xref: /freebsd/sys/geom/raid/md_nvidia.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
5  * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/bio.h>
35 #include <sys/endian.h>
36 #include <sys/kernel.h>
37 #include <sys/kobj.h>
38 #include <sys/limits.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/mutex.h>
42 #include <sys/systm.h>
43 #include <sys/taskqueue.h>
44 #include <geom/geom.h>
45 #include <geom/geom_dbg.h>
46 #include "geom/raid/g_raid.h"
47 #include "g_raid_md_if.h"
48 
49 static MALLOC_DEFINE(M_MD_NVIDIA, "md_nvidia_data", "GEOM_RAID NVIDIA metadata");
50 
51 struct nvidia_raid_conf {
52 	uint8_t		nvidia_id[8];
53 #define NVIDIA_MAGIC                "NVIDIA  "
54 
55 	uint32_t	config_size;
56 	uint32_t	checksum;
57 	uint16_t	version;
58 	uint8_t		disk_number;
59 	uint8_t		dummy_0;
60 	uint32_t	total_sectors;
61 	uint32_t	sector_size;
62 	uint8_t		name[16];
63 	uint8_t		revision[4];
64 	uint32_t	disk_status;
65 
66 	uint32_t	magic_0;
67 #define NVIDIA_MAGIC0		0x00640044
68 
69 	uint64_t	volume_id[2];
70 	uint8_t		state;
71 #define NVIDIA_S_IDLE		0
72 #define NVIDIA_S_INIT		2
73 #define NVIDIA_S_REBUILD	3
74 #define NVIDIA_S_UPGRADE	4
75 #define NVIDIA_S_SYNC		5
76 	uint8_t		array_width;
77 	uint8_t		total_disks;
78 	uint8_t		orig_array_width;
79 	uint16_t	type;
80 #define NVIDIA_T_RAID0		0x0080
81 #define NVIDIA_T_RAID1		0x0081
82 #define NVIDIA_T_RAID3		0x0083
83 #define NVIDIA_T_RAID5		0x0085	/* RLQ = 00/02? */
84 #define NVIDIA_T_RAID5_SYM	0x0095	/* RLQ = 03 */
85 #define NVIDIA_T_RAID10		0x008a
86 #define NVIDIA_T_RAID01		0x8180
87 #define NVIDIA_T_CONCAT		0x00ff
88 
89 	uint16_t	dummy_3;
90 	uint32_t	strip_sectors;
91 	uint32_t	strip_bytes;
92 	uint32_t	strip_shift;
93 	uint32_t	strip_mask;
94 	uint32_t	stripe_sectors;
95 	uint32_t	stripe_bytes;
96 	uint32_t	rebuild_lba;
97 	uint32_t	orig_type;
98 	uint32_t	orig_total_sectors;
99 	uint32_t	status;
100 #define NVIDIA_S_BOOTABLE	0x00000001
101 #define NVIDIA_S_DEGRADED	0x00000002
102 
103 	uint32_t	filler[98];
104 } __packed;
105 
106 struct g_raid_md_nvidia_perdisk {
107 	struct nvidia_raid_conf	*pd_meta;
108 	int			 pd_disk_pos;
109 	off_t			 pd_disk_size;
110 };
111 
112 struct g_raid_md_nvidia_object {
113 	struct g_raid_md_object	 mdio_base;
114 	uint64_t		 mdio_volume_id[2];
115 	struct nvidia_raid_conf	*mdio_meta;
116 	struct callout		 mdio_start_co;	/* STARTING state timer. */
117 	int			 mdio_total_disks;
118 	int			 mdio_disks_present;
119 	int			 mdio_started;
120 	int			 mdio_incomplete;
121 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
122 };
123 
124 static g_raid_md_create_t g_raid_md_create_nvidia;
125 static g_raid_md_taste_t g_raid_md_taste_nvidia;
126 static g_raid_md_event_t g_raid_md_event_nvidia;
127 static g_raid_md_ctl_t g_raid_md_ctl_nvidia;
128 static g_raid_md_write_t g_raid_md_write_nvidia;
129 static g_raid_md_fail_disk_t g_raid_md_fail_disk_nvidia;
130 static g_raid_md_free_disk_t g_raid_md_free_disk_nvidia;
131 static g_raid_md_free_t g_raid_md_free_nvidia;
132 
133 static kobj_method_t g_raid_md_nvidia_methods[] = {
134 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_nvidia),
135 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_nvidia),
136 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_nvidia),
137 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_nvidia),
138 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_nvidia),
139 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_nvidia),
140 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_nvidia),
141 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_nvidia),
142 	{ 0, 0 }
143 };
144 
145 static struct g_raid_md_class g_raid_md_nvidia_class = {
146 	"NVIDIA",
147 	g_raid_md_nvidia_methods,
148 	sizeof(struct g_raid_md_nvidia_object),
149 	.mdc_enable = 1,
150 	.mdc_priority = 100
151 };
152 
153 static int NVIDIANodeID = 1;
154 
155 static void
156 g_raid_md_nvidia_print(struct nvidia_raid_conf *meta)
157 {
158 
159 	if (g_raid_debug < 1)
160 		return;
161 
162 	printf("********* ATA NVIDIA RAID Metadata *********\n");
163 	printf("nvidia_id           <%.8s>\n", meta->nvidia_id);
164 	printf("config_size         %u\n", meta->config_size);
165 	printf("checksum            0x%08x\n", meta->checksum);
166 	printf("version             0x%04x\n", meta->version);
167 	printf("disk_number         %d\n", meta->disk_number);
168 	printf("dummy_0             0x%02x\n", meta->dummy_0);
169 	printf("total_sectors       %u\n", meta->total_sectors);
170 	printf("sector_size         %u\n", meta->sector_size);
171 	printf("name                <%.16s>\n", meta->name);
172 	printf("revision            0x%02x%02x%02x%02x\n",
173 	    meta->revision[0], meta->revision[1],
174 	    meta->revision[2], meta->revision[3]);
175 	printf("disk_status         0x%08x\n", meta->disk_status);
176 	printf("magic_0             0x%08x\n", meta->magic_0);
177 	printf("volume_id           0x%016jx%016jx\n",
178 	    meta->volume_id[1], meta->volume_id[0]);
179 	printf("state               0x%02x\n", meta->state);
180 	printf("array_width         %u\n", meta->array_width);
181 	printf("total_disks         %u\n", meta->total_disks);
182 	printf("orig_array_width    %u\n", meta->orig_array_width);
183 	printf("type                0x%04x\n", meta->type);
184 	printf("dummy_3             0x%04x\n", meta->dummy_3);
185 	printf("strip_sectors       %u\n", meta->strip_sectors);
186 	printf("strip_bytes         %u\n", meta->strip_bytes);
187 	printf("strip_shift         %u\n", meta->strip_shift);
188 	printf("strip_mask          0x%08x\n", meta->strip_mask);
189 	printf("stripe_sectors      %u\n", meta->stripe_sectors);
190 	printf("stripe_bytes        %u\n", meta->stripe_bytes);
191 	printf("rebuild_lba         %u\n", meta->rebuild_lba);
192 	printf("orig_type           0x%04x\n", meta->orig_type);
193 	printf("orig_total_sectors  %u\n", meta->orig_total_sectors);
194 	printf("status              0x%08x\n", meta->status);
195 	printf("=================================================\n");
196 }
197 
198 static struct nvidia_raid_conf *
199 nvidia_meta_copy(struct nvidia_raid_conf *meta)
200 {
201 	struct nvidia_raid_conf *nmeta;
202 
203 	nmeta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
204 	memcpy(nmeta, meta, sizeof(*meta));
205 	return (nmeta);
206 }
207 
208 static int
209 nvidia_meta_translate_disk(struct nvidia_raid_conf *meta, int md_disk_pos)
210 {
211 	int disk_pos;
212 
213 	if (md_disk_pos >= 0 && meta->type == NVIDIA_T_RAID01) {
214 		disk_pos = (md_disk_pos / meta->array_width) +
215 		    (md_disk_pos % meta->array_width) * meta->array_width;
216 	} else
217 		disk_pos = md_disk_pos;
218 	return (disk_pos);
219 }
220 
221 static void
222 nvidia_meta_get_name(struct nvidia_raid_conf *meta, char *buf)
223 {
224 	int i;
225 
226 	strncpy(buf, meta->name, 16);
227 	buf[16] = 0;
228 	for (i = 15; i >= 0; i--) {
229 		if (buf[i] > 0x20)
230 			break;
231 		buf[i] = 0;
232 	}
233 }
234 
235 static void
236 nvidia_meta_put_name(struct nvidia_raid_conf *meta, char *buf)
237 {
238 
239 	memset(meta->name, 0x20, 16);
240 	memcpy(meta->name, buf, MIN(strlen(buf), 16));
241 }
242 
243 static struct nvidia_raid_conf *
244 nvidia_meta_read(struct g_consumer *cp)
245 {
246 	struct g_provider *pp;
247 	struct nvidia_raid_conf *meta;
248 	char *buf;
249 	int error, i;
250 	uint32_t checksum, *ptr;
251 
252 	pp = cp->provider;
253 	if (pp->sectorsize < sizeof(*meta))
254 		return (NULL);
255 	/* Read the anchor sector. */
256 	buf = g_read_data(cp,
257 	    pp->mediasize - 2 * pp->sectorsize, pp->sectorsize, &error);
258 	if (buf == NULL) {
259 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
260 		    pp->name, error);
261 		return (NULL);
262 	}
263 	meta = (struct nvidia_raid_conf *)buf;
264 
265 	/* Check if this is an NVIDIA RAID struct */
266 	if (strncmp(meta->nvidia_id, NVIDIA_MAGIC, strlen(NVIDIA_MAGIC))) {
267 		G_RAID_DEBUG(1, "NVIDIA signature check failed on %s", pp->name);
268 		g_free(buf);
269 		return (NULL);
270 	}
271 	if (meta->config_size > 128 ||
272 	    meta->config_size < 30) {
273 		G_RAID_DEBUG(1, "NVIDIA metadata size looks wrong: %d",
274 		    meta->config_size);
275 		g_free(buf);
276 		return (NULL);
277 	}
278 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
279 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
280 	g_free(buf);
281 
282 	/* Check metadata checksum. */
283 	for (checksum = 0, ptr = (uint32_t *)meta,
284 	    i = 0; i < meta->config_size; i++)
285 		checksum += *ptr++;
286 	if (checksum != 0) {
287 		G_RAID_DEBUG(1, "NVIDIA checksum check failed on %s", pp->name);
288 		free(meta, M_MD_NVIDIA);
289 		return (NULL);
290 	}
291 
292 	/* Check volume state. */
293 	if (meta->state != NVIDIA_S_IDLE && meta->state != NVIDIA_S_INIT &&
294 	    meta->state != NVIDIA_S_REBUILD && meta->state != NVIDIA_S_SYNC) {
295 		G_RAID_DEBUG(1, "NVIDIA unknown state on %s (0x%02x)",
296 		    pp->name, meta->state);
297 		free(meta, M_MD_NVIDIA);
298 		return (NULL);
299 	}
300 
301 	/* Check raid type. */
302 	if (meta->type != NVIDIA_T_RAID0 && meta->type != NVIDIA_T_RAID1 &&
303 	    meta->type != NVIDIA_T_RAID3 && meta->type != NVIDIA_T_RAID5 &&
304 	    meta->type != NVIDIA_T_RAID5_SYM &&
305 	    meta->type != NVIDIA_T_RAID01 && meta->type != NVIDIA_T_CONCAT) {
306 		G_RAID_DEBUG(1, "NVIDIA unknown RAID level on %s (0x%02x)",
307 		    pp->name, meta->type);
308 		free(meta, M_MD_NVIDIA);
309 		return (NULL);
310 	}
311 
312 	return (meta);
313 }
314 
315 static int
316 nvidia_meta_write(struct g_consumer *cp, struct nvidia_raid_conf *meta)
317 {
318 	struct g_provider *pp;
319 	char *buf;
320 	int error, i;
321 	uint32_t checksum, *ptr;
322 
323 	pp = cp->provider;
324 
325 	/* Recalculate checksum for case if metadata were changed. */
326 	meta->checksum = 0;
327 	for (checksum = 0, ptr = (uint32_t *)meta,
328 	    i = 0; i < meta->config_size; i++)
329 		checksum += *ptr++;
330 	meta->checksum -= checksum;
331 
332 	/* Create and fill buffer. */
333 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
334 	memcpy(buf, meta, sizeof(*meta));
335 
336 	/* Write metadata. */
337 	error = g_write_data(cp,
338 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
339 	if (error != 0) {
340 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
341 		    pp->name, error);
342 	}
343 
344 	free(buf, M_MD_NVIDIA);
345 	return (error);
346 }
347 
348 static int
349 nvidia_meta_erase(struct g_consumer *cp)
350 {
351 	struct g_provider *pp;
352 	char *buf;
353 	int error;
354 
355 	pp = cp->provider;
356 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
357 	error = g_write_data(cp,
358 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
359 	if (error != 0) {
360 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
361 		    pp->name, error);
362 	}
363 	free(buf, M_MD_NVIDIA);
364 	return (error);
365 }
366 
367 static struct g_raid_disk *
368 g_raid_md_nvidia_get_disk(struct g_raid_softc *sc, int id)
369 {
370 	struct g_raid_disk	*disk;
371 	struct g_raid_md_nvidia_perdisk *pd;
372 
373 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
374 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
375 		if (pd->pd_disk_pos == id)
376 			break;
377 	}
378 	return (disk);
379 }
380 
381 static int
382 g_raid_md_nvidia_supported(int level, int qual, int disks, int force)
383 {
384 
385 	switch (level) {
386 	case G_RAID_VOLUME_RL_RAID0:
387 		if (disks < 1)
388 			return (0);
389 		if (!force && (disks < 2 || disks > 6))
390 			return (0);
391 		break;
392 	case G_RAID_VOLUME_RL_RAID1:
393 		if (disks < 1)
394 			return (0);
395 		if (!force && (disks != 2))
396 			return (0);
397 		break;
398 	case G_RAID_VOLUME_RL_RAID1E:
399 		if (disks < 2)
400 			return (0);
401 		if (disks % 2 != 0)
402 			return (0);
403 		if (!force && (disks < 4))
404 			return (0);
405 		break;
406 	case G_RAID_VOLUME_RL_SINGLE:
407 		if (disks != 1)
408 			return (0);
409 		break;
410 	case G_RAID_VOLUME_RL_CONCAT:
411 		if (disks < 2)
412 			return (0);
413 		break;
414 	case G_RAID_VOLUME_RL_RAID5:
415 		if (disks < 3)
416 			return (0);
417 		if (qual != G_RAID_VOLUME_RLQ_R5LA &&
418 		    qual != G_RAID_VOLUME_RLQ_R5LS)
419 			return (0);
420 		break;
421 	default:
422 		return (0);
423 	}
424 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
425 		return (0);
426 	return (1);
427 }
428 
429 static int
430 g_raid_md_nvidia_start_disk(struct g_raid_disk *disk)
431 {
432 	struct g_raid_softc *sc;
433 	struct g_raid_subdisk *sd, *tmpsd;
434 	struct g_raid_disk *olddisk, *tmpdisk;
435 	struct g_raid_md_object *md;
436 	struct g_raid_md_nvidia_object *mdi;
437 	struct g_raid_md_nvidia_perdisk *pd, *oldpd;
438 	struct nvidia_raid_conf *meta;
439 	int disk_pos, resurrection = 0;
440 
441 	sc = disk->d_softc;
442 	md = sc->sc_md;
443 	mdi = (struct g_raid_md_nvidia_object *)md;
444 	meta = mdi->mdio_meta;
445 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
446 	olddisk = NULL;
447 
448 	/* Find disk position in metadata by its serial. */
449 	if (pd->pd_meta != NULL) {
450 		disk_pos = pd->pd_meta->disk_number;
451 		if (disk_pos >= meta->total_disks || mdi->mdio_started)
452 			disk_pos = -3;
453 	} else
454 		disk_pos = -3;
455 	/* For RAID0+1 we need to translate order. */
456 	disk_pos = nvidia_meta_translate_disk(meta, disk_pos);
457 	if (disk_pos < 0) {
458 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
459 		/* If we are in the start process, that's all for now. */
460 		if (!mdi->mdio_started)
461 			goto nofit;
462 		/*
463 		 * If we have already started - try to get use of the disk.
464 		 * Try to replace OFFLINE disks first, then FAILED.
465 		 */
466 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
467 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
468 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
469 				continue;
470 			/* Make sure this disk is big enough. */
471 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
472 				if (sd->sd_offset + sd->sd_size + 2 * 512 >
473 				    pd->pd_disk_size) {
474 					G_RAID_DEBUG1(1, sc,
475 					    "Disk too small (%ju < %ju)",
476 					    pd->pd_disk_size,
477 					    sd->sd_offset + sd->sd_size + 512);
478 					break;
479 				}
480 			}
481 			if (sd != NULL)
482 				continue;
483 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
484 				olddisk = tmpdisk;
485 				break;
486 			} else if (olddisk == NULL)
487 				olddisk = tmpdisk;
488 		}
489 		if (olddisk == NULL) {
490 nofit:
491 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
492 			return (1);
493 		}
494 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
495 		disk_pos = oldpd->pd_disk_pos;
496 		resurrection = 1;
497 	}
498 
499 	if (olddisk == NULL) {
500 		/* Find placeholder by position. */
501 		olddisk = g_raid_md_nvidia_get_disk(sc, disk_pos);
502 		if (olddisk == NULL)
503 			panic("No disk at position %d!", disk_pos);
504 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
505 			G_RAID_DEBUG1(1, sc, "More than one disk for pos %d",
506 			    disk_pos);
507 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
508 			return (0);
509 		}
510 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
511 	}
512 
513 	/* Replace failed disk or placeholder with new disk. */
514 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
515 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
516 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
517 		sd->sd_disk = disk;
518 	}
519 	oldpd->pd_disk_pos = -2;
520 	pd->pd_disk_pos = disk_pos;
521 
522 	/* If it was placeholder -- destroy it. */
523 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
524 		g_raid_destroy_disk(olddisk);
525 	} else {
526 		/* Otherwise, make it STALE_FAILED. */
527 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
528 	}
529 
530 	/* Welcome the new disk. */
531 	if (resurrection)
532 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
533 	else// if (pd->pd_meta->disk_status == NVIDIA_S_CURRENT ||
534 	    //pd->pd_meta->disk_status == NVIDIA_S_REBUILD)
535 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
536 //	else
537 //		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
538 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
539 		/*
540 		 * Different disks may have different sizes,
541 		 * in concat mode. Update from real disk size.
542 		 */
543 		if (meta->type == NVIDIA_T_CONCAT)
544 			sd->sd_size = pd->pd_disk_size - 0x800 * 512;
545 
546 		if (resurrection) {
547 			/* New or ex-spare disk. */
548 			g_raid_change_subdisk_state(sd,
549 			    G_RAID_SUBDISK_S_NEW);
550 		} else if (meta->state == NVIDIA_S_REBUILD &&
551 		    (pd->pd_meta->disk_status & 0x100)) {
552 			/* Rebuilding disk. */
553 			g_raid_change_subdisk_state(sd,
554 			    G_RAID_SUBDISK_S_REBUILD);
555 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
556 			    meta->array_width * pd->pd_meta->sector_size;
557 		} else if (meta->state == NVIDIA_S_SYNC) {
558 			/* Resyncing/dirty disk. */
559 			g_raid_change_subdisk_state(sd,
560 			    G_RAID_SUBDISK_S_RESYNC);
561 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
562 			    meta->array_width * pd->pd_meta->sector_size;
563 		} else {
564 			/* Up to date disk. */
565 			g_raid_change_subdisk_state(sd,
566 			    G_RAID_SUBDISK_S_ACTIVE);
567 		}
568 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
569 		    G_RAID_EVENT_SUBDISK);
570 	}
571 
572 	/* Update status of our need for spare. */
573 	if (mdi->mdio_started) {
574 		mdi->mdio_incomplete =
575 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
576 		     mdi->mdio_total_disks);
577 	}
578 
579 	return (resurrection);
580 }
581 
582 static void
583 g_disk_md_nvidia_retaste(void *arg, int pending)
584 {
585 
586 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
587 	g_retaste(&g_raid_class);
588 	free(arg, M_MD_NVIDIA);
589 }
590 
591 static void
592 g_raid_md_nvidia_refill(struct g_raid_softc *sc)
593 {
594 	struct g_raid_md_object *md;
595 	struct g_raid_md_nvidia_object *mdi;
596 	struct g_raid_disk *disk;
597 	struct task *task;
598 	int update, na;
599 
600 	md = sc->sc_md;
601 	mdi = (struct g_raid_md_nvidia_object *)md;
602 	update = 0;
603 	do {
604 		/* Make sure we miss anything. */
605 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
606 		if (na == mdi->mdio_total_disks)
607 			break;
608 
609 		G_RAID_DEBUG1(1, md->mdo_softc,
610 		    "Array is not complete (%d of %d), "
611 		    "trying to refill.", na, mdi->mdio_total_disks);
612 
613 		/* Try to get use some of STALE disks. */
614 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
615 			if (disk->d_state == G_RAID_DISK_S_STALE) {
616 				update += g_raid_md_nvidia_start_disk(disk);
617 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
618 					break;
619 			}
620 		}
621 		if (disk != NULL)
622 			continue;
623 
624 		/* Try to get use some of SPARE disks. */
625 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
626 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
627 				update += g_raid_md_nvidia_start_disk(disk);
628 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
629 					break;
630 			}
631 		}
632 	} while (disk != NULL);
633 
634 	/* Write new metadata if we changed something. */
635 	if (update)
636 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
637 
638 	/* Update status of our need for spare. */
639 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
640 	    mdi->mdio_total_disks);
641 
642 	/* Request retaste hoping to find spare. */
643 	if (mdi->mdio_incomplete) {
644 		task = malloc(sizeof(struct task),
645 		    M_MD_NVIDIA, M_WAITOK | M_ZERO);
646 		TASK_INIT(task, 0, g_disk_md_nvidia_retaste, task);
647 		taskqueue_enqueue(taskqueue_swi, task);
648 	}
649 }
650 
651 static void
652 g_raid_md_nvidia_start(struct g_raid_softc *sc)
653 {
654 	struct g_raid_md_object *md;
655 	struct g_raid_md_nvidia_object *mdi;
656 	struct g_raid_md_nvidia_perdisk *pd;
657 	struct nvidia_raid_conf *meta;
658 	struct g_raid_volume *vol;
659 	struct g_raid_subdisk *sd;
660 	struct g_raid_disk *disk;
661 	off_t size;
662 	int j, disk_pos;
663 	char buf[17];
664 
665 	md = sc->sc_md;
666 	mdi = (struct g_raid_md_nvidia_object *)md;
667 	meta = mdi->mdio_meta;
668 
669 	/* Create volumes and subdisks. */
670 	nvidia_meta_get_name(meta, buf);
671 	vol = g_raid_create_volume(sc, buf, -1);
672 	vol->v_mediasize = (off_t)meta->total_sectors * 512;
673 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
674 	if (meta->type == NVIDIA_T_RAID0) {
675 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
676 		size = vol->v_mediasize / mdi->mdio_total_disks;
677 	} else if (meta->type == NVIDIA_T_RAID1) {
678 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
679 		size = vol->v_mediasize;
680 	} else if (meta->type == NVIDIA_T_RAID01) {
681 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
682 		size = vol->v_mediasize / (mdi->mdio_total_disks / 2);
683 	} else if (meta->type == NVIDIA_T_CONCAT) {
684 		if (mdi->mdio_total_disks == 1)
685 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
686 		else
687 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
688 		size = 0;
689 	} else if (meta->type == NVIDIA_T_RAID5) {
690 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
691 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
692 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
693 	} else if (meta->type == NVIDIA_T_RAID5_SYM) {
694 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
695 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LS;
696 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
697 	} else {
698 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
699 		size = 0;
700 	}
701 	vol->v_strip_size = meta->strip_sectors * 512; //ZZZ
702 	vol->v_disks_count = mdi->mdio_total_disks;
703 	vol->v_sectorsize = 512; //ZZZ
704 	for (j = 0; j < vol->v_disks_count; j++) {
705 		sd = &vol->v_subdisks[j];
706 		sd->sd_offset = 0;
707 		sd->sd_size = size;
708 	}
709 	g_raid_start_volume(vol);
710 
711 	/* Create disk placeholders to store data for later writing. */
712 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
713 		pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
714 		pd->pd_disk_pos = disk_pos;
715 		disk = g_raid_create_disk(sc);
716 		disk->d_md_data = (void *)pd;
717 		disk->d_state = G_RAID_DISK_S_OFFLINE;
718 		sd = &vol->v_subdisks[disk_pos];
719 		sd->sd_disk = disk;
720 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
721 	}
722 
723 	/* Make all disks found till the moment take their places. */
724 	do {
725 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
726 			if (disk->d_state == G_RAID_DISK_S_NONE) {
727 				g_raid_md_nvidia_start_disk(disk);
728 				break;
729 			}
730 		}
731 	} while (disk != NULL);
732 
733 	mdi->mdio_started = 1;
734 	G_RAID_DEBUG1(0, sc, "Array started.");
735 	g_raid_md_write_nvidia(md, NULL, NULL, NULL);
736 
737 	/* Pickup any STALE/SPARE disks to refill array if needed. */
738 	g_raid_md_nvidia_refill(sc);
739 
740 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
741 
742 	callout_stop(&mdi->mdio_start_co);
743 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
744 	root_mount_rel(mdi->mdio_rootmount);
745 	mdi->mdio_rootmount = NULL;
746 }
747 
748 static void
749 g_raid_md_nvidia_new_disk(struct g_raid_disk *disk)
750 {
751 	struct g_raid_softc *sc;
752 	struct g_raid_md_object *md;
753 	struct g_raid_md_nvidia_object *mdi;
754 	struct nvidia_raid_conf *pdmeta;
755 	struct g_raid_md_nvidia_perdisk *pd;
756 
757 	sc = disk->d_softc;
758 	md = sc->sc_md;
759 	mdi = (struct g_raid_md_nvidia_object *)md;
760 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
761 	pdmeta = pd->pd_meta;
762 
763 	if (mdi->mdio_started) {
764 		if (g_raid_md_nvidia_start_disk(disk))
765 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
766 	} else {
767 		if (mdi->mdio_meta == NULL ||
768 		    mdi->mdio_meta->disk_number >= mdi->mdio_meta->total_disks) {
769 			G_RAID_DEBUG1(1, sc, "Newer disk");
770 			if (mdi->mdio_meta != NULL)
771 				free(mdi->mdio_meta, M_MD_NVIDIA);
772 			mdi->mdio_meta = nvidia_meta_copy(pdmeta);
773 			mdi->mdio_total_disks = pdmeta->total_disks;
774 			mdi->mdio_disks_present = 1;
775 		} else if (pdmeta->disk_number < mdi->mdio_meta->total_disks) {
776 			mdi->mdio_disks_present++;
777 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
778 			    mdi->mdio_disks_present,
779 			    mdi->mdio_total_disks);
780 		} else
781 			G_RAID_DEBUG1(1, sc, "Spare disk");
782 
783 		/* If we collected all needed disks - start array. */
784 		if (mdi->mdio_disks_present == mdi->mdio_total_disks)
785 			g_raid_md_nvidia_start(sc);
786 	}
787 }
788 
789 static void
790 g_raid_nvidia_go(void *arg)
791 {
792 	struct g_raid_softc *sc;
793 	struct g_raid_md_object *md;
794 	struct g_raid_md_nvidia_object *mdi;
795 
796 	sc = arg;
797 	md = sc->sc_md;
798 	mdi = (struct g_raid_md_nvidia_object *)md;
799 	if (!mdi->mdio_started) {
800 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
801 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
802 	}
803 }
804 
805 static int
806 g_raid_md_create_nvidia(struct g_raid_md_object *md, struct g_class *mp,
807     struct g_geom **gp)
808 {
809 	struct g_raid_softc *sc;
810 	struct g_raid_md_nvidia_object *mdi;
811 	char name[32];
812 
813 	mdi = (struct g_raid_md_nvidia_object *)md;
814 	arc4rand(&mdi->mdio_volume_id, 16, 0);
815 	snprintf(name, sizeof(name), "NVIDIA-%d",
816 	    atomic_fetchadd_int(&NVIDIANodeID, 1));
817 	sc = g_raid_create_node(mp, name, md);
818 	if (sc == NULL)
819 		return (G_RAID_MD_TASTE_FAIL);
820 	md->mdo_softc = sc;
821 	*gp = sc->sc_geom;
822 	return (G_RAID_MD_TASTE_NEW);
823 }
824 
825 static int
826 g_raid_md_taste_nvidia(struct g_raid_md_object *md, struct g_class *mp,
827                               struct g_consumer *cp, struct g_geom **gp)
828 {
829 	struct g_consumer *rcp;
830 	struct g_provider *pp;
831 	struct g_raid_md_nvidia_object *mdi, *mdi1;
832 	struct g_raid_softc *sc;
833 	struct g_raid_disk *disk;
834 	struct nvidia_raid_conf *meta;
835 	struct g_raid_md_nvidia_perdisk *pd;
836 	struct g_geom *geom;
837 	int result, spare, len;
838 	char name[32];
839 	uint16_t vendor;
840 
841 	G_RAID_DEBUG(1, "Tasting NVIDIA on %s", cp->provider->name);
842 	mdi = (struct g_raid_md_nvidia_object *)md;
843 	pp = cp->provider;
844 
845 	/* Read metadata from device. */
846 	meta = NULL;
847 	g_topology_unlock();
848 	vendor = 0xffff;
849 	len = sizeof(vendor);
850 	if (pp->geom->rank == 1)
851 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
852 	meta = nvidia_meta_read(cp);
853 	g_topology_lock();
854 	if (meta == NULL) {
855 		if (g_raid_aggressive_spare) {
856 			if (vendor == 0x10de) {
857 				G_RAID_DEBUG(1,
858 				    "No NVIDIA metadata, forcing spare.");
859 				spare = 2;
860 				goto search;
861 			} else {
862 				G_RAID_DEBUG(1,
863 				    "NVIDIA vendor mismatch 0x%04x != 0x10de",
864 				    vendor);
865 			}
866 		}
867 		return (G_RAID_MD_TASTE_FAIL);
868 	}
869 
870 	/* Metadata valid. Print it. */
871 	g_raid_md_nvidia_print(meta);
872 	G_RAID_DEBUG(1, "NVIDIA disk position %d", meta->disk_number);
873 	spare = 0;//(meta->type == NVIDIA_T_SPARE) ? 1 : 0;
874 
875 search:
876 	/* Search for matching node. */
877 	sc = NULL;
878 	mdi1 = NULL;
879 	LIST_FOREACH(geom, &mp->geom, geom) {
880 		sc = geom->softc;
881 		if (sc == NULL)
882 			continue;
883 		if (sc->sc_stopping != 0)
884 			continue;
885 		if (sc->sc_md->mdo_class != md->mdo_class)
886 			continue;
887 		mdi1 = (struct g_raid_md_nvidia_object *)sc->sc_md;
888 		if (spare) {
889 			if (mdi1->mdio_incomplete)
890 				break;
891 		} else {
892 			if (memcmp(&mdi1->mdio_volume_id,
893 			     &meta->volume_id, 16) == 0)
894 				break;
895 		}
896 	}
897 
898 	/* Found matching node. */
899 	if (geom != NULL) {
900 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
901 		result = G_RAID_MD_TASTE_EXISTING;
902 
903 	} else if (spare) { /* Not found needy node -- left for later. */
904 		G_RAID_DEBUG(1, "Spare is not needed at this time");
905 		goto fail1;
906 
907 	} else { /* Not found matching node -- create one. */
908 		result = G_RAID_MD_TASTE_NEW;
909 		memcpy(&mdi->mdio_volume_id, &meta->volume_id, 16);
910 		snprintf(name, sizeof(name), "NVIDIA-%d",
911 		    atomic_fetchadd_int(&NVIDIANodeID, 1));
912 		sc = g_raid_create_node(mp, name, md);
913 		md->mdo_softc = sc;
914 		geom = sc->sc_geom;
915 		callout_init(&mdi->mdio_start_co, 1);
916 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
917 		    g_raid_nvidia_go, sc);
918 		mdi->mdio_rootmount = root_mount_hold("GRAID-NVIDIA");
919 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
920 	}
921 
922 	/* There is no return after this point, so we close passed consumer. */
923 	g_access(cp, -1, 0, 0);
924 
925 	rcp = g_new_consumer(geom);
926 	rcp->flags |= G_CF_DIRECT_RECEIVE;
927 	g_attach(rcp, pp);
928 	if (g_access(rcp, 1, 1, 1) != 0)
929 		; //goto fail1;
930 
931 	g_topology_unlock();
932 	sx_xlock(&sc->sc_lock);
933 
934 	pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
935 	pd->pd_meta = meta;
936 	if (spare == 2) {
937 		pd->pd_disk_pos = -3;
938 	} else {
939 		pd->pd_disk_pos = -1;
940 	}
941 	pd->pd_disk_size = pp->mediasize;
942 	disk = g_raid_create_disk(sc);
943 	disk->d_md_data = (void *)pd;
944 	disk->d_consumer = rcp;
945 	rcp->private = disk;
946 
947 	g_raid_get_disk_info(disk);
948 
949 	g_raid_md_nvidia_new_disk(disk);
950 
951 	sx_xunlock(&sc->sc_lock);
952 	g_topology_lock();
953 	*gp = geom;
954 	return (result);
955 fail1:
956 	free(meta, M_MD_NVIDIA);
957 	return (G_RAID_MD_TASTE_FAIL);
958 }
959 
960 static int
961 g_raid_md_event_nvidia(struct g_raid_md_object *md,
962     struct g_raid_disk *disk, u_int event)
963 {
964 	struct g_raid_softc *sc;
965 	struct g_raid_subdisk *sd;
966 	struct g_raid_md_nvidia_object *mdi;
967 	struct g_raid_md_nvidia_perdisk *pd;
968 
969 	sc = md->mdo_softc;
970 	mdi = (struct g_raid_md_nvidia_object *)md;
971 	if (disk == NULL) {
972 		switch (event) {
973 		case G_RAID_NODE_E_START:
974 			if (!mdi->mdio_started) {
975 				/* Bump volume ID to drop missing disks. */
976 				arc4rand(&mdi->mdio_volume_id, 16, 0);
977 				g_raid_md_nvidia_start(sc);
978 			}
979 			return (0);
980 		}
981 		return (-1);
982 	}
983 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
984 	switch (event) {
985 	case G_RAID_DISK_E_DISCONNECTED:
986 		/* If disk was assigned, just update statuses. */
987 		if (pd->pd_disk_pos >= 0) {
988 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
989 			if (disk->d_consumer) {
990 				g_raid_kill_consumer(sc, disk->d_consumer);
991 				disk->d_consumer = NULL;
992 			}
993 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
994 				g_raid_change_subdisk_state(sd,
995 				    G_RAID_SUBDISK_S_NONE);
996 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
997 				    G_RAID_EVENT_SUBDISK);
998 			}
999 		} else {
1000 			/* Otherwise -- delete. */
1001 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1002 			g_raid_destroy_disk(disk);
1003 		}
1004 
1005 		if (mdi->mdio_started) {
1006 			/* Bump volume ID to prevent disk resurrection. */
1007 			if (pd->pd_disk_pos >= 0)
1008 				arc4rand(&mdi->mdio_volume_id, 16, 0);
1009 
1010 			/* Write updated metadata to all disks. */
1011 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1012 		}
1013 
1014 		/* Check if anything left except placeholders. */
1015 		if (g_raid_ndisks(sc, -1) ==
1016 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1017 			g_raid_destroy_node(sc, 0);
1018 		else
1019 			g_raid_md_nvidia_refill(sc);
1020 		return (0);
1021 	}
1022 	return (-2);
1023 }
1024 
1025 static int
1026 g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
1027     struct gctl_req *req)
1028 {
1029 	struct g_raid_softc *sc;
1030 	struct g_raid_volume *vol;
1031 	struct g_raid_subdisk *sd;
1032 	struct g_raid_disk *disk;
1033 	struct g_raid_md_nvidia_object *mdi;
1034 	struct g_raid_md_nvidia_perdisk *pd;
1035 	struct g_consumer *cp;
1036 	struct g_provider *pp;
1037 	char arg[16];
1038 	const char *verb, *volname, *levelname, *diskname;
1039 	int *nargs, *force;
1040 	off_t size, sectorsize, strip, volsize;
1041 	intmax_t *sizearg, *striparg;
1042 	int numdisks, i, len, level, qual, update;
1043 	int error;
1044 
1045 	sc = md->mdo_softc;
1046 	mdi = (struct g_raid_md_nvidia_object *)md;
1047 	verb = gctl_get_param(req, "verb", NULL);
1048 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1049 	error = 0;
1050 	if (strcmp(verb, "label") == 0) {
1051 		if (*nargs < 4) {
1052 			gctl_error(req, "Invalid number of arguments.");
1053 			return (-1);
1054 		}
1055 		volname = gctl_get_asciiparam(req, "arg1");
1056 		if (volname == NULL) {
1057 			gctl_error(req, "No volume name.");
1058 			return (-2);
1059 		}
1060 		levelname = gctl_get_asciiparam(req, "arg2");
1061 		if (levelname == NULL) {
1062 			gctl_error(req, "No RAID level.");
1063 			return (-3);
1064 		}
1065 		if (strcasecmp(levelname, "RAID5") == 0)
1066 			levelname = "RAID5-LS";
1067 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1068 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1069 			return (-4);
1070 		}
1071 		numdisks = *nargs - 3;
1072 		force = gctl_get_paraml(req, "force", sizeof(*force));
1073 		if (!g_raid_md_nvidia_supported(level, qual, numdisks,
1074 		    force ? *force : 0)) {
1075 			gctl_error(req, "Unsupported RAID level "
1076 			    "(0x%02x/0x%02x), or number of disks (%d).",
1077 			    level, qual, numdisks);
1078 			return (-5);
1079 		}
1080 
1081 		/* Search for disks, connect them and probe. */
1082 		size = 0x7fffffffffffffffllu;
1083 		sectorsize = 0;
1084 		for (i = 0; i < numdisks; i++) {
1085 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1086 			diskname = gctl_get_asciiparam(req, arg);
1087 			if (diskname == NULL) {
1088 				gctl_error(req, "No disk name (%s).", arg);
1089 				error = -6;
1090 				break;
1091 			}
1092 			if (strcmp(diskname, "NONE") == 0) {
1093 				cp = NULL;
1094 				pp = NULL;
1095 			} else {
1096 				g_topology_lock();
1097 				cp = g_raid_open_consumer(sc, diskname);
1098 				if (cp == NULL) {
1099 					gctl_error(req, "Can't open '%s'.",
1100 					    diskname);
1101 					g_topology_unlock();
1102 					error = -7;
1103 					break;
1104 				}
1105 				pp = cp->provider;
1106 			}
1107 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1108 			pd->pd_disk_pos = i;
1109 			disk = g_raid_create_disk(sc);
1110 			disk->d_md_data = (void *)pd;
1111 			disk->d_consumer = cp;
1112 			if (cp == NULL)
1113 				continue;
1114 			cp->private = disk;
1115 			g_topology_unlock();
1116 
1117 			g_raid_get_disk_info(disk);
1118 
1119 			pd->pd_disk_size = pp->mediasize;
1120 			if (size > pp->mediasize)
1121 				size = pp->mediasize;
1122 			if (sectorsize < pp->sectorsize)
1123 				sectorsize = pp->sectorsize;
1124 		}
1125 		if (error != 0)
1126 			return (error);
1127 
1128 		if (sectorsize <= 0) {
1129 			gctl_error(req, "Can't get sector size.");
1130 			return (-8);
1131 		}
1132 
1133 		/* Reserve space for metadata. */
1134 		size -= 2 * sectorsize;
1135 
1136 		/* Handle size argument. */
1137 		len = sizeof(*sizearg);
1138 		sizearg = gctl_get_param(req, "size", &len);
1139 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1140 		    *sizearg > 0) {
1141 			if (*sizearg > size) {
1142 				gctl_error(req, "Size too big %lld > %lld.",
1143 				    (long long)*sizearg, (long long)size);
1144 				return (-9);
1145 			}
1146 			size = *sizearg;
1147 		}
1148 
1149 		/* Handle strip argument. */
1150 		strip = 131072;
1151 		len = sizeof(*striparg);
1152 		striparg = gctl_get_param(req, "strip", &len);
1153 		if (striparg != NULL && len == sizeof(*striparg) &&
1154 		    *striparg > 0) {
1155 			if (*striparg < sectorsize) {
1156 				gctl_error(req, "Strip size too small.");
1157 				return (-10);
1158 			}
1159 			if (*striparg % sectorsize != 0) {
1160 				gctl_error(req, "Incorrect strip size.");
1161 				return (-11);
1162 			}
1163 			if (strip > 65535 * sectorsize) {
1164 				gctl_error(req, "Strip size too big.");
1165 				return (-12);
1166 			}
1167 			strip = *striparg;
1168 		}
1169 
1170 		/* Round size down to strip or sector. */
1171 		if (level == G_RAID_VOLUME_RL_RAID1)
1172 			size -= (size % sectorsize);
1173 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1174 		    (numdisks & 1) != 0)
1175 			size -= (size % (2 * strip));
1176 		else
1177 			size -= (size % strip);
1178 		if (size <= 0) {
1179 			gctl_error(req, "Size too small.");
1180 			return (-13);
1181 		}
1182 
1183 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1184 		    level == G_RAID_VOLUME_RL_CONCAT ||
1185 		    level == G_RAID_VOLUME_RL_SINGLE)
1186 			volsize = size * numdisks;
1187 		else if (level == G_RAID_VOLUME_RL_RAID1)
1188 			volsize = size;
1189 		else if (level == G_RAID_VOLUME_RL_RAID5)
1190 			volsize = size * (numdisks - 1);
1191 		else { /* RAID1E */
1192 			volsize = ((size * numdisks) / strip / 2) *
1193 			    strip;
1194 		}
1195 		if (volsize > 0xffffffffllu * sectorsize) {
1196 			gctl_error(req, "Size too big.");
1197 			return (-14);
1198 		}
1199 
1200 		/* We have all we need, create things: volume, ... */
1201 		mdi->mdio_total_disks = numdisks;
1202 		mdi->mdio_started = 1;
1203 		vol = g_raid_create_volume(sc, volname, -1);
1204 		vol->v_md_data = (void *)(intptr_t)0;
1205 		vol->v_raid_level = level;
1206 		vol->v_raid_level_qualifier = qual;
1207 		vol->v_strip_size = strip;
1208 		vol->v_disks_count = numdisks;
1209 		vol->v_mediasize = volsize;
1210 		vol->v_sectorsize = sectorsize;
1211 		g_raid_start_volume(vol);
1212 
1213 		/* , and subdisks. */
1214 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1215 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1216 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1217 			sd->sd_disk = disk;
1218 			sd->sd_offset = 0;
1219 			sd->sd_size = size;
1220 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1221 			if (sd->sd_disk->d_consumer != NULL) {
1222 				g_raid_change_disk_state(disk,
1223 				    G_RAID_DISK_S_ACTIVE);
1224 				g_raid_change_subdisk_state(sd,
1225 				    G_RAID_SUBDISK_S_ACTIVE);
1226 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1227 				    G_RAID_EVENT_SUBDISK);
1228 			} else {
1229 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1230 			}
1231 		}
1232 
1233 		/* Write metadata based on created entities. */
1234 		G_RAID_DEBUG1(0, sc, "Array started.");
1235 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1236 
1237 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1238 		g_raid_md_nvidia_refill(sc);
1239 
1240 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1241 		    G_RAID_EVENT_VOLUME);
1242 		return (0);
1243 	}
1244 	if (strcmp(verb, "delete") == 0) {
1245 		/* Check if some volume is still open. */
1246 		force = gctl_get_paraml(req, "force", sizeof(*force));
1247 		if (force != NULL && *force == 0 &&
1248 		    g_raid_nopens(sc) != 0) {
1249 			gctl_error(req, "Some volume is still open.");
1250 			return (-4);
1251 		}
1252 
1253 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1254 			if (disk->d_consumer)
1255 				nvidia_meta_erase(disk->d_consumer);
1256 		}
1257 		g_raid_destroy_node(sc, 0);
1258 		return (0);
1259 	}
1260 	if (strcmp(verb, "remove") == 0 ||
1261 	    strcmp(verb, "fail") == 0) {
1262 		if (*nargs < 2) {
1263 			gctl_error(req, "Invalid number of arguments.");
1264 			return (-1);
1265 		}
1266 		for (i = 1; i < *nargs; i++) {
1267 			snprintf(arg, sizeof(arg), "arg%d", i);
1268 			diskname = gctl_get_asciiparam(req, arg);
1269 			if (diskname == NULL) {
1270 				gctl_error(req, "No disk name (%s).", arg);
1271 				error = -2;
1272 				break;
1273 			}
1274 			if (strncmp(diskname, _PATH_DEV, 5) == 0)
1275 				diskname += 5;
1276 
1277 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1278 				if (disk->d_consumer != NULL &&
1279 				    disk->d_consumer->provider != NULL &&
1280 				    strcmp(disk->d_consumer->provider->name,
1281 				     diskname) == 0)
1282 					break;
1283 			}
1284 			if (disk == NULL) {
1285 				gctl_error(req, "Disk '%s' not found.",
1286 				    diskname);
1287 				error = -3;
1288 				break;
1289 			}
1290 
1291 			if (strcmp(verb, "fail") == 0) {
1292 				g_raid_md_fail_disk_nvidia(md, NULL, disk);
1293 				continue;
1294 			}
1295 
1296 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1297 
1298 			/* Erase metadata on deleting disk. */
1299 			nvidia_meta_erase(disk->d_consumer);
1300 
1301 			/* If disk was assigned, just update statuses. */
1302 			if (pd->pd_disk_pos >= 0) {
1303 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1304 				g_raid_kill_consumer(sc, disk->d_consumer);
1305 				disk->d_consumer = NULL;
1306 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1307 					g_raid_change_subdisk_state(sd,
1308 					    G_RAID_SUBDISK_S_NONE);
1309 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1310 					    G_RAID_EVENT_SUBDISK);
1311 				}
1312 			} else {
1313 				/* Otherwise -- delete. */
1314 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1315 				g_raid_destroy_disk(disk);
1316 			}
1317 		}
1318 
1319 		/* Write updated metadata to remaining disks. */
1320 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1321 
1322 		/* Check if anything left except placeholders. */
1323 		if (g_raid_ndisks(sc, -1) ==
1324 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1325 			g_raid_destroy_node(sc, 0);
1326 		else
1327 			g_raid_md_nvidia_refill(sc);
1328 		return (error);
1329 	}
1330 	if (strcmp(verb, "insert") == 0) {
1331 		if (*nargs < 2) {
1332 			gctl_error(req, "Invalid number of arguments.");
1333 			return (-1);
1334 		}
1335 		update = 0;
1336 		for (i = 1; i < *nargs; i++) {
1337 			/* Get disk name. */
1338 			snprintf(arg, sizeof(arg), "arg%d", i);
1339 			diskname = gctl_get_asciiparam(req, arg);
1340 			if (diskname == NULL) {
1341 				gctl_error(req, "No disk name (%s).", arg);
1342 				error = -3;
1343 				break;
1344 			}
1345 
1346 			/* Try to find provider with specified name. */
1347 			g_topology_lock();
1348 			cp = g_raid_open_consumer(sc, diskname);
1349 			if (cp == NULL) {
1350 				gctl_error(req, "Can't open disk '%s'.",
1351 				    diskname);
1352 				g_topology_unlock();
1353 				error = -4;
1354 				break;
1355 			}
1356 			pp = cp->provider;
1357 
1358 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1359 			pd->pd_disk_pos = -3;
1360 			pd->pd_disk_size = pp->mediasize;
1361 
1362 			disk = g_raid_create_disk(sc);
1363 			disk->d_consumer = cp;
1364 			disk->d_md_data = (void *)pd;
1365 			cp->private = disk;
1366 			g_topology_unlock();
1367 
1368 			g_raid_get_disk_info(disk);
1369 
1370 			/* Welcome the "new" disk. */
1371 			update += g_raid_md_nvidia_start_disk(disk);
1372 			if (disk->d_state != G_RAID_DISK_S_SPARE &&
1373 			    disk->d_state != G_RAID_DISK_S_ACTIVE) {
1374 				gctl_error(req, "Disk '%s' doesn't fit.",
1375 				    diskname);
1376 				g_raid_destroy_disk(disk);
1377 				error = -8;
1378 				break;
1379 			}
1380 		}
1381 
1382 		/* Write new metadata if we changed something. */
1383 		if (update)
1384 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1385 		return (error);
1386 	}
1387 	gctl_error(req, "Command '%s' is not supported.", verb);
1388 	return (-100);
1389 }
1390 
1391 static int
1392 g_raid_md_write_nvidia(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1393     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1394 {
1395 	struct g_raid_softc *sc;
1396 	struct g_raid_volume *vol;
1397 	struct g_raid_subdisk *sd;
1398 	struct g_raid_disk *disk;
1399 	struct g_raid_md_nvidia_object *mdi;
1400 	struct g_raid_md_nvidia_perdisk *pd;
1401 	struct nvidia_raid_conf *meta;
1402 	int i, spares;
1403 
1404 	sc = md->mdo_softc;
1405 	mdi = (struct g_raid_md_nvidia_object *)md;
1406 
1407 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1408 		return (0);
1409 
1410 	/* There is only one volume. */
1411 	vol = TAILQ_FIRST(&sc->sc_volumes);
1412 
1413 	/* Fill global fields. */
1414 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1415 	if (mdi->mdio_meta)
1416 		memcpy(meta, mdi->mdio_meta, sizeof(*meta));
1417 	memcpy(meta->nvidia_id, NVIDIA_MAGIC, sizeof(NVIDIA_MAGIC) - 1);
1418 	meta->config_size = 30;
1419 	meta->version = 0x0064;
1420 	meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1421 	meta->sector_size = vol->v_sectorsize;
1422 	nvidia_meta_put_name(meta, vol->v_name);
1423 	meta->magic_0 = NVIDIA_MAGIC0;
1424 	memcpy(&meta->volume_id, &mdi->mdio_volume_id, 16);
1425 	meta->state = NVIDIA_S_IDLE;
1426 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1427 		meta->array_width = 1;
1428 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1429 		meta->array_width = vol->v_disks_count / 2;
1430 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1431 		meta->array_width = vol->v_disks_count - 1;
1432 	else
1433 		meta->array_width = vol->v_disks_count;
1434 	meta->total_disks = vol->v_disks_count;
1435 	meta->orig_array_width = meta->array_width;
1436 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
1437 		meta->type = NVIDIA_T_RAID0;
1438 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1439 		meta->type = NVIDIA_T_RAID1;
1440 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1441 		meta->type = NVIDIA_T_RAID01;
1442 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1443 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1444 		meta->type = NVIDIA_T_CONCAT;
1445 	else if (vol->v_raid_level_qualifier == G_RAID_VOLUME_RLQ_R5LA)
1446 		meta->type = NVIDIA_T_RAID5;
1447 	else
1448 		meta->type = NVIDIA_T_RAID5_SYM;
1449 	meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize;
1450 	meta->strip_bytes = vol->v_strip_size;
1451 	meta->strip_shift = ffs(meta->strip_sectors) - 1;
1452 	meta->strip_mask = meta->strip_sectors - 1;
1453 	meta->stripe_sectors = meta->strip_sectors * meta->orig_array_width;
1454 	meta->stripe_bytes = meta->stripe_sectors * vol->v_sectorsize;
1455 	meta->rebuild_lba = 0;
1456 	meta->orig_type = meta->type;
1457 	meta->orig_total_sectors = meta->total_sectors;
1458 	meta->status = 0;
1459 
1460 	for (i = 0; i < vol->v_disks_count; i++) {
1461 		sd = &vol->v_subdisks[i];
1462 		if ((sd->sd_state == G_RAID_SUBDISK_S_STALE ||
1463 		     sd->sd_state == G_RAID_SUBDISK_S_RESYNC ||
1464 		     vol->v_dirty) &&
1465 		     meta->state != NVIDIA_S_REBUILD)
1466 			meta->state = NVIDIA_S_SYNC;
1467 		else if (sd->sd_state == G_RAID_SUBDISK_S_NEW ||
1468 		     sd->sd_state == G_RAID_SUBDISK_S_REBUILD)
1469 			meta->state = NVIDIA_S_REBUILD;
1470 	}
1471 
1472 	/* We are done. Print meta data and store them to disks. */
1473 	if (mdi->mdio_meta != NULL)
1474 		free(mdi->mdio_meta, M_MD_NVIDIA);
1475 	mdi->mdio_meta = meta;
1476 	spares = 0;
1477 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1478 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1479 		if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1480 		    disk->d_state != G_RAID_DISK_S_SPARE)
1481 			continue;
1482 		if (pd->pd_meta != NULL) {
1483 			free(pd->pd_meta, M_MD_NVIDIA);
1484 			pd->pd_meta = NULL;
1485 		}
1486 		pd->pd_meta = nvidia_meta_copy(meta);
1487 		if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1488 			/* For RAID0+1 we need to translate order. */
1489 			pd->pd_meta->disk_number =
1490 			    nvidia_meta_translate_disk(meta, sd->sd_pos);
1491 			if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE) {
1492 				pd->pd_meta->disk_status = 0x100;
1493 				pd->pd_meta->rebuild_lba =
1494 				    sd->sd_rebuild_pos / vol->v_sectorsize *
1495 				    meta->array_width;
1496 			}
1497 		} else
1498 			pd->pd_meta->disk_number = meta->total_disks + spares++;
1499 		G_RAID_DEBUG(1, "Writing NVIDIA metadata to %s",
1500 		    g_raid_get_diskname(disk));
1501 		g_raid_md_nvidia_print(pd->pd_meta);
1502 		nvidia_meta_write(disk->d_consumer, pd->pd_meta);
1503 	}
1504 	return (0);
1505 }
1506 
1507 static int
1508 g_raid_md_fail_disk_nvidia(struct g_raid_md_object *md,
1509     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1510 {
1511 	struct g_raid_softc *sc;
1512 	struct g_raid_md_nvidia_perdisk *pd;
1513 	struct g_raid_subdisk *sd;
1514 
1515 	sc = md->mdo_softc;
1516 	pd = (struct g_raid_md_nvidia_perdisk *)tdisk->d_md_data;
1517 
1518 	/* We can't fail disk that is not a part of array now. */
1519 	if (pd->pd_disk_pos < 0)
1520 		return (-1);
1521 
1522 	/* Erase metadata to prevent disks's later resurrection. */
1523 	if (tdisk->d_consumer)
1524 		nvidia_meta_erase(tdisk->d_consumer);
1525 
1526 	/* Change states. */
1527 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1528 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1529 		g_raid_change_subdisk_state(sd,
1530 		    G_RAID_SUBDISK_S_FAILED);
1531 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1532 		    G_RAID_EVENT_SUBDISK);
1533 	}
1534 
1535 	/* Write updated metadata to remaining disks. */
1536 	g_raid_md_write_nvidia(md, NULL, NULL, tdisk);
1537 
1538 	/* Check if anything left except placeholders. */
1539 	if (g_raid_ndisks(sc, -1) ==
1540 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1541 		g_raid_destroy_node(sc, 0);
1542 	else
1543 		g_raid_md_nvidia_refill(sc);
1544 	return (0);
1545 }
1546 
1547 static int
1548 g_raid_md_free_disk_nvidia(struct g_raid_md_object *md,
1549     struct g_raid_disk *disk)
1550 {
1551 	struct g_raid_md_nvidia_perdisk *pd;
1552 
1553 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1554 	if (pd->pd_meta != NULL) {
1555 		free(pd->pd_meta, M_MD_NVIDIA);
1556 		pd->pd_meta = NULL;
1557 	}
1558 	free(pd, M_MD_NVIDIA);
1559 	disk->d_md_data = NULL;
1560 	return (0);
1561 }
1562 
1563 static int
1564 g_raid_md_free_nvidia(struct g_raid_md_object *md)
1565 {
1566 	struct g_raid_md_nvidia_object *mdi;
1567 
1568 	mdi = (struct g_raid_md_nvidia_object *)md;
1569 	if (!mdi->mdio_started) {
1570 		mdi->mdio_started = 0;
1571 		callout_stop(&mdi->mdio_start_co);
1572 		G_RAID_DEBUG1(1, md->mdo_softc,
1573 		    "root_mount_rel %p", mdi->mdio_rootmount);
1574 		root_mount_rel(mdi->mdio_rootmount);
1575 		mdi->mdio_rootmount = NULL;
1576 	}
1577 	if (mdi->mdio_meta != NULL) {
1578 		free(mdi->mdio_meta, M_MD_NVIDIA);
1579 		mdi->mdio_meta = NULL;
1580 	}
1581 	return (0);
1582 }
1583 
1584 G_RAID_MD_DECLARE(nvidia, "NVIDIA");
1585