(&g_raid_class);
1176 	free(arg, M_MD_INTEL);
1177 }
1178 
1179 static void
1180 g_raid_md_intel_refill(struct g_raid_softc *sc)
1181 {
1182 	struct g_raid_md_object *md;
1183 	struct g_raid_md_intel_object *mdi;
1184 	struct intel_raid_conf *meta;
1185 	struct g_raid_disk *disk;
1186 	struct task *task;
1187 	int update, na;
1188 
1189 	md = sc->sc_md;
1190 	mdi = (struct g_raid_md_intel_object *)md;
1191 	meta = mdi->mdio_meta;
1192 	update = 0;
1193 	do {
1194 		/* Make sure we miss anything. */
1195 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) +
1196 		    g_raid_ndisks(sc, G_RAID_DISK_S_DISABLED);
1197 		if (na == meta->total_disks)
1198 			break;
1199 
1200 		G_RAID_DEBUG1(1, md->mdo_softc,
1201 		    "Array is not complete (%d of %d), "
1202 		    "trying to refill.", na, meta->total_disks);
1203 
1204 		/* Try to get use some of STALE disks. */
1205 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1206 			if (disk->d_state == G_RAID_DISK_S_STALE) {
1207 				update += g_raid_md_intel_start_disk(disk);
1208 				if (disk->d_state == G_RAID_DISK_S_ACTIVE ||
1209 				    disk->d_state == G_RAID_DISK_S_DISABLED)
1210 					break;
1211 			}
1212 		}
1213 		if (disk != NULL)
1214 			continue;
1215 
1216 		/* Try to get use some of SPARE disks. */
1217 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1218 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
1219 				update += g_raid_md_intel_start_disk(disk);
1220 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
1221 					break;
1222 			}
1223 		}
1224 	} while (disk != NULL);
1225 
1226 	/* Write new metadata if we changed something. */
1227 	if (update) {
1228 		g_raid_md_write_intel(md, NULL, NULL, NULL);
1229 		meta = mdi->mdio_meta;
1230 	}
1231 
1232 	/* Update status of our need for spare. */
1233 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) +
1234 	    g_raid_ndisks(sc, G_RAID_DISK_S_DISABLED) < meta->total_disks);
1235 
1236 	/* Request retaste hoping to find spare. */
1237 	if (mdi->mdio_incomplete) {
1238 		task = malloc(sizeof(struct task),
1239 		    M_MD_INTEL, M_WAITOK | M_ZERO);
1240 		TASK_INIT(task, 0, g_disk_md_intel_retaste, task);
1241 		taskqueue_enqueue(taskqueue_swi, task);
1242 	}
1243 }
1244 
1245 static void
1246 g_raid_md_intel_start(struct g_raid_softc *sc)
1247 {
1248 	struct g_raid_md_object *md;
1249 	struct g_raid_md_intel_object *mdi;
1250 	struct g_raid_md_intel_pervolume *pv;
1251 	struct g_raid_md_intel_perdisk *pd;
1252 	struct intel_raid_conf *meta;
1253 	struct intel_raid_vol *mvol;
1254 	struct intel_raid_map *mmap;
1255 	struct g_raid_volume *vol;
1256 	struct g_raid_subdisk *sd;
1257 	struct g_raid_disk *disk;
1258 	int i, j, disk_pos;
1259 
1260 	md = sc->sc_md;
1261 	mdi = (struct g_raid_md_intel_object *)md;
1262 	meta = mdi->mdio_meta;
1263 
1264 	/* Create volumes and subdisks. */
1265 	for (i = 0; i < meta->total_volumes; i++) {
1266 		mvol = intel_get_volume(meta, i);
1267 		mmap = intel_get_map(mvol, 0);
1268 		vol = g_raid_create_volume(sc, mvol->name, mvol->tid - 1);
1269 		pv = malloc(sizeof(*pv), M_MD_INTEL, M_WAITOK | M_ZERO);
1270 		pv->pv_volume_pos = i;
1271 		pv->pv_cng = (mvol->state & INTEL_ST_CLONE_N_GO) != 0;
1272 		pv->pv_cng_man_sync = (mvol->state & INTEL_ST_CLONE_MAN_SYNC) != 0;
1273 		if (mvol->cng_master_disk < mmap->total_disks)
1274 			pv->pv_cng_master_disk = mvol->cng_master_disk;
1275 		vol->v_md_data = pv;
1276 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
1277 		if (mmap->type == INTEL_T_RAID0)
1278 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
1279 		else if (mmap->type == INTEL_T_RAID1 &&
1280 		    mmap->total_domains >= 2 &&
1281 		    mmap->total_domains <= mmap->total_disks) {
1282 			/* Assume total_domains is correct. */
1283 			if (mmap->total_domains == mmap->total_disks)
1284 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
1285 			else
1286 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
1287 		} else if (mmap->type == INTEL_T_RAID1) {
1288 			/* total_domains looks wrong. */
1289 			if (mmap->total_disks <= 2)
1290 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
1291 			else
1292 				vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
1293 		} else if (mmap->type == INTEL_T_RAID5) {
1294 			vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
1295 			vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
1296 		} else
1297 			vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1298 		vol->v_strip_size = (u_int)mmap->strip_sectors * 512; //ZZZ
1299 		vol->v_disks_count = mmap->total_disks;
1300 		vol->v_mediasize = (off_t)mvol->total_sectors * 512; //ZZZ
1301 		vol->v_sectorsize = 512; //ZZZ
1302 		for (j = 0; j < vol->v_disks_count; j++) {
1303 			sd = &vol->v_subdisks[j];
1304 			sd->sd_offset = intel_get_map_offset(mmap) * 512; //ZZZ
1305 			sd->sd_size = intel_get_map_disk_sectors(mmap) * 512; //ZZZ
1306 		}
1307 		g_raid_start_volume(vol);
1308 	}
1309 
1310 	/* Create disk placeholders to store data for later writing. */
1311 	for (disk_pos = 0; disk_pos < meta->total_disks; disk_pos++) {
1312 		pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
1313 		pd->pd_disk_pos = disk_pos;
1314 		pd->pd_disk_meta = meta->disk[disk_pos];
1315 		disk = g_raid_create_disk(sc);
1316 		disk->d_md_data = (void *)pd;
1317 		disk->d_state = G_RAID_DISK_S_OFFLINE;
1318 		for (i = 0; i < meta->total_volumes; i++) {
1319 			mvol = intel_get_volume(meta, i);
1320 			mmap = intel_get_map(mvol, 0);
1321 			for (j = 0; j < mmap->total_disks; j++) {
1322 				if ((mmap->disk_idx[j] & INTEL_DI_IDX) == disk_pos)
1323 					break;
1324 			}
1325 			if (j == mmap->total_disks)
1326 				continue;
1327 			vol = g_raid_md_intel_get_volume(sc, i);
1328 			sd = &vol->v_subdisks[j];
1329 			sd->sd_disk = disk;
1330 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1331 		}
1332 	}
1333 
1334 	/* Make all disks found till the moment take their places. */
1335 	do {
1336 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1337 			if (disk->d_state == G_RAID_DISK_S_NONE) {
1338 				g_raid_md_intel_start_disk(disk);
1339 				break;
1340 			}
1341 		}
1342 	} while (disk != NULL);
1343 
1344 	mdi->mdio_started = 1;
1345 	G_RAID_DEBUG1(0, sc, "Array started.");
1346 	g_raid_md_write_intel(md, NULL, NULL, NULL);
1347 
1348 	/* Pickup any STALE/SPARE disks to refill array if needed. */
1349 	g_raid_md_intel_refill(sc);
1350 
1351 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1352 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1353 		    G_RAID_EVENT_VOLUME);
1354 	}
1355 
1356 	callout_stop(&mdi->mdio_start_co);
1357 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
1358 	root_mount_rel(mdi->mdio_rootmount);
1359 	mdi->mdio_rootmount = NULL;
1360 }
1361 
1362 static void
1363 g_raid_md_intel_new_disk(struct g_raid_disk *disk)
1364 {
1365 	struct g_raid_softc *sc;
1366 	struct g_raid_md_object *md;
1367 	struct g_raid_md_intel_object *mdi;
1368 	struct intel_raid_conf *pdmeta;
1369 	struct g_raid_md_intel_perdisk *pd;
1370 
1371 	sc = disk->d_softc;
1372 	md = sc->sc_md;
1373 	mdi = (struct g_raid_md_intel_object *)md;
1374 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
1375 	pdmeta = pd->pd_meta;
1376 
1377 	if (mdi->mdio_started) {
1378 		if (g_raid_md_intel_start_disk(disk))
1379 			g_raid_md_write_intel(md, NULL, NULL, NULL);
1380 	} else {
1381 		/* If we haven't started yet - check metadata freshness. */
1382 		if (mdi->mdio_meta == NULL ||
1383 		    ((int32_t)(pdmeta->generation - mdi->mdio_generation)) > 0) {
1384 			G_RAID_DEBUG1(1, sc, "Newer disk");
1385 			if (mdi->mdio_meta != NULL)
1386 				free(mdi->mdio_meta, M_MD_INTEL);
1387 			mdi->mdio_meta = intel_meta_copy(pdmeta);
1388 			mdi->mdio_generation = mdi->mdio_meta->generation;
1389 			mdi->mdio_disks_present = 1;
1390 		} else if (pdmeta->generation == mdi->mdio_generation) {
1391 			mdi->mdio_disks_present++;
1392 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
1393 			    mdi->mdio_disks_present,
1394 			    mdi->mdio_meta->total_disks);
1395 		} else {
1396 			G_RAID_DEBUG1(1, sc, "Older disk");
1397 		}
1398 		/* If we collected all needed disks - start array. */
1399 		if (mdi->mdio_disks_present == mdi->mdio_meta->total_disks)
1400 			g_raid_md_intel_start(sc);
1401 	}
1402 }
1403 
1404 static void
1405 g_raid_intel_go(void *arg)
1406 {
1407 	struct g_raid_softc *sc;
1408 	struct g_raid_md_object *md;
1409 	struct g_raid_md_intel_object *mdi;
1410 
1411 	sc = arg;
1412 	md = sc->sc_md;
1413 	mdi = (struct g_raid_md_intel_object *)md;
1414 	if (!mdi->mdio_started) {
1415 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
1416 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
1417 	}
1418 }
1419 
1420 static int
1421 g_raid_md_create_intel(struct g_raid_md_object *md, struct g_class *mp,
1422     struct g_geom **gp)
1423 {
1424 	struct g_raid_softc *sc;
1425 	struct g_raid_md_intel_object *mdi;
1426 	char name[16];
1427 
1428 	mdi = (struct g_raid_md_intel_object *)md;
1429 	mdi->mdio_config_id = mdi->mdio_orig_config_id = arc4random();
1430 	mdi->mdio_generation = 0;
1431 	snprintf(name, sizeof(name), "Intel-%08x", mdi->mdio_config_id);
1432 	sc = g_raid_create_node(mp, name, md);
1433 	if (sc == NULL)
1434 		return (G_RAID_MD_TASTE_FAIL);
1435 	md->mdo_softc = sc;
1436 	*gp = sc->sc_geom;
1437 	return (G_RAID_MD_TASTE_NEW);
1438 }
1439 
1440 /*
1441  * Return the last N characters of the serial label.  The Linux and
1442  * ataraid(7) code always uses the last 16 characters of the label to
1443  * store into the Intel meta format.  Generalize this to N characters
1444  * since that's easy.  Labels can be up to 20 characters for SATA drives
1445  * and up 251 characters for SAS drives.  Since intel controllers don't
1446  * support SAS drives, just stick with the SATA limits for stack friendliness.
1447  */
1448 static int
1449 g_raid_md_get_label(struct g_consumer *cp, char *serial, int serlen)
1450 {
1451 	char serial_buffer[24];
1452 	int len, error;
1453 
1454 	len = sizeof(serial_buffer);
1455 	error = g_io_getattr("GEOM::ident", cp, &len, serial_buffer);
1456 	if (error != 0)
1457 		return (error);
1458 	len = strlen(serial_buffer);
1459 	if (len > serlen)
1460 		len -= serlen;
1461 	else
1462 		len = 0;
1463 	strncpy(serial, serial_buffer + len, serlen);
1464 	return (0);
1465 }
1466 
1467 static int
1468 g_raid_md_taste_intel(struct g_raid_md_object *md, struct g_class *mp,
1469                               struct g_consumer *cp, struct g_geom **gp)
1470 {
1471 	struct g_consumer *rcp;
1472 	struct g_provider *pp;
1473 	struct g_raid_md_intel_object *mdi, *mdi1;
1474 	struct g_raid_softc *sc;
1475 	struct g_raid_disk *disk;
1476 	struct intel_raid_conf *meta;
1477 	struct g_raid_md_intel_perdisk *pd;
1478 	struct g_geom *geom;
1479 	int error, disk_pos, result, spare, len;
1480 	char serial[INTEL_SERIAL_LEN];
1481 	char name[16];
1482 	uint16_t vendor;
1483 
1484 	G_RAID_DEBUG(1, "Tasting Intel on %s", cp->provider->name);
1485 	mdi = (struct g_raid_md_intel_object *)md;
1486 	pp = cp->provider;
1487 
1488 	/* Read metadata from device. */
1489 	meta = NULL;
1490 	disk_pos = 0;
1491 	g_topology_unlock();
1492 	error = g_raid_md_get_label(cp, serial, sizeof(serial));
1493 	if (error != 0) {
1494 		G_RAID_DEBUG(1, "Cannot get serial number from %s (error=%d).",
1495 		    pp->name, error);
1496 		goto fail2;
1497 	}
1498 	vendor = 0xffff;
1499 	len = sizeof(vendor);
1500 	if (pp->geom->rank == 1)
1501 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1502 	meta = intel_meta_read(cp);
1503 	g_topology_lock();
1504 	if (meta == NULL) {
1505 		if (g_raid_aggressive_spare) {
1506 			if (vendor != 0x8086) {
1507 				G_RAID_DEBUG(1,
1508 				    "Intel vendor mismatch 0x%04x != 0x8086",
1509 				    vendor);
1510 			} else {
1511 				G_RAID_DEBUG(1,
1512 				    "No Intel metadata, forcing spare.");
1513 				spare = 2;
1514 				goto search;
1515 			}
1516 		}
1517 		return (G_RAID_MD_TASTE_FAIL);
1518 	}
1519 
1520 	/* Check this disk position in obtained metadata. */
1521 	disk_pos = intel_meta_find_disk(meta, serial);
1522 	if (disk_pos < 0) {
1523 		G_RAID_DEBUG(1, "Intel serial '%s' not found", serial);
1524 		goto fail1;
1525 	}
1526 	if (intel_get_disk_sectors(&meta->disk[disk_pos]) !=
1527 	    (pp->mediasize / pp->sectorsize)) {
1528 		G_RAID_DEBUG(1, "Intel size mismatch %ju != %ju",
1529 		    intel_get_disk_sectors(&meta->disk[disk_pos]),
1530 		    (off_t)(pp->mediasize / pp->sectorsize));
1531 		goto fail1;
1532 	}
1533 
1534 	G_RAID_DEBUG(1, "Intel disk position %d", disk_pos);
1535 	spare = meta->disk[disk_pos].flags & INTEL_F_SPARE;
1536 
1537 search:
1538 	/* Search for matching node. */
1539 	sc = NULL;
1540 	mdi1 = NULL;
1541 	LIST_FOREACH(geom, &mp->geom, geom) {
1542 		sc = geom->softc;
1543 		if (sc == NULL)
1544 			continue;
1545 		if (sc->sc_stopping != 0)
1546 			continue;
1547 		if (sc->sc_md->mdo_class != md->mdo_class)
1548 			continue;
1549 		mdi1 = (struct g_raid_md_intel_object *)sc->sc_md;
1550 		if (spare) {
1551 			if (mdi1->mdio_incomplete)
1552 				break;
1553 		} else {
1554 			if (mdi1->mdio_config_id == meta->config_id)
1555 				break;
1556 		}
1557 	}
1558 
1559 	/* Found matching node. */
1560 	if (geom != NULL) {
1561 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1562 		result = G_RAID_MD_TASTE_EXISTING;
1563 
1564 	} else if (spare) { /* Not found needy node -- left for later. */
1565 		G_RAID_DEBUG(1, "Spare is not needed at this time");
1566 		goto fail1;
1567 
1568 	} else { /* Not found matching node -- create one. */
1569 		result = G_RAID_MD_TASTE_NEW;
1570 		mdi->mdio_config_id = meta->config_id;
1571 		mdi->mdio_orig_config_id = meta->orig_config_id;
1572 		snprintf(name, sizeof(name), "Intel-%08x", meta->config_id);
1573 		sc = g_raid_create_node(mp, name, md);
1574 		md->mdo_softc = sc;
1575 		geom = sc->sc_geom;
1576 		callout_init(&mdi->mdio_start_co, 1);
1577 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
1578 		    g_raid_intel_go, sc);
1579 		mdi->mdio_rootmount = root_mount_hold("GRAID-Intel");
1580 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
1581 	}
1582 
1583 	/* There is no return after this point, so we close passed consumer. */
1584 	g_access(cp, -1, 0, 0);
1585 
1586 	rcp = g_new_consumer(geom);
1587 	rcp->flags |= G_CF_DIRECT_RECEIVE;
1588 	g_attach(rcp, pp);
1589 	if (g_access(rcp, 1, 1, 1) != 0)
1590 		; //goto fail1;
1591 
1592 	g_topology_unlock();
1593 	sx_xlock(&sc->sc_lock);
1594 
1595 	pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
1596 	pd->pd_meta = meta;
1597 	pd->pd_disk_pos = -1;
1598 	if (spare == 2) {
1599 		memcpy(&pd->pd_disk_meta.serial[0], serial, INTEL_SERIAL_LEN);
1600 		intel_set_disk_sectors(&pd->pd_disk_meta,
1601 		    pp->mediasize / pp->sectorsize);
1602 		pd->pd_disk_meta.id = 0;
1603 		pd->pd_disk_meta.flags = INTEL_F_SPARE;
1604 	} else {
1605 		pd->pd_disk_meta = meta->disk[disk_pos];
1606 	}
1607 	disk = g_raid_create_disk(sc);
1608 	disk->d_md_data = (void *)pd;
1609 	disk->d_consumer = rcp;
1610 	rcp->private = disk;
1611 
1612 	g_raid_get_disk_info(disk);
1613 
1614 	g_raid_md_intel_new_disk(disk);
1615 
1616 	sx_xunlock(&sc->sc_lock);
1617 	g_topology_lock();
1618 	*gp = geom;
1619 	return (result);
1620 fail2:
1621 	g_topology_lock();
1622 fail1:
1623 	free(meta, M_MD_INTEL);
1624 	return (G_RAID_MD_TASTE_FAIL);
1625 }
1626 
1627 static int
1628 g_raid_md_event_intel(struct g_raid_md_object *md,
1629     struct g_raid_disk *disk, u_int event)
1630 {
1631 	struct g_raid_softc *sc;
1632 	struct g_raid_subdisk *sd;
1633 	struct g_raid_md_intel_object *mdi;
1634 	struct g_raid_md_intel_perdisk *pd;
1635 
1636 	sc = md->mdo_softc;
1637 	mdi = (struct g_raid_md_intel_object *)md;
1638 	if (disk == NULL) {
1639 		switch (event) {
1640 		case G_RAID_NODE_E_START:
1641 			if (!mdi->mdio_started)
1642 				g_raid_md_intel_start(sc);
1643 			return (0);
1644 		}
1645 		return (-1);
1646 	}
1647 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
1648 	switch (event) {
1649 	case G_RAID_DISK_E_DISCONNECTED:
1650 		/* If disk was assigned, just update statuses. */
1651 		if (pd->pd_disk_pos >= 0) {
1652 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1653 			if (disk->d_consumer) {
1654 				g_raid_kill_consumer(sc, disk->d_consumer);
1655 				disk->d_consumer = NULL;
1656 			}
1657 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1658 				g_raid_change_subdisk_state(sd,
1659 				    G_RAID_SUBDISK_S_NONE);
1660 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1661 				    G_RAID_EVENT_SUBDISK);
1662 			}
1663 		} else {
1664 			/* Otherwise -- delete. */
1665 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1666 			g_raid_destroy_disk(disk);
1667 		}
1668 
1669 		/* Write updated metadata to all disks. */
1670 		g_raid_md_write_intel(md, NULL, NULL, NULL);
1671 
1672 		/* Check if anything left except placeholders. */
1673 		if (g_raid_ndisks(sc, -1) ==
1674 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1675 			g_raid_destroy_node(sc, 0);
1676 		else
1677 			g_raid_md_intel_refill(sc);
1678 		return (0);
1679 	}
1680 	return (-2);
1681 }
1682 
1683 static int
1684 g_raid_md_ctl_intel(struct g_raid_md_object *md,
1685     struct gctl_req *req)
1686 {
1687 	struct g_raid_softc *sc;
1688 	struct g_raid_volume *vol, *vol1;
1689 	struct g_raid_subdisk *sd;
1690 	struct g_raid_disk *disk;
1691 	struct g_raid_md_intel_object *mdi;
1692 	struct g_raid_md_intel_pervolume *pv;
1693 	struct g_raid_md_intel_perdisk *pd;
1694 	struct g_consumer *cp;
1695 	struct g_provider *pp;
1696 	char arg[16], serial[INTEL_SERIAL_LEN];
1697 	const char *nodename, *verb, *volname, *levelname, *diskname;
1698 	char *tmp;
1699 	int *nargs, *force;
1700 	off_t off, size, sectorsize, strip, disk_sectors;
1701 	intmax_t *sizearg, *striparg;
1702 	int numdisks, i, len, level, qual, update;
1703 	int error;
1704 
1705 	sc = md->mdo_softc;
1706 	mdi = (struct g_raid_md_intel_object *)md;
1707 	verb = gctl_get_param(req, "verb", NULL);
1708 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1709 	error = 0;
1710 	if (strcmp(verb, "label") == 0) {
1711 
1712 		if (*nargs < 4) {
1713 			gctl_error(req, "Invalid number of arguments.");
1714 			return (-1);
1715 		}
1716 		volname = gctl_get_asciiparam(req, "arg1");
1717 		if (volname == NULL) {
1718 			gctl_error(req, "No volume name.");
1719 			return (-2);
1720 		}
1721 		levelname = gctl_get_asciiparam(req, "arg2");
1722 		if (levelname == NULL) {
1723 			gctl_error(req, "No RAID level.");
1724 			return (-3);
1725 		}
1726 		if (strcasecmp(levelname, "RAID5") == 0)
1727 			levelname = "RAID5-LA";
1728 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1729 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1730 			return (-4);
1731 		}
1732 		numdisks = *nargs - 3;
1733 		force = gctl_get_paraml(req, "force", sizeof(*force));
1734 		if (!g_raid_md_intel_supported(level, qual, numdisks,
1735 		    force ? *force : 0)) {
1736 			gctl_error(req, "Unsupported RAID level "
1737 			    "(0x%02x/0x%02x), or number of disks (%d).",
1738 			    level, qual, numdisks);
1739 			return (-5);
1740 		}
1741 
1742 		/* Search for disks, connect them and probe. */
1743 		size = 0x7fffffffffffffffllu;
1744 		sectorsize = 0;
1745 		for (i = 0; i < numdisks; i++) {
1746 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1747 			diskname = gctl_get_asciiparam(req, arg);
1748 			if (diskname == NULL) {
1749 				gctl_error(req, "No disk name (%s).", arg);
1750 				error = -6;
1751 				break;
1752 			}
1753 			if (strcmp(diskname, "NONE") == 0) {
1754 				cp = NULL;
1755 				pp = NULL;
1756 			} else {
1757 				g_topology_lock();
1758 				cp = g_raid_open_consumer(sc, diskname);
1759 				if (cp == NULL) {
1760 					gctl_error(req, "Can't open disk '%s'.",
1761 					    diskname);
1762 					g_topology_unlock();
1763 					error = -7;
1764 					break;
1765 				}
1766 				pp = cp->provider;
1767 			}
1768 			pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
1769 			pd->pd_disk_pos = i;
1770 			disk = g_raid_create_disk(sc);
1771 			disk->d_md_data = (void *)pd;
1772 			disk->d_consumer = cp;
1773 			if (cp == NULL) {
1774 				strcpy(&pd->pd_disk_meta.serial[0], "NONE");
1775 				pd->pd_disk_meta.id = 0xffffffff;
1776 				pd->pd_disk_meta.flags = INTEL_F_ASSIGNED;
1777 				continue;
1778 			}
1779 			cp->private = disk;
1780 			g_topology_unlock();
1781 
1782 			error = g_raid_md_get_label(cp,
1783 			    &pd->pd_disk_meta.serial[0], INTEL_SERIAL_LEN);
1784 			if (error != 0) {
1785 				gctl_error(req,
1786 				    "Can't get serial for provider '%s'.",
1787 				    diskname);
1788 				error = -8;
1789 				break;
1790 			}
1791 
1792 			g_raid_get_disk_info(disk);
1793 
1794 			intel_set_disk_sectors(&pd->pd_disk_meta,
1795 			    pp->mediasize / pp->sectorsize);
1796 			if (size > pp->mediasize)
1797 				size = pp->mediasize;
1798 			if (sectorsize < pp->sectorsize)
1799 				sectorsize = pp->sectorsize;
1800 			pd->pd_disk_meta.id = 0;
1801 			pd->pd_disk_meta.flags = INTEL_F_ASSIGNED | INTEL_F_ONLINE;
1802 		}
1803 		if (error != 0)
1804 			return (error);
1805 
1806 		if (sectorsize <= 0) {
1807 			gctl_error(req, "Can't get sector size.");
1808 			return (-8);
1809 		}
1810 
1811 		/* Reserve some space for metadata. */
1812 		size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize;
1813 
1814 		/* Handle size argument. */
1815 		len = sizeof(*sizearg);
1816 		sizearg = gctl_get_param(req, "size", &len);
1817 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1818 		    *sizearg > 0) {
1819 			if (*sizearg > size) {
1820 				gctl_error(req, "Size too big %lld > %lld.",
1821 				    (long long)*sizearg, (long long)size);
1822 				return (-9);
1823 			}
1824 			size = *sizearg;
1825 		}
1826 
1827 		/* Handle strip argument. */
1828 		strip = 131072;
1829 		len = sizeof(*striparg);
1830 		striparg = gctl_get_param(req, "strip", &len);
1831 		if (striparg != NULL && len == sizeof(*striparg) &&
1832 		    *striparg > 0) {
1833 			if (*striparg < sectorsize) {
1834 				gctl_error(req, "Strip size too small.");
1835 				return (-10);
1836 			}
1837 			if (*striparg % sectorsize != 0) {
1838 				gctl_error(req, "Incorrect strip size.");
1839 				return (-11);
1840 			}
1841 			if (strip > 65535 * sectorsize) {
1842 				gctl_error(req, "Strip size too big.");
1843 				return (-12);
1844 			}
1845 			strip = *striparg;
1846 		}
1847 
1848 		/* Round size down to strip or sector. */
1849 		if (level == G_RAID_VOLUME_RL_RAID1)
1850 			size -= (size % sectorsize);
1851 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1852 		    (numdisks & 1) != 0)
1853 			size -= (size % (2 * strip));
1854 		else
1855 			size -= (size % strip);
1856 		if (size <= 0) {
1857 			gctl_error(req, "Size too small.");
1858 			return (-13);
1859 		}
1860 
1861 		/* We have all we need, create things: volume, ... */
1862 		mdi->mdio_started = 1;
1863 		vol = g_raid_create_volume(sc, volname, -1);
1864 		pv = malloc(sizeof(*pv), M_MD_INTEL, M_WAITOK | M_ZERO);
1865 		pv->pv_volume_pos = 0;
1866 		vol->v_md_data = pv;
1867 		vol->v_raid_level = level;
1868 		vol->v_raid_level_qualifier = qual;
1869 		vol->v_strip_size = strip;
1870 		vol->v_disks_count = numdisks;
1871 		if (level == G_RAID_VOLUME_RL_RAID0)
1872 			vol->v_mediasize = size * numdisks;
1873 		else if (level == G_RAID_VOLUME_RL_RAID1)
1874 			vol->v_mediasize = size;
1875 		else if (level == G_RAID_VOLUME_RL_RAID5)
1876 			vol->v_mediasize = size * (numdisks - 1);
1877 		else { /* RAID1E */
1878 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1879 			    strip;
1880 		}
1881 		vol->v_sectorsize = sectorsize;
1882 		g_raid_start_volume(vol);
1883 
1884 		/* , and subdisks. */
1885 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1886 			pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
1887 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1888 			sd->sd_disk = disk;
1889 			sd->sd_offset = 0;
1890 			sd->sd_size = size;
1891 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1892 			if (sd->sd_disk->d_consumer != NULL) {
1893 				g_raid_change_disk_state(disk,
1894 				    G_RAID_DISK_S_ACTIVE);
1895 				if (level == G_RAID_VOLUME_RL_RAID5)
1896 					g_raid_change_subdisk_state(sd,
1897 					    G_RAID_SUBDISK_S_UNINITIALIZED);
1898 				else
1899 					g_raid_change_subdisk_state(sd,
1900 					    G_RAID_SUBDISK_S_ACTIVE);
1901 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1902 				    G_RAID_EVENT_SUBDISK);
1903 			} else {
1904 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1905 			}
1906 		}
1907 
1908 		/* Write metadata based on created entities. */
1909 		G_RAID_DEBUG1(0, sc, "Array started.");
1910 		g_raid_md_write_intel(md, NULL, NULL, NULL);
1911 
1912 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1913 		g_raid_md_intel_refill(sc);
1914 
1915 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1916 		    G_RAID_EVENT_VOLUME);
1917 		return (0);
1918 	}
1919 	if (strcmp(verb, "add") == 0) {
1920 
1921 		if (*nargs != 3) {
1922 			gctl_error(req, "Invalid number of arguments.");
1923 			return (-1);
1924 		}
1925 		volname = gctl_get_asciiparam(req, "arg1");
1926 		if (volname == NULL) {
1927 			gctl_error(req, "No volume name.");
1928 			return (-2);
1929 		}
1930 		levelname = gctl_get_asciiparam(req, "arg2");
1931 		if (levelname == NULL) {
1932 			gctl_error(req, "No RAID level.");
1933 			return (-3);
1934 		}
1935 		if (strcasecmp(levelname, "RAID5") == 0)
1936 			levelname = "RAID5-LA";
1937 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1938 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1939 			return (-4);
1940 		}
1941 
1942 		/* Look for existing volumes. */
1943 		i = 0;
1944 		vol1 = NULL;
1945 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1946 			vol1 = vol;
1947 			i++;
1948 		}
1949 		if (i > 1) {
1950 			gctl_error(req, "Maximum two volumes supported.");
1951 			return (-6);
1952 		}
1953 		if (vol1 == NULL) {
1954 			gctl_error(req, "At least one volume must exist.");
1955 			return (-7);
1956 		}
1957 
1958 		numdisks = vol1->v_disks_count;
1959 		force = gctl_get_paraml(req, "force", sizeof(*force));
1960 		if (!g_raid_md_intel_supported(level, qual, numdisks,
1961 		    force ? *force : 0)) {
1962 			gctl_error(req, "Unsupported RAID level "
1963 			    "(0x%02x/0x%02x), or number of disks (%d).",
1964 			    level, qual, numdisks);
1965 			return (-5);
1966 		}
1967 
1968 		/* Collect info about present disks. */
1969 		size = 0x7fffffffffffffffllu;
1970 		sectorsize = 512;
1971 		for (i = 0; i < numdisks; i++) {
1972 			disk = vol1->v_subdisks[i].sd_disk;
1973 			pd = (struct g_raid_md_intel_perdisk *)
1974 			    disk->d_md_data;
1975 			disk_sectors =
1976 			    intel_get_disk_sectors(&pd->pd_disk_meta);
1977 
1978 			if (disk_sectors * 512 < size)
1979 				size = disk_sectors * 512;
1980 			if (disk->d_consumer != NULL &&
1981 			    disk->d_consumer->provider != NULL &&
1982 			    disk->d_consumer->provider->sectorsize >
1983 			     sectorsize) {
1984 				sectorsize =
1985 				    disk->d_consumer->provider->sectorsize;
1986 			}
1987 		}
1988 
1989 		/* Reserve some space for metadata. */
1990 		size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize;
1991 
1992 		/* Decide insert before or after. */
1993 		sd = &vol1->v_subdisks[0];
1994 		if (sd->sd_offset >
1995 		    size - (sd->sd_offset + sd->sd_size)) {
1996 			off = 0;
1997 			size = sd->sd_offset;
1998 		} else {
1999 			off = sd->sd_offset + sd->sd_size;
2000 			size = size - (sd->sd_offset + sd->sd_size);
2001 		}
2002 
2003 		/* Handle strip argument. */
2004 		strip = 131072;
2005 		len = sizeof(*striparg);
2006 		striparg = gctl_get_param(req, "strip", &len);
2007 		if (striparg != NULL && len == sizeof(*striparg) &&
2008 		    *striparg > 0) {
2009 			if (*striparg < sectorsize) {
2010 				gctl_error(req, "Strip size too small.");
2011 				return (-10);
2012 			}
2013 			if (*striparg % sectorsize != 0) {
2014 				gctl_error(req, "Incorrect strip size.");
2015 				return (-11);
2016 			}
2017 			if (strip > 65535 * sectorsize) {
2018 				gctl_error(req, "Strip size too big.");
2019 				return (-12);
2020 			}
2021 			strip = *striparg;
2022 		}
2023 
2024 		/* Round offset up to strip. */
2025 		if (off % strip != 0) {
2026 			size -= strip - off % strip;
2027 			off += strip - off % strip;
2028 		}
2029 
2030 		/* Handle size argument. */
2031 		len = sizeof(*sizearg);
2032 		sizearg = gctl_get_param(req, "size", &len);
2033 		if (sizearg != NULL && len == sizeof(*sizearg) &&
2034 		    *sizearg > 0) {
2035 			if (*sizearg > size) {
2036 				gctl_error(req, "Size too big %lld > %lld.",
2037 				    (long long)*sizearg, (long long)size);
2038 				return (-9);
2039 			}
2040 			size = *sizearg;
2041 		}
2042 
2043 		/* Round size down to strip or sector. */
2044 		if (level == G_RAID_VOLUME_RL_RAID1)
2045 			size -= (size % sectorsize);
2046 		else
2047 			size -= (size % strip);
2048 		if (size <= 0) {
2049 			gctl_error(req, "Size too small.");
2050 			return (-13);
2051 		}
2052 		if (size > 0xffffffffllu * sectorsize) {
2053 			gctl_error(req, "Size too big.");
2054 			return (-14);
2055 		}
2056 
2057 		/* We have all we need, create things: volume, ... */
2058 		vol = g_raid_create_volume(sc, volname, -1);
2059 		pv = malloc(sizeof(*pv), M_MD_INTEL, M_WAITOK | M_ZERO);
2060 		pv->pv_volume_pos = i;
2061 		vol->v_md_data = pv;
2062 		vol->v_raid_level = level;
2063 		vol->v_raid_level_qualifier = qual;
2064 		vol->v_strip_size = strip;
2065 		vol->v_disks_count = numdisks;
2066 		if (level == G_RAID_VOLUME_RL_RAID0)
2067 			vol->v_mediasize = size * numdisks;
2068 		else if (level == G_RAID_VOLUME_RL_RAID1)
2069 			vol->v_mediasize = size;
2070 		else if (level == G_RAID_VOLUME_RL_RAID5)
2071 			vol->v_mediasize = size * (numdisks - 1);
2072 		else { /* RAID1E */
2073 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
2074 			    strip;
2075 		}
2076 		vol->v_sectorsize = sectorsize;
2077 		g_raid_start_volume(vol);
2078 
2079 		/* , and subdisks. */
2080 		for (i = 0; i < numdisks; i++) {
2081 			disk = vol1->v_subdisks[i].sd_disk;
2082 			sd = &vol->v_subdisks[i];
2083 			sd->sd_disk = disk;
2084 			sd->sd_offset = off;
2085 			sd->sd_size = size;
2086 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
2087 			if (disk->d_state == G_RAID_DISK_S_ACTIVE) {
2088 				if (level == G_RAID_VOLUME_RL_RAID5)
2089 					g_raid_change_subdisk_state(sd,
2090 					    G_RAID_SUBDISK_S_UNINITIALIZED);
2091 				else
2092 					g_raid_change_subdisk_state(sd,
2093 					    G_RAID_SUBDISK_S_ACTIVE);
2094 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
2095 				    G_RAID_EVENT_SUBDISK);
2096 			}
2097 		}
2098 
2099 		/* Write metadata based on created entities. */
2100 		g_raid_md_write_intel(md, NULL, NULL, NULL);
2101 
2102 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
2103 		    G_RAID_EVENT_VOLUME);
2104 		return (0);
2105 	}
2106 	if (strcmp(verb, "delete") == 0) {
2107 
2108 		nodename = gctl_get_asciiparam(req, "arg0");
2109 		if (nodename != NULL && strcasecmp(sc->sc_name, nodename) != 0)
2110 			nodename = NULL;
2111 
2112 		/* Full node destruction. */
2113 		if (*nargs == 1 && nodename != NULL) {
2114 			/* Check if some volume is still open. */
2115 			force = gctl_get_paraml(req, "force", sizeof(*force));
2116 			if (force != NULL && *force == 0 &&
2117 			    g_raid_nopens(sc) != 0) {
2118 				gctl_error(req, "Some volume is still open.");
2119 				return (-4);
2120 			}
2121 
2122 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2123 				if (disk->d_consumer)
2124 					intel_meta_erase(disk->d_consumer);
2125 			}
2126 			g_raid_destroy_node(sc, 0);
2127 			return (0);
2128 		}
2129 
2130 		/* Destroy specified volume. If it was last - all node. */
2131 		if (*nargs > 2) {
2132 			gctl_error(req, "Invalid number of arguments.");
2133 			return (-1);
2134 		}
2135 		volname = gctl_get_asciiparam(req,
2136 		    nodename != NULL ? "arg1" : "arg0");
2137 		if (volname == NULL) {
2138 			gctl_error(req, "No volume name.");
2139 			return (-2);
2140 		}
2141 
2142 		/* Search for volume. */
2143 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2144 			if (strcmp(vol->v_name, volname) == 0)
2145 				break;
2146 			pp = vol->v_provider;
2147 			if (pp == NULL)
2148 				continue;
2149 			if (strcmp(pp->name, volname) == 0)
2150 				break;
2151 			if (strncmp(pp->name, "raid/", 5) == 0 &&
2152 			    strcmp(pp->name + 5, volname) == 0)
2153 				break;
2154 		}
2155 		if (vol == NULL) {
2156 			i = strtol(volname, &tmp, 10);
2157 			if (verb != volname && tmp[0] == 0) {
2158 				TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2159 					if (vol->v_global_id == i)
2160 						break;
2161 				}
2162 			}
2163 		}
2164 		if (vol == NULL) {
2165 			gctl_error(req, "Volume '%s' not found.", volname);
2166 			return (-3);
2167 		}
2168 
2169 		/* Check if volume is still open. */
2170 		force = gctl_get_paraml(req, "force", sizeof(*force));
2171 		if (force != NULL && *force == 0 &&
2172 		    vol->v_provider_open != 0) {
2173 			gctl_error(req, "Volume is still open.");
2174 			return (-4);
2175 		}
2176 
2177 		/* Destroy volume and potentially node. */
2178 		i = 0;
2179 		TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
2180 			i++;
2181 		if (i >= 2) {
2182 			g_raid_destroy_volume(vol);
2183 			g_raid_md_write_intel(md, NULL, NULL, NULL);
2184 		} else {
2185 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2186 				if (disk->d_consumer)
2187 					intel_meta_erase(disk->d_consumer);
2188 			}
2189 			g_raid_destroy_node(sc, 0);
2190 		}
2191 		return (0);
2192 	}
2193 	if (strcmp(verb, "remove") == 0 ||
2194 	    strcmp(verb, "fail") == 0) {
2195 		if (*nargs < 2) {
2196 			gctl_error(req, "Invalid number of arguments.");
2197 			return (-1);
2198 		}
2199 		for (i = 1; i < *nargs; i++) {
2200 			snprintf(arg, sizeof(arg), "arg%d", i);
2201 			diskname = gctl_get_asciiparam(req, arg);
2202 			if (diskname == NULL) {
2203 				gctl_error(req, "No disk name (%s).", arg);
2204 				error = -2;
2205 				break;
2206 			}
2207 			if (strncmp(diskname, "/dev/", 5) == 0)
2208 				diskname += 5;
2209 
2210 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2211 				if (disk->d_consumer != NULL &&
2212 				    disk->d_consumer->provider != NULL &&
2213 				    strcmp(disk->d_consumer->provider->name,
2214 				     diskname) == 0)
2215 					break;
2216 			}
2217 			if (disk == NULL) {
2218 				gctl_error(req, "Disk '%s' not found.",
2219 				    diskname);
2220 				error = -3;
2221 				break;
2222 			}
2223 
2224 			if (strcmp(verb, "fail") == 0) {
2225 				g_raid_md_fail_disk_intel(md, NULL, disk);
2226 				continue;
2227 			}
2228 
2229 			pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2230 
2231 			/* Erase metadata on deleting disk. */
2232 			intel_meta_erase(disk->d_consumer);
2233 
2234 			/* If disk was assigned, just update statuses. */
2235 			if (pd->pd_disk_pos >= 0) {
2236 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
2237 				g_raid_kill_consumer(sc, disk->d_consumer);
2238 				disk->d_consumer = NULL;
2239 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2240 					g_raid_change_subdisk_state(sd,
2241 					    G_RAID_SUBDISK_S_NONE);
2242 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2243 					    G_RAID_EVENT_SUBDISK);
2244 				}
2245 			} else {
2246 				/* Otherwise -- delete. */
2247 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
2248 				g_raid_destroy_disk(disk);
2249 			}
2250 		}
2251 
2252 		/* Write updated metadata to remaining disks. */
2253 		g_raid_md_write_intel(md, NULL, NULL, NULL);
2254 
2255 		/* Check if anything left except placeholders. */
2256 		if (g_raid_ndisks(sc, -1) ==
2257 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
2258 			g_raid_destroy_node(sc, 0);
2259 		else
2260 			g_raid_md_intel_refill(sc);
2261 		return (error);
2262 	}
2263 	if (strcmp(verb, "insert") == 0) {
2264 		if (*nargs < 2) {
2265 			gctl_error(req, "Invalid number of arguments.");
2266 			return (-1);
2267 		}
2268 		update = 0;
2269 		for (i = 1; i < *nargs; i++) {
2270 			/* Get disk name. */
2271 			snprintf(arg, sizeof(arg), "arg%d", i);
2272 			diskname = gctl_get_asciiparam(req, arg);
2273 			if (diskname == NULL) {
2274 				gctl_error(req, "No disk name (%s).", arg);
2275 				error = -3;
2276 				break;
2277 			}
2278 
2279 			/* Try to find provider with specified name. */
2280 			g_topology_lock();
2281 			cp = g_raid_open_consumer(sc, diskname);
2282 			if (cp == NULL) {
2283 				gctl_error(req, "Can't open disk '%s'.",
2284 				    diskname);
2285 				g_topology_unlock();
2286 				error = -4;
2287 				break;
2288 			}
2289 			pp = cp->provider;
2290 			g_topology_unlock();
2291 
2292 			/* Read disk serial. */
2293 			error = g_raid_md_get_label(cp,
2294 			    &serial[0], INTEL_SERIAL_LEN);
2295 			if (error != 0) {
2296 				gctl_error(req,
2297 				    "Can't get serial for provider '%s'.",
2298 				    diskname);
2299 				g_raid_kill_consumer(sc, cp);
2300 				error = -7;
2301 				break;
2302 			}
2303 
2304 			pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO);
2305 			pd->pd_disk_pos = -1;
2306 
2307 			disk = g_raid_create_disk(sc);
2308 			disk->d_consumer = cp;
2309 			disk->d_md_data = (void *)pd;
2310 			cp->private = disk;
2311 
2312 			g_raid_get_disk_info(disk);
2313 
2314 			memcpy(&pd->pd_disk_meta.serial[0], &serial[0],
2315 			    INTEL_SERIAL_LEN);
2316 			intel_set_disk_sectors(&pd->pd_disk_meta,
2317 			    pp->mediasize / pp->sectorsize);
2318 			pd->pd_disk_meta.id = 0;
2319 			pd->pd_disk_meta.flags = INTEL_F_SPARE;
2320 
2321 			/* Welcome the "new" disk. */
2322 			update += g_raid_md_intel_start_disk(disk);
2323 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
2324 				intel_meta_write_spare(cp, &pd->pd_disk_meta);
2325 				g_raid_destroy_disk(disk);
2326 			} else if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2327 				gctl_error(req, "Disk '%s' doesn't fit.",
2328 				    diskname);
2329 				g_raid_destroy_disk(disk);
2330 				error = -8;
2331 				break;
2332 			}
2333 		}
2334 
2335 		/* Write new metadata if we changed something. */
2336 		if (update)
2337 			g_raid_md_write_intel(md, NULL, NULL, NULL);
2338 		return (error);
2339 	}
2340 	return (-100);
2341 }
2342 
2343 static int
2344 g_raid_md_write_intel(struct g_raid_md_object *md, struct g_raid_volume *tvol,
2345     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
2346 {
2347 	struct g_raid_softc *sc;
2348 	struct g_raid_volume *vol;
2349 	struct g_raid_subdisk *sd;
2350 	struct g_raid_disk *disk;
2351 	struct g_raid_md_intel_object *mdi;
2352 	struct g_raid_md_intel_pervolume *pv;
2353 	struct g_raid_md_intel_perdisk *pd;
2354 	struct intel_raid_conf *meta;
2355 	struct intel_raid_vol *mvol;
2356 	struct intel_raid_map *mmap0, *mmap1;
2357 	off_t sectorsize = 512, pos;
2358 	const char *version, *cv;
2359 	int vi, sdi, numdisks, len, state, stale;
2360 
2361 	sc = md->mdo_softc;
2362 	mdi = (struct g_raid_md_intel_object *)md;
2363 
2364 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2365 		return (0);
2366 
2367 	/* Bump generation. Newly written metadata may differ from previous. */
2368 	mdi->mdio_generation++;
2369 
2370 	/* Count number of disks. */
2371 	numdisks = 0;
2372 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2373 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2374 		if (pd->pd_disk_pos < 0)
2375 			continue;
2376 		numdisks++;
2377 		if (disk->d_state == G_RAID_DISK_S_ACTIVE) {
2378 			pd->pd_disk_meta.flags =
2379 			    INTEL_F_ONLINE | INTEL_F_ASSIGNED;
2380 		} else if (disk->d_state == G_RAID_DISK_S_FAILED) {
2381 			pd->pd_disk_meta.flags = INTEL_F_FAILED |
2382 			    INTEL_F_ASSIGNED;
2383 		} else if (disk->d_state == G_RAID_DISK_S_DISABLED) {
2384 			pd->pd_disk_meta.flags = INTEL_F_FAILED |
2385 			    INTEL_F_ASSIGNED | INTEL_F_DISABLED;
2386 		} else {
2387 			if (!(pd->pd_disk_meta.flags & INTEL_F_DISABLED))
2388 				pd->pd_disk_meta.flags = INTEL_F_ASSIGNED;
2389 			if (pd->pd_disk_meta.id != 0xffffffff) {
2390 				pd->pd_disk_meta.id = 0xffffffff;
2391 				len = strlen(pd->pd_disk_meta.serial);
2392 				len = min(len, INTEL_SERIAL_LEN - 3);
2393 				strcpy(pd->pd_disk_meta.serial + len, ":0");
2394 			}
2395 		}
2396 	}
2397 
2398 	/* Fill anchor and disks. */
2399 	meta = malloc(INTEL_MAX_MD_SIZE(numdisks),
2400 	    M_MD_INTEL, M_WAITOK | M_ZERO);
2401 	memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1);
2402 	meta->config_size = INTEL_MAX_MD_SIZE(numdisks);
2403 	meta->config_id = mdi->mdio_config_id;
2404 	meta->orig_config_id = mdi->mdio_orig_config_id;
2405 	meta->generation = mdi->mdio_generation;
2406 	meta->attributes = INTEL_ATTR_CHECKSUM;
2407 	meta->total_disks = numdisks;
2408 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2409 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2410 		if (pd->pd_disk_pos < 0)
2411 			continue;
2412 		meta->disk[pd->pd_disk_pos] = pd->pd_disk_meta;
2413 		if (pd->pd_disk_meta.sectors_hi != 0)
2414 			meta->attributes |= INTEL_ATTR_2TB_DISK;
2415 	}
2416 
2417 	/* Fill volumes and maps. */
2418 	vi = 0;
2419 	version = INTEL_VERSION_1000;
2420 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2421 		pv = vol->v_md_data;
2422 		if (vol->v_stopping)
2423 			continue;
2424 		mvol = intel_get_volume(meta, vi);
2425 
2426 		/* New metadata may have different volumes order. */
2427 		pv->pv_volume_pos = vi;
2428 
2429 		for (sdi = 0; sdi < vol->v_disks_count; sdi++) {
2430 			sd = &vol->v_subdisks[sdi];
2431 			if (sd->sd_disk != NULL)
2432 				break;
2433 		}
2434 		if (sdi >= vol->v_disks_count)
2435 			panic("No any filled subdisk in volume");
2436 		if (vol->v_mediasize >= 0x20000000000llu)
2437 			meta->attributes |= INTEL_ATTR_2TB;
2438 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
2439 			meta->attributes |= INTEL_ATTR_RAID0;
2440 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
2441 			meta->attributes |= INTEL_ATTR_RAID1;
2442 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
2443 			meta->attributes |= INTEL_ATTR_RAID5;
2444 		else if ((vol->v_disks_count & 1) == 0)
2445 			meta->attributes |= INTEL_ATTR_RAID10;
2446 		else
2447 			meta->attributes |= INTEL_ATTR_RAID1E;
2448 		if (pv->pv_cng)
2449 			meta->attributes |= INTEL_ATTR_RAIDCNG;
2450 		if (vol->v_strip_size > 131072)
2451 			meta->attributes |= INTEL_ATTR_EXT_STRIP;
2452 
2453 		if (pv->pv_cng)
2454 			cv = INTEL_VERSION_1206;
2455 		else if (vol->v_disks_count > 4)
2456 			cv = INTEL_VERSION_1204;
2457 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
2458 			cv = INTEL_VERSION_1202;
2459 		else if (vol->v_disks_count > 2)
2460 			cv = INTEL_VERSION_1201;
2461 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
2462 			cv = INTEL_VERSION_1100;
2463 		else
2464 			cv = INTEL_VERSION_1000;
2465 		if (strcmp(cv, version) > 0)
2466 			version = cv;
2467 
2468 		strlcpy(&mvol->name[0], vol->v_name, sizeof(mvol->name));
2469 		mvol->total_sectors = vol->v_mediasize / sectorsize;
2470 		mvol->state = (INTEL_ST_READ_COALESCING |
2471 		    INTEL_ST_WRITE_COALESCING);
2472 		mvol->tid = vol->v_global_id + 1;
2473 		if (pv->pv_cng) {
2474 			mvol->state |= INTEL_ST_CLONE_N_GO;
2475 			if (pv->pv_cng_man_sync)
2476 				mvol->state |= INTEL_ST_CLONE_MAN_SYNC;
2477 			mvol->cng_master_disk = pv->pv_cng_master_disk;
2478 			if (vol->v_subdisks[pv->pv_cng_master_disk].sd_state ==
2479 			    G_RAID_SUBDISK_S_NONE)
2480 				mvol->cng_state = INTEL_CNGST_MASTER_MISSING;
2481 			else if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL)
2482 				mvol->cng_state = INTEL_CNGST_NEEDS_UPDATE;
2483 			else
2484 				mvol->cng_state = INTEL_CNGST_UPDATED;
2485 		}
2486 
2487 		/* Check for any recovery in progress. */
2488 		state = G_RAID_SUBDISK_S_ACTIVE;
2489 		pos = 0x7fffffffffffffffllu;
2490 		stale = 0;
2491 		for (sdi = 0; sdi < vol->v_disks_count; sdi++) {
2492 			sd = &vol->v_subdisks[sdi];
2493 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD)
2494 				state = G_RAID_SUBDISK_S_REBUILD;
2495 			else if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC &&
2496 			    state != G_RAID_SUBDISK_S_REBUILD)
2497 				state = G_RAID_SUBDISK_S_RESYNC;
2498 			else if (sd->sd_state == G_RAID_SUBDISK_S_STALE)
2499 				stale = 1;
2500 			if ((sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2501 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) &&
2502 			     sd->sd_rebuild_pos < pos)
2503 			        pos = sd->sd_rebuild_pos;
2504 		}
2505 		if (state == G_RAID_SUBDISK_S_REBUILD) {
2506 			mvol->migr_state = 1;
2507 			mvol->migr_type = INTEL_MT_REBUILD;
2508 		} else if (state == G_RAID_SUBDISK_S_RESYNC) {
2509 			mvol->migr_state = 1;
2510 			/* mvol->migr_type = INTEL_MT_REPAIR; */
2511 			mvol->migr_type = INTEL_MT_VERIFY;
2512 			mvol->state |= INTEL_ST_VERIFY_AND_FIX;
2513 		} else
2514 			mvol->migr_state = 0;
2515 		mvol->dirty = (vol->v_dirty || stale);
2516 
2517 		mmap0 = intel_get_map(mvol, 0);
2518 
2519 		/* Write map / common part of two maps. */
2520 		intel_set_map_offset(mmap0, sd->sd_offset / sectorsize);
2521 		intel_set_map_disk_sectors(mmap0, sd->sd_size / sectorsize);
2522 		mmap0->strip_sectors = vol->v_strip_size / sectorsize;
2523 		if (vol->v_state == G_RAID_VOLUME_S_BROKEN)
2524 			mmap0->status = INTEL_S_FAILURE;
2525 		else if (vol->v_state == G_RAID_VOLUME_S_DEGRADED)
2526 			mmap0->status = INTEL_S_DEGRADED;
2527 		else if (g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED)
2528 		    == g_raid_nsubdisks(vol, -1))
2529 			mmap0->status = INTEL_S_UNINITIALIZED;
2530 		else
2531 			mmap0->status = INTEL_S_READY;
2532 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
2533 			mmap0->type = INTEL_T_RAID0;
2534 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
2535 		    vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
2536 			mmap0->type = INTEL_T_RAID1;
2537 		else
2538 			mmap0->type = INTEL_T_RAID5;
2539 		mmap0->total_disks = vol->v_disks_count;
2540 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
2541 			mmap0->total_domains = vol->v_disks_count;
2542 		else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
2543 			mmap0->total_domains = 2;
2544 		else
2545 			mmap0->total_domains = 1;
2546 		intel_set_map_stripe_count(mmap0,
2547 		    sd->sd_size / vol->v_strip_size / mmap0->total_domains);
2548 		mmap0->failed_disk_num = 0xff;
2549 		mmap0->ddf = 1;
2550 
2551 		/* If there are two maps - copy common and update. */
2552 		if (mvol->migr_state) {
2553 			intel_set_vol_curr_migr_unit(mvol,
2554 			    pos / vol->v_strip_size / mmap0->total_domains);
2555 			mmap1 = intel_get_map(mvol, 1);
2556 			memcpy(mmap1, mmap0, sizeof(struct intel_raid_map));
2557 			mmap0->status = INTEL_S_READY;
2558 		} else
2559 			mmap1 = NULL;
2560 
2561 		/* Write disk indexes and put rebuild flags. */
2562 		for (sdi = 0; sdi < vol->v_disks_count; sdi++) {
2563 			sd = &vol->v_subdisks[sdi];
2564 			pd = (struct g_raid_md_intel_perdisk *)
2565 			    sd->sd_disk->d_md_data;
2566 			mmap0->disk_idx[sdi] = pd->pd_disk_pos;
2567 			if (mvol->migr_state)
2568 				mmap1->disk_idx[sdi] = pd->pd_disk_pos;
2569 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2570 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2571 				mmap1->disk_idx[sdi] |= INTEL_DI_RBLD;
2572 			} else if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE &&
2573 			    sd->sd_state != G_RAID_SUBDISK_S_STALE &&
2574 			    sd->sd_state != G_RAID_SUBDISK_S_UNINITIALIZED) {
2575 				mmap0->disk_idx[sdi] |= INTEL_DI_RBLD;
2576 				if (mvol->migr_state)
2577 					mmap1->disk_idx[sdi] |= INTEL_DI_RBLD;
2578 			}
2579 			if ((sd->sd_state == G_RAID_SUBDISK_S_NONE ||
2580 			     sd->sd_state == G_RAID_SUBDISK_S_FAILED ||
2581 			     sd->sd_state == G_RAID_SUBDISK_S_REBUILD) &&
2582 			    mmap0->failed_disk_num == 0xff) {
2583 				mmap0->failed_disk_num = sdi;
2584 				if (mvol->migr_state)
2585 					mmap1->failed_disk_num = sdi;
2586 			}
2587 		}
2588 		vi++;
2589 	}
2590 	meta->total_volumes = vi;
2591 	if (vi > 1 || meta->attributes &
2592 	     (INTEL_ATTR_EXT_STRIP | INTEL_ATTR_2TB_DISK | INTEL_ATTR_2TB))
2593 		version = INTEL_VERSION_1300;
2594 	if (strcmp(version, INTEL_VERSION_1300) < 0)
2595 		meta->attributes &= INTEL_ATTR_CHECKSUM;
2596 	memcpy(&meta->version[0], version, sizeof(INTEL_VERSION_1000) - 1);
2597 
2598 	/* We are done. Print meta data and store them to disks. */
2599 	g_raid_md_intel_print(meta);
2600 	if (mdi->mdio_meta != NULL)
2601 		free(mdi->mdio_meta, M_MD_INTEL);
2602 	mdi->mdio_meta = meta;
2603 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
2604 		pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2605 		if (disk->d_state != G_RAID_DISK_S_ACTIVE)
2606 			continue;
2607 		if (pd->pd_meta != NULL) {
2608 			free(pd->pd_meta, M_MD_INTEL);
2609 			pd->pd_meta = NULL;
2610 		}
2611 		pd->pd_meta = intel_meta_copy(meta);
2612 		intel_meta_write(disk->d_consumer, meta);
2613 	}
2614 	return (0);
2615 }
2616 
2617 static int
2618 g_raid_md_fail_disk_intel(struct g_raid_md_object *md,
2619     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
2620 {
2621 	struct g_raid_softc *sc;
2622 	struct g_raid_md_intel_object *mdi;
2623 	struct g_raid_md_intel_perdisk *pd;
2624 	struct g_raid_subdisk *sd;
2625 
2626 	sc = md->mdo_softc;
2627 	mdi = (struct g_raid_md_intel_object *)md;
2628 	pd = (struct g_raid_md_intel_perdisk *)tdisk->d_md_data;
2629 
2630 	/* We can't fail disk that is not a part of array now. */
2631 	if (pd->pd_disk_pos < 0)
2632 		return (-1);
2633 
2634 	/*
2635 	 * Mark disk as failed in metadata and try to write that metadata
2636 	 * to the disk itself to prevent it's later resurrection as STALE.
2637 	 */
2638 	mdi->mdio_meta->disk[pd->pd_disk_pos].flags = INTEL_F_FAILED;
2639 	pd->pd_disk_meta.flags = INTEL_F_FAILED;
2640 	g_raid_md_intel_print(mdi->mdio_meta);
2641 	if (tdisk->d_consumer)
2642 		intel_meta_write(tdisk->d_consumer, mdi->mdio_meta);
2643 
2644 	/* Change states. */
2645 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
2646 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
2647 		g_raid_change_subdisk_state(sd,
2648 		    G_RAID_SUBDISK_S_FAILED);
2649 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
2650 		    G_RAID_EVENT_SUBDISK);
2651 	}
2652 
2653 	/* Write updated metadata to remaining disks. */
2654 	g_raid_md_write_intel(md, NULL, NULL, tdisk);
2655 
2656 	/* Check if anything left except placeholders. */
2657 	if (g_raid_ndisks(sc, -1) ==
2658 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
2659 		g_raid_destroy_node(sc, 0);
2660 	else
2661 		g_raid_md_intel_refill(sc);
2662 	return (0);
2663 }
2664 
2665 static int
2666 g_raid_md_free_disk_intel(struct g_raid_md_object *md,
2667     struct g_raid_disk *disk)
2668 {
2669 	struct g_raid_md_intel_perdisk *pd;
2670 
2671 	pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data;
2672 	if (pd->pd_meta != NULL) {
2673 		free(pd->pd_meta, M_MD_INTEL);
2674 		pd->pd_meta = NULL;
2675 	}
2676 	free(pd, M_MD_INTEL);
2677 	disk->d_md_data = NULL;
2678 	return (0);
2679 }
2680 
2681 static int
2682 g_raid_md_free_volume_intel(struct g_raid_md_object *md,
2683     struct g_raid_volume *vol)
2684 {
2685 	struct g_raid_md_intel_pervolume *pv;
2686 
2687 	pv = (struct g_raid_md_intel_pervolume *)vol->v_md_data;
2688 	free(pv, M_MD_INTEL);
2689 	vol->v_md_data = NULL;
2690 	return (0);
2691 }
2692 
2693 static int
2694 g_raid_md_free_intel(struct g_raid_md_object *md)
2695 {
2696 	struct g_raid_md_intel_object *mdi;
2697 
2698 	mdi = (struct g_raid_md_intel_object *)md;
2699 	if (!mdi->mdio_started) {
2700 		mdi->mdio_started = 0;
2701 		callout_stop(&mdi->mdio_start_co);
2702 		G_RAID_DEBUG1(1, md->mdo_softc,
2703 		    "root_mount_rel %p", mdi->mdio_rootmount);
2704 		root_mount_rel(mdi->mdio_rootmount);
2705 		mdi->mdio_rootmount = NULL;
2706 	}
2707 	if (mdi->mdio_meta != NULL) {
2708 		free(mdi->mdio_meta, M_MD_INTEL);
2709 		mdi->mdio_meta = NULL;
2710 	}
2711 	return (0);
2712 }
2713 
2714 G_RAID_MD_DECLARE(intel, "Intel");
2715