xref: /freebsd/sys/geom/raid/md_nvidia.c (revision 1f474190)
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 
254 	/* Read the anchor sector. */
255 	buf = g_read_data(cp,
256 	    pp->mediasize - 2 * pp->sectorsize, pp->sectorsize, &error);
257 	if (buf == NULL) {
258 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
259 		    pp->name, error);
260 		return (NULL);
261 	}
262 	meta = (struct nvidia_raid_conf *)buf;
263 
264 	/* Check if this is an NVIDIA RAID struct */
265 	if (strncmp(meta->nvidia_id, NVIDIA_MAGIC, strlen(NVIDIA_MAGIC))) {
266 		G_RAID_DEBUG(1, "NVIDIA signature check failed on %s", pp->name);
267 		g_free(buf);
268 		return (NULL);
269 	}
270 	if (meta->config_size > 128 ||
271 	    meta->config_size < 30) {
272 		G_RAID_DEBUG(1, "NVIDIA metadata size looks wrong: %d",
273 		    meta->config_size);
274 		g_free(buf);
275 		return (NULL);
276 	}
277 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
278 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
279 	g_free(buf);
280 
281 	/* Check metadata checksum. */
282 	for (checksum = 0, ptr = (uint32_t *)meta,
283 	    i = 0; i < meta->config_size; i++)
284 		checksum += *ptr++;
285 	if (checksum != 0) {
286 		G_RAID_DEBUG(1, "NVIDIA checksum check failed on %s", pp->name);
287 		free(meta, M_MD_NVIDIA);
288 		return (NULL);
289 	}
290 
291 	/* Check volume state. */
292 	if (meta->state != NVIDIA_S_IDLE && meta->state != NVIDIA_S_INIT &&
293 	    meta->state != NVIDIA_S_REBUILD && meta->state != NVIDIA_S_SYNC) {
294 		G_RAID_DEBUG(1, "NVIDIA unknown state on %s (0x%02x)",
295 		    pp->name, meta->state);
296 		free(meta, M_MD_NVIDIA);
297 		return (NULL);
298 	}
299 
300 	/* Check raid type. */
301 	if (meta->type != NVIDIA_T_RAID0 && meta->type != NVIDIA_T_RAID1 &&
302 	    meta->type != NVIDIA_T_RAID3 && meta->type != NVIDIA_T_RAID5 &&
303 	    meta->type != NVIDIA_T_RAID5_SYM &&
304 	    meta->type != NVIDIA_T_RAID01 && meta->type != NVIDIA_T_CONCAT) {
305 		G_RAID_DEBUG(1, "NVIDIA unknown RAID level on %s (0x%02x)",
306 		    pp->name, meta->type);
307 		free(meta, M_MD_NVIDIA);
308 		return (NULL);
309 	}
310 
311 	return (meta);
312 }
313 
314 static int
315 nvidia_meta_write(struct g_consumer *cp, struct nvidia_raid_conf *meta)
316 {
317 	struct g_provider *pp;
318 	char *buf;
319 	int error, i;
320 	uint32_t checksum, *ptr;
321 
322 	pp = cp->provider;
323 
324 	/* Recalculate checksum for case if metadata were changed. */
325 	meta->checksum = 0;
326 	for (checksum = 0, ptr = (uint32_t *)meta,
327 	    i = 0; i < meta->config_size; i++)
328 		checksum += *ptr++;
329 	meta->checksum -= checksum;
330 
331 	/* Create and fill buffer. */
332 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
333 	memcpy(buf, meta, sizeof(*meta));
334 
335 	/* Write metadata. */
336 	error = g_write_data(cp,
337 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
338 	if (error != 0) {
339 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
340 		    pp->name, error);
341 	}
342 
343 	free(buf, M_MD_NVIDIA);
344 	return (error);
345 }
346 
347 static int
348 nvidia_meta_erase(struct g_consumer *cp)
349 {
350 	struct g_provider *pp;
351 	char *buf;
352 	int error;
353 
354 	pp = cp->provider;
355 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
356 	error = g_write_data(cp,
357 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
358 	if (error != 0) {
359 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
360 		    pp->name, error);
361 	}
362 	free(buf, M_MD_NVIDIA);
363 	return (error);
364 }
365 
366 static struct g_raid_disk *
367 g_raid_md_nvidia_get_disk(struct g_raid_softc *sc, int id)
368 {
369 	struct g_raid_disk	*disk;
370 	struct g_raid_md_nvidia_perdisk *pd;
371 
372 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
373 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
374 		if (pd->pd_disk_pos == id)
375 			break;
376 	}
377 	return (disk);
378 }
379 
380 static int
381 g_raid_md_nvidia_supported(int level, int qual, int disks, int force)
382 {
383 
384 	switch (level) {
385 	case G_RAID_VOLUME_RL_RAID0:
386 		if (disks < 1)
387 			return (0);
388 		if (!force && (disks < 2 || disks > 6))
389 			return (0);
390 		break;
391 	case G_RAID_VOLUME_RL_RAID1:
392 		if (disks < 1)
393 			return (0);
394 		if (!force && (disks != 2))
395 			return (0);
396 		break;
397 	case G_RAID_VOLUME_RL_RAID1E:
398 		if (disks < 2)
399 			return (0);
400 		if (disks % 2 != 0)
401 			return (0);
402 		if (!force && (disks < 4))
403 			return (0);
404 		break;
405 	case G_RAID_VOLUME_RL_SINGLE:
406 		if (disks != 1)
407 			return (0);
408 		break;
409 	case G_RAID_VOLUME_RL_CONCAT:
410 		if (disks < 2)
411 			return (0);
412 		break;
413 	case G_RAID_VOLUME_RL_RAID5:
414 		if (disks < 3)
415 			return (0);
416 		if (qual != G_RAID_VOLUME_RLQ_R5LA &&
417 		    qual != G_RAID_VOLUME_RLQ_R5LS)
418 			return (0);
419 		break;
420 	default:
421 		return (0);
422 	}
423 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
424 		return (0);
425 	return (1);
426 }
427 
428 static int
429 g_raid_md_nvidia_start_disk(struct g_raid_disk *disk)
430 {
431 	struct g_raid_softc *sc;
432 	struct g_raid_subdisk *sd, *tmpsd;
433 	struct g_raid_disk *olddisk, *tmpdisk;
434 	struct g_raid_md_object *md;
435 	struct g_raid_md_nvidia_object *mdi;
436 	struct g_raid_md_nvidia_perdisk *pd, *oldpd;
437 	struct nvidia_raid_conf *meta;
438 	int disk_pos, resurrection = 0;
439 
440 	sc = disk->d_softc;
441 	md = sc->sc_md;
442 	mdi = (struct g_raid_md_nvidia_object *)md;
443 	meta = mdi->mdio_meta;
444 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
445 	olddisk = NULL;
446 
447 	/* Find disk position in metadata by its serial. */
448 	if (pd->pd_meta != NULL) {
449 		disk_pos = pd->pd_meta->disk_number;
450 		if (disk_pos >= meta->total_disks || mdi->mdio_started)
451 			disk_pos = -3;
452 	} else
453 		disk_pos = -3;
454 	/* For RAID0+1 we need to translate order. */
455 	disk_pos = nvidia_meta_translate_disk(meta, disk_pos);
456 	if (disk_pos < 0) {
457 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
458 		/* If we are in the start process, that's all for now. */
459 		if (!mdi->mdio_started)
460 			goto nofit;
461 		/*
462 		 * If we have already started - try to get use of the disk.
463 		 * Try to replace OFFLINE disks first, then FAILED.
464 		 */
465 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
466 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
467 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
468 				continue;
469 			/* Make sure this disk is big enough. */
470 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
471 				if (sd->sd_offset + sd->sd_size + 2 * 512 >
472 				    pd->pd_disk_size) {
473 					G_RAID_DEBUG1(1, sc,
474 					    "Disk too small (%ju < %ju)",
475 					    pd->pd_disk_size,
476 					    sd->sd_offset + sd->sd_size + 512);
477 					break;
478 				}
479 			}
480 			if (sd != NULL)
481 				continue;
482 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
483 				olddisk = tmpdisk;
484 				break;
485 			} else if (olddisk == NULL)
486 				olddisk = tmpdisk;
487 		}
488 		if (olddisk == NULL) {
489 nofit:
490 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
491 			return (1);
492 		}
493 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
494 		disk_pos = oldpd->pd_disk_pos;
495 		resurrection = 1;
496 	}
497 
498 	if (olddisk == NULL) {
499 		/* Find placeholder by position. */
500 		olddisk = g_raid_md_nvidia_get_disk(sc, disk_pos);
501 		if (olddisk == NULL)
502 			panic("No disk at position %d!", disk_pos);
503 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
504 			G_RAID_DEBUG1(1, sc, "More than one disk for pos %d",
505 			    disk_pos);
506 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
507 			return (0);
508 		}
509 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
510 	}
511 
512 	/* Replace failed disk or placeholder with new disk. */
513 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
514 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
515 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
516 		sd->sd_disk = disk;
517 	}
518 	oldpd->pd_disk_pos = -2;
519 	pd->pd_disk_pos = disk_pos;
520 
521 	/* If it was placeholder -- destroy it. */
522 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
523 		g_raid_destroy_disk(olddisk);
524 	} else {
525 		/* Otherwise, make it STALE_FAILED. */
526 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
527 	}
528 
529 	/* Welcome the new disk. */
530 	if (resurrection)
531 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
532 	else// if (pd->pd_meta->disk_status == NVIDIA_S_CURRENT ||
533 	    //pd->pd_meta->disk_status == NVIDIA_S_REBUILD)
534 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
535 //	else
536 //		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
537 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
538 		/*
539 		 * Different disks may have different sizes,
540 		 * in concat mode. Update from real disk size.
541 		 */
542 		if (meta->type == NVIDIA_T_CONCAT)
543 			sd->sd_size = pd->pd_disk_size - 0x800 * 512;
544 
545 		if (resurrection) {
546 			/* New or ex-spare disk. */
547 			g_raid_change_subdisk_state(sd,
548 			    G_RAID_SUBDISK_S_NEW);
549 		} else if (meta->state == NVIDIA_S_REBUILD &&
550 		    (pd->pd_meta->disk_status & 0x100)) {
551 			/* Rebuilding disk. */
552 			g_raid_change_subdisk_state(sd,
553 			    G_RAID_SUBDISK_S_REBUILD);
554 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
555 			    meta->array_width * pd->pd_meta->sector_size;
556 		} else if (meta->state == NVIDIA_S_SYNC) {
557 			/* Resyncing/dirty disk. */
558 			g_raid_change_subdisk_state(sd,
559 			    G_RAID_SUBDISK_S_RESYNC);
560 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
561 			    meta->array_width * pd->pd_meta->sector_size;
562 		} else {
563 			/* Up to date disk. */
564 			g_raid_change_subdisk_state(sd,
565 			    G_RAID_SUBDISK_S_ACTIVE);
566 		}
567 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
568 		    G_RAID_EVENT_SUBDISK);
569 	}
570 
571 	/* Update status of our need for spare. */
572 	if (mdi->mdio_started) {
573 		mdi->mdio_incomplete =
574 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
575 		     mdi->mdio_total_disks);
576 	}
577 
578 	return (resurrection);
579 }
580 
581 static void
582 g_disk_md_nvidia_retaste(void *arg, int pending)
583 {
584 
585 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
586 	g_retaste(&g_raid_class);
587 	free(arg, M_MD_NVIDIA);
588 }
589 
590 static void
591 g_raid_md_nvidia_refill(struct g_raid_softc *sc)
592 {
593 	struct g_raid_md_object *md;
594 	struct g_raid_md_nvidia_object *mdi;
595 	struct g_raid_disk *disk;
596 	struct task *task;
597 	int update, na;
598 
599 	md = sc->sc_md;
600 	mdi = (struct g_raid_md_nvidia_object *)md;
601 	update = 0;
602 	do {
603 		/* Make sure we miss anything. */
604 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
605 		if (na == mdi->mdio_total_disks)
606 			break;
607 
608 		G_RAID_DEBUG1(1, md->mdo_softc,
609 		    "Array is not complete (%d of %d), "
610 		    "trying to refill.", na, mdi->mdio_total_disks);
611 
612 		/* Try to get use some of STALE disks. */
613 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
614 			if (disk->d_state == G_RAID_DISK_S_STALE) {
615 				update += g_raid_md_nvidia_start_disk(disk);
616 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
617 					break;
618 			}
619 		}
620 		if (disk != NULL)
621 			continue;
622 
623 		/* Try to get use some of SPARE disks. */
624 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
625 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
626 				update += g_raid_md_nvidia_start_disk(disk);
627 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
628 					break;
629 			}
630 		}
631 	} while (disk != NULL);
632 
633 	/* Write new metadata if we changed something. */
634 	if (update)
635 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
636 
637 	/* Update status of our need for spare. */
638 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
639 	    mdi->mdio_total_disks);
640 
641 	/* Request retaste hoping to find spare. */
642 	if (mdi->mdio_incomplete) {
643 		task = malloc(sizeof(struct task),
644 		    M_MD_NVIDIA, M_WAITOK | M_ZERO);
645 		TASK_INIT(task, 0, g_disk_md_nvidia_retaste, task);
646 		taskqueue_enqueue(taskqueue_swi, task);
647 	}
648 }
649 
650 static void
651 g_raid_md_nvidia_start(struct g_raid_softc *sc)
652 {
653 	struct g_raid_md_object *md;
654 	struct g_raid_md_nvidia_object *mdi;
655 	struct g_raid_md_nvidia_perdisk *pd;
656 	struct nvidia_raid_conf *meta;
657 	struct g_raid_volume *vol;
658 	struct g_raid_subdisk *sd;
659 	struct g_raid_disk *disk;
660 	off_t size;
661 	int j, disk_pos;
662 	char buf[17];
663 
664 	md = sc->sc_md;
665 	mdi = (struct g_raid_md_nvidia_object *)md;
666 	meta = mdi->mdio_meta;
667 
668 	/* Create volumes and subdisks. */
669 	nvidia_meta_get_name(meta, buf);
670 	vol = g_raid_create_volume(sc, buf, -1);
671 	vol->v_mediasize = (off_t)meta->total_sectors * 512;
672 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
673 	if (meta->type == NVIDIA_T_RAID0) {
674 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
675 		size = vol->v_mediasize / mdi->mdio_total_disks;
676 	} else if (meta->type == NVIDIA_T_RAID1) {
677 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
678 		size = vol->v_mediasize;
679 	} else if (meta->type == NVIDIA_T_RAID01) {
680 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
681 		size = vol->v_mediasize / (mdi->mdio_total_disks / 2);
682 	} else if (meta->type == NVIDIA_T_CONCAT) {
683 		if (mdi->mdio_total_disks == 1)
684 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
685 		else
686 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
687 		size = 0;
688 	} else if (meta->type == NVIDIA_T_RAID5) {
689 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
690 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
691 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
692 	} else if (meta->type == NVIDIA_T_RAID5_SYM) {
693 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
694 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LS;
695 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
696 	} else {
697 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
698 		size = 0;
699 	}
700 	vol->v_strip_size = meta->strip_sectors * 512; //ZZZ
701 	vol->v_disks_count = mdi->mdio_total_disks;
702 	vol->v_sectorsize = 512; //ZZZ
703 	for (j = 0; j < vol->v_disks_count; j++) {
704 		sd = &vol->v_subdisks[j];
705 		sd->sd_offset = 0;
706 		sd->sd_size = size;
707 	}
708 	g_raid_start_volume(vol);
709 
710 	/* Create disk placeholders to store data for later writing. */
711 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
712 		pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
713 		pd->pd_disk_pos = disk_pos;
714 		disk = g_raid_create_disk(sc);
715 		disk->d_md_data = (void *)pd;
716 		disk->d_state = G_RAID_DISK_S_OFFLINE;
717 		sd = &vol->v_subdisks[disk_pos];
718 		sd->sd_disk = disk;
719 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
720 	}
721 
722 	/* Make all disks found till the moment take their places. */
723 	do {
724 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
725 			if (disk->d_state == G_RAID_DISK_S_NONE) {
726 				g_raid_md_nvidia_start_disk(disk);
727 				break;
728 			}
729 		}
730 	} while (disk != NULL);
731 
732 	mdi->mdio_started = 1;
733 	G_RAID_DEBUG1(0, sc, "Array started.");
734 	g_raid_md_write_nvidia(md, NULL, NULL, NULL);
735 
736 	/* Pickup any STALE/SPARE disks to refill array if needed. */
737 	g_raid_md_nvidia_refill(sc);
738 
739 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
740 
741 	callout_stop(&mdi->mdio_start_co);
742 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
743 	root_mount_rel(mdi->mdio_rootmount);
744 	mdi->mdio_rootmount = NULL;
745 }
746 
747 static void
748 g_raid_md_nvidia_new_disk(struct g_raid_disk *disk)
749 {
750 	struct g_raid_softc *sc;
751 	struct g_raid_md_object *md;
752 	struct g_raid_md_nvidia_object *mdi;
753 	struct nvidia_raid_conf *pdmeta;
754 	struct g_raid_md_nvidia_perdisk *pd;
755 
756 	sc = disk->d_softc;
757 	md = sc->sc_md;
758 	mdi = (struct g_raid_md_nvidia_object *)md;
759 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
760 	pdmeta = pd->pd_meta;
761 
762 	if (mdi->mdio_started) {
763 		if (g_raid_md_nvidia_start_disk(disk))
764 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
765 	} else {
766 		if (mdi->mdio_meta == NULL ||
767 		    mdi->mdio_meta->disk_number >= mdi->mdio_meta->total_disks) {
768 			G_RAID_DEBUG1(1, sc, "Newer disk");
769 			if (mdi->mdio_meta != NULL)
770 				free(mdi->mdio_meta, M_MD_NVIDIA);
771 			mdi->mdio_meta = nvidia_meta_copy(pdmeta);
772 			mdi->mdio_total_disks = pdmeta->total_disks;
773 			mdi->mdio_disks_present = 1;
774 		} else if (pdmeta->disk_number < mdi->mdio_meta->total_disks) {
775 			mdi->mdio_disks_present++;
776 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
777 			    mdi->mdio_disks_present,
778 			    mdi->mdio_total_disks);
779 		} else
780 			G_RAID_DEBUG1(1, sc, "Spare disk");
781 
782 		/* If we collected all needed disks - start array. */
783 		if (mdi->mdio_disks_present == mdi->mdio_total_disks)
784 			g_raid_md_nvidia_start(sc);
785 	}
786 }
787 
788 static void
789 g_raid_nvidia_go(void *arg)
790 {
791 	struct g_raid_softc *sc;
792 	struct g_raid_md_object *md;
793 	struct g_raid_md_nvidia_object *mdi;
794 
795 	sc = arg;
796 	md = sc->sc_md;
797 	mdi = (struct g_raid_md_nvidia_object *)md;
798 	if (!mdi->mdio_started) {
799 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
800 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
801 	}
802 }
803 
804 static int
805 g_raid_md_create_nvidia(struct g_raid_md_object *md, struct g_class *mp,
806     struct g_geom **gp)
807 {
808 	struct g_raid_softc *sc;
809 	struct g_raid_md_nvidia_object *mdi;
810 	char name[32];
811 
812 	mdi = (struct g_raid_md_nvidia_object *)md;
813 	arc4rand(&mdi->mdio_volume_id, 16, 0);
814 	snprintf(name, sizeof(name), "NVIDIA-%d",
815 	    atomic_fetchadd_int(&NVIDIANodeID, 1));
816 	sc = g_raid_create_node(mp, name, md);
817 	if (sc == NULL)
818 		return (G_RAID_MD_TASTE_FAIL);
819 	md->mdo_softc = sc;
820 	*gp = sc->sc_geom;
821 	return (G_RAID_MD_TASTE_NEW);
822 }
823 
824 static int
825 g_raid_md_taste_nvidia(struct g_raid_md_object *md, struct g_class *mp,
826                               struct g_consumer *cp, struct g_geom **gp)
827 {
828 	struct g_consumer *rcp;
829 	struct g_provider *pp;
830 	struct g_raid_md_nvidia_object *mdi, *mdi1;
831 	struct g_raid_softc *sc;
832 	struct g_raid_disk *disk;
833 	struct nvidia_raid_conf *meta;
834 	struct g_raid_md_nvidia_perdisk *pd;
835 	struct g_geom *geom;
836 	int result, spare, len;
837 	char name[32];
838 	uint16_t vendor;
839 
840 	G_RAID_DEBUG(1, "Tasting NVIDIA on %s", cp->provider->name);
841 	mdi = (struct g_raid_md_nvidia_object *)md;
842 	pp = cp->provider;
843 
844 	/* Read metadata from device. */
845 	meta = NULL;
846 	g_topology_unlock();
847 	vendor = 0xffff;
848 	len = sizeof(vendor);
849 	if (pp->geom->rank == 1)
850 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
851 	meta = nvidia_meta_read(cp);
852 	g_topology_lock();
853 	if (meta == NULL) {
854 		if (g_raid_aggressive_spare) {
855 			if (vendor == 0x10de) {
856 				G_RAID_DEBUG(1,
857 				    "No NVIDIA metadata, forcing spare.");
858 				spare = 2;
859 				goto search;
860 			} else {
861 				G_RAID_DEBUG(1,
862 				    "NVIDIA vendor mismatch 0x%04x != 0x10de",
863 				    vendor);
864 			}
865 		}
866 		return (G_RAID_MD_TASTE_FAIL);
867 	}
868 
869 	/* Metadata valid. Print it. */
870 	g_raid_md_nvidia_print(meta);
871 	G_RAID_DEBUG(1, "NVIDIA disk position %d", meta->disk_number);
872 	spare = 0;//(meta->type == NVIDIA_T_SPARE) ? 1 : 0;
873 
874 search:
875 	/* Search for matching node. */
876 	sc = NULL;
877 	mdi1 = NULL;
878 	LIST_FOREACH(geom, &mp->geom, geom) {
879 		sc = geom->softc;
880 		if (sc == NULL)
881 			continue;
882 		if (sc->sc_stopping != 0)
883 			continue;
884 		if (sc->sc_md->mdo_class != md->mdo_class)
885 			continue;
886 		mdi1 = (struct g_raid_md_nvidia_object *)sc->sc_md;
887 		if (spare) {
888 			if (mdi1->mdio_incomplete)
889 				break;
890 		} else {
891 			if (memcmp(&mdi1->mdio_volume_id,
892 			     &meta->volume_id, 16) == 0)
893 				break;
894 		}
895 	}
896 
897 	/* Found matching node. */
898 	if (geom != NULL) {
899 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
900 		result = G_RAID_MD_TASTE_EXISTING;
901 
902 	} else if (spare) { /* Not found needy node -- left for later. */
903 		G_RAID_DEBUG(1, "Spare is not needed at this time");
904 		goto fail1;
905 
906 	} else { /* Not found matching node -- create one. */
907 		result = G_RAID_MD_TASTE_NEW;
908 		memcpy(&mdi->mdio_volume_id, &meta->volume_id, 16);
909 		snprintf(name, sizeof(name), "NVIDIA-%d",
910 		    atomic_fetchadd_int(&NVIDIANodeID, 1));
911 		sc = g_raid_create_node(mp, name, md);
912 		md->mdo_softc = sc;
913 		geom = sc->sc_geom;
914 		callout_init(&mdi->mdio_start_co, 1);
915 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
916 		    g_raid_nvidia_go, sc);
917 		mdi->mdio_rootmount = root_mount_hold("GRAID-NVIDIA");
918 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
919 	}
920 
921 	/* There is no return after this point, so we close passed consumer. */
922 	g_access(cp, -1, 0, 0);
923 
924 	rcp = g_new_consumer(geom);
925 	rcp->flags |= G_CF_DIRECT_RECEIVE;
926 	g_attach(rcp, pp);
927 	if (g_access(rcp, 1, 1, 1) != 0)
928 		; //goto fail1;
929 
930 	g_topology_unlock();
931 	sx_xlock(&sc->sc_lock);
932 
933 	pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
934 	pd->pd_meta = meta;
935 	if (spare == 2) {
936 		pd->pd_disk_pos = -3;
937 	} else {
938 		pd->pd_disk_pos = -1;
939 	}
940 	pd->pd_disk_size = pp->mediasize;
941 	disk = g_raid_create_disk(sc);
942 	disk->d_md_data = (void *)pd;
943 	disk->d_consumer = rcp;
944 	rcp->private = disk;
945 
946 	g_raid_get_disk_info(disk);
947 
948 	g_raid_md_nvidia_new_disk(disk);
949 
950 	sx_xunlock(&sc->sc_lock);
951 	g_topology_lock();
952 	*gp = geom;
953 	return (result);
954 fail1:
955 	free(meta, M_MD_NVIDIA);
956 	return (G_RAID_MD_TASTE_FAIL);
957 }
958 
959 static int
960 g_raid_md_event_nvidia(struct g_raid_md_object *md,
961     struct g_raid_disk *disk, u_int event)
962 {
963 	struct g_raid_softc *sc;
964 	struct g_raid_subdisk *sd;
965 	struct g_raid_md_nvidia_object *mdi;
966 	struct g_raid_md_nvidia_perdisk *pd;
967 
968 	sc = md->mdo_softc;
969 	mdi = (struct g_raid_md_nvidia_object *)md;
970 	if (disk == NULL) {
971 		switch (event) {
972 		case G_RAID_NODE_E_START:
973 			if (!mdi->mdio_started) {
974 				/* Bump volume ID to drop missing disks. */
975 				arc4rand(&mdi->mdio_volume_id, 16, 0);
976 				g_raid_md_nvidia_start(sc);
977 			}
978 			return (0);
979 		}
980 		return (-1);
981 	}
982 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
983 	switch (event) {
984 	case G_RAID_DISK_E_DISCONNECTED:
985 		/* If disk was assigned, just update statuses. */
986 		if (pd->pd_disk_pos >= 0) {
987 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
988 			if (disk->d_consumer) {
989 				g_raid_kill_consumer(sc, disk->d_consumer);
990 				disk->d_consumer = NULL;
991 			}
992 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
993 				g_raid_change_subdisk_state(sd,
994 				    G_RAID_SUBDISK_S_NONE);
995 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
996 				    G_RAID_EVENT_SUBDISK);
997 			}
998 		} else {
999 			/* Otherwise -- delete. */
1000 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1001 			g_raid_destroy_disk(disk);
1002 		}
1003 
1004 		if (mdi->mdio_started) {
1005 			/* Bump volume ID to prevent disk resurrection. */
1006 			if (pd->pd_disk_pos >= 0)
1007 				arc4rand(&mdi->mdio_volume_id, 16, 0);
1008 
1009 			/* Write updated metadata to all disks. */
1010 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1011 		}
1012 
1013 		/* Check if anything left except placeholders. */
1014 		if (g_raid_ndisks(sc, -1) ==
1015 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1016 			g_raid_destroy_node(sc, 0);
1017 		else
1018 			g_raid_md_nvidia_refill(sc);
1019 		return (0);
1020 	}
1021 	return (-2);
1022 }
1023 
1024 static int
1025 g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
1026     struct gctl_req *req)
1027 {
1028 	struct g_raid_softc *sc;
1029 	struct g_raid_volume *vol;
1030 	struct g_raid_subdisk *sd;
1031 	struct g_raid_disk *disk;
1032 	struct g_raid_md_nvidia_object *mdi;
1033 	struct g_raid_md_nvidia_perdisk *pd;
1034 	struct g_consumer *cp;
1035 	struct g_provider *pp;
1036 	char arg[16];
1037 	const char *verb, *volname, *levelname, *diskname;
1038 	int *nargs, *force;
1039 	off_t size, sectorsize, strip, volsize;
1040 	intmax_t *sizearg, *striparg;
1041 	int numdisks, i, len, level, qual, update;
1042 	int error;
1043 
1044 	sc = md->mdo_softc;
1045 	mdi = (struct g_raid_md_nvidia_object *)md;
1046 	verb = gctl_get_param(req, "verb", NULL);
1047 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1048 	error = 0;
1049 	if (strcmp(verb, "label") == 0) {
1050 		if (*nargs < 4) {
1051 			gctl_error(req, "Invalid number of arguments.");
1052 			return (-1);
1053 		}
1054 		volname = gctl_get_asciiparam(req, "arg1");
1055 		if (volname == NULL) {
1056 			gctl_error(req, "No volume name.");
1057 			return (-2);
1058 		}
1059 		levelname = gctl_get_asciiparam(req, "arg2");
1060 		if (levelname == NULL) {
1061 			gctl_error(req, "No RAID level.");
1062 			return (-3);
1063 		}
1064 		if (strcasecmp(levelname, "RAID5") == 0)
1065 			levelname = "RAID5-LS";
1066 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1067 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1068 			return (-4);
1069 		}
1070 		numdisks = *nargs - 3;
1071 		force = gctl_get_paraml(req, "force", sizeof(*force));
1072 		if (!g_raid_md_nvidia_supported(level, qual, numdisks,
1073 		    force ? *force : 0)) {
1074 			gctl_error(req, "Unsupported RAID level "
1075 			    "(0x%02x/0x%02x), or number of disks (%d).",
1076 			    level, qual, numdisks);
1077 			return (-5);
1078 		}
1079 
1080 		/* Search for disks, connect them and probe. */
1081 		size = 0x7fffffffffffffffllu;
1082 		sectorsize = 0;
1083 		for (i = 0; i < numdisks; i++) {
1084 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1085 			diskname = gctl_get_asciiparam(req, arg);
1086 			if (diskname == NULL) {
1087 				gctl_error(req, "No disk name (%s).", arg);
1088 				error = -6;
1089 				break;
1090 			}
1091 			if (strcmp(diskname, "NONE") == 0) {
1092 				cp = NULL;
1093 				pp = NULL;
1094 			} else {
1095 				g_topology_lock();
1096 				cp = g_raid_open_consumer(sc, diskname);
1097 				if (cp == NULL) {
1098 					gctl_error(req, "Can't open '%s'.",
1099 					    diskname);
1100 					g_topology_unlock();
1101 					error = -7;
1102 					break;
1103 				}
1104 				pp = cp->provider;
1105 			}
1106 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1107 			pd->pd_disk_pos = i;
1108 			disk = g_raid_create_disk(sc);
1109 			disk->d_md_data = (void *)pd;
1110 			disk->d_consumer = cp;
1111 			if (cp == NULL)
1112 				continue;
1113 			cp->private = disk;
1114 			g_topology_unlock();
1115 
1116 			g_raid_get_disk_info(disk);
1117 
1118 			pd->pd_disk_size = pp->mediasize;
1119 			if (size > pp->mediasize)
1120 				size = pp->mediasize;
1121 			if (sectorsize < pp->sectorsize)
1122 				sectorsize = pp->sectorsize;
1123 		}
1124 		if (error != 0)
1125 			return (error);
1126 
1127 		if (sectorsize <= 0) {
1128 			gctl_error(req, "Can't get sector size.");
1129 			return (-8);
1130 		}
1131 
1132 		/* Reserve space for metadata. */
1133 		size -= 2 * sectorsize;
1134 
1135 		/* Handle size argument. */
1136 		len = sizeof(*sizearg);
1137 		sizearg = gctl_get_param(req, "size", &len);
1138 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1139 		    *sizearg > 0) {
1140 			if (*sizearg > size) {
1141 				gctl_error(req, "Size too big %lld > %lld.",
1142 				    (long long)*sizearg, (long long)size);
1143 				return (-9);
1144 			}
1145 			size = *sizearg;
1146 		}
1147 
1148 		/* Handle strip argument. */
1149 		strip = 131072;
1150 		len = sizeof(*striparg);
1151 		striparg = gctl_get_param(req, "strip", &len);
1152 		if (striparg != NULL && len == sizeof(*striparg) &&
1153 		    *striparg > 0) {
1154 			if (*striparg < sectorsize) {
1155 				gctl_error(req, "Strip size too small.");
1156 				return (-10);
1157 			}
1158 			if (*striparg % sectorsize != 0) {
1159 				gctl_error(req, "Incorrect strip size.");
1160 				return (-11);
1161 			}
1162 			if (strip > 65535 * sectorsize) {
1163 				gctl_error(req, "Strip size too big.");
1164 				return (-12);
1165 			}
1166 			strip = *striparg;
1167 		}
1168 
1169 		/* Round size down to strip or sector. */
1170 		if (level == G_RAID_VOLUME_RL_RAID1)
1171 			size -= (size % sectorsize);
1172 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1173 		    (numdisks & 1) != 0)
1174 			size -= (size % (2 * strip));
1175 		else
1176 			size -= (size % strip);
1177 		if (size <= 0) {
1178 			gctl_error(req, "Size too small.");
1179 			return (-13);
1180 		}
1181 
1182 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1183 		    level == G_RAID_VOLUME_RL_CONCAT ||
1184 		    level == G_RAID_VOLUME_RL_SINGLE)
1185 			volsize = size * numdisks;
1186 		else if (level == G_RAID_VOLUME_RL_RAID1)
1187 			volsize = size;
1188 		else if (level == G_RAID_VOLUME_RL_RAID5)
1189 			volsize = size * (numdisks - 1);
1190 		else { /* RAID1E */
1191 			volsize = ((size * numdisks) / strip / 2) *
1192 			    strip;
1193 		}
1194 		if (volsize > 0xffffffffllu * sectorsize) {
1195 			gctl_error(req, "Size too big.");
1196 			return (-14);
1197 		}
1198 
1199 		/* We have all we need, create things: volume, ... */
1200 		mdi->mdio_total_disks = numdisks;
1201 		mdi->mdio_started = 1;
1202 		vol = g_raid_create_volume(sc, volname, -1);
1203 		vol->v_md_data = (void *)(intptr_t)0;
1204 		vol->v_raid_level = level;
1205 		vol->v_raid_level_qualifier = qual;
1206 		vol->v_strip_size = strip;
1207 		vol->v_disks_count = numdisks;
1208 		vol->v_mediasize = volsize;
1209 		vol->v_sectorsize = sectorsize;
1210 		g_raid_start_volume(vol);
1211 
1212 		/* , and subdisks. */
1213 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1214 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1215 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1216 			sd->sd_disk = disk;
1217 			sd->sd_offset = 0;
1218 			sd->sd_size = size;
1219 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1220 			if (sd->sd_disk->d_consumer != NULL) {
1221 				g_raid_change_disk_state(disk,
1222 				    G_RAID_DISK_S_ACTIVE);
1223 				g_raid_change_subdisk_state(sd,
1224 				    G_RAID_SUBDISK_S_ACTIVE);
1225 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1226 				    G_RAID_EVENT_SUBDISK);
1227 			} else {
1228 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1229 			}
1230 		}
1231 
1232 		/* Write metadata based on created entities. */
1233 		G_RAID_DEBUG1(0, sc, "Array started.");
1234 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1235 
1236 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1237 		g_raid_md_nvidia_refill(sc);
1238 
1239 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1240 		    G_RAID_EVENT_VOLUME);
1241 		return (0);
1242 	}
1243 	if (strcmp(verb, "delete") == 0) {
1244 		/* Check if some volume is still open. */
1245 		force = gctl_get_paraml(req, "force", sizeof(*force));
1246 		if (force != NULL && *force == 0 &&
1247 		    g_raid_nopens(sc) != 0) {
1248 			gctl_error(req, "Some volume is still open.");
1249 			return (-4);
1250 		}
1251 
1252 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1253 			if (disk->d_consumer)
1254 				nvidia_meta_erase(disk->d_consumer);
1255 		}
1256 		g_raid_destroy_node(sc, 0);
1257 		return (0);
1258 	}
1259 	if (strcmp(verb, "remove") == 0 ||
1260 	    strcmp(verb, "fail") == 0) {
1261 		if (*nargs < 2) {
1262 			gctl_error(req, "Invalid number of arguments.");
1263 			return (-1);
1264 		}
1265 		for (i = 1; i < *nargs; i++) {
1266 			snprintf(arg, sizeof(arg), "arg%d", i);
1267 			diskname = gctl_get_asciiparam(req, arg);
1268 			if (diskname == NULL) {
1269 				gctl_error(req, "No disk name (%s).", arg);
1270 				error = -2;
1271 				break;
1272 			}
1273 			if (strncmp(diskname, _PATH_DEV, 5) == 0)
1274 				diskname += 5;
1275 
1276 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1277 				if (disk->d_consumer != NULL &&
1278 				    disk->d_consumer->provider != NULL &&
1279 				    strcmp(disk->d_consumer->provider->name,
1280 				     diskname) == 0)
1281 					break;
1282 			}
1283 			if (disk == NULL) {
1284 				gctl_error(req, "Disk '%s' not found.",
1285 				    diskname);
1286 				error = -3;
1287 				break;
1288 			}
1289 
1290 			if (strcmp(verb, "fail") == 0) {
1291 				g_raid_md_fail_disk_nvidia(md, NULL, disk);
1292 				continue;
1293 			}
1294 
1295 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1296 
1297 			/* Erase metadata on deleting disk. */
1298 			nvidia_meta_erase(disk->d_consumer);
1299 
1300 			/* If disk was assigned, just update statuses. */
1301 			if (pd->pd_disk_pos >= 0) {
1302 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1303 				g_raid_kill_consumer(sc, disk->d_consumer);
1304 				disk->d_consumer = NULL;
1305 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1306 					g_raid_change_subdisk_state(sd,
1307 					    G_RAID_SUBDISK_S_NONE);
1308 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1309 					    G_RAID_EVENT_SUBDISK);
1310 				}
1311 			} else {
1312 				/* Otherwise -- delete. */
1313 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1314 				g_raid_destroy_disk(disk);
1315 			}
1316 		}
1317 
1318 		/* Write updated metadata to remaining disks. */
1319 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1320 
1321 		/* Check if anything left except placeholders. */
1322 		if (g_raid_ndisks(sc, -1) ==
1323 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1324 			g_raid_destroy_node(sc, 0);
1325 		else
1326 			g_raid_md_nvidia_refill(sc);
1327 		return (error);
1328 	}
1329 	if (strcmp(verb, "insert") == 0) {
1330 		if (*nargs < 2) {
1331 			gctl_error(req, "Invalid number of arguments.");
1332 			return (-1);
1333 		}
1334 		update = 0;
1335 		for (i = 1; i < *nargs; i++) {
1336 			/* Get disk name. */
1337 			snprintf(arg, sizeof(arg), "arg%d", i);
1338 			diskname = gctl_get_asciiparam(req, arg);
1339 			if (diskname == NULL) {
1340 				gctl_error(req, "No disk name (%s).", arg);
1341 				error = -3;
1342 				break;
1343 			}
1344 
1345 			/* Try to find provider with specified name. */
1346 			g_topology_lock();
1347 			cp = g_raid_open_consumer(sc, diskname);
1348 			if (cp == NULL) {
1349 				gctl_error(req, "Can't open disk '%s'.",
1350 				    diskname);
1351 				g_topology_unlock();
1352 				error = -4;
1353 				break;
1354 			}
1355 			pp = cp->provider;
1356 
1357 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1358 			pd->pd_disk_pos = -3;
1359 			pd->pd_disk_size = pp->mediasize;
1360 
1361 			disk = g_raid_create_disk(sc);
1362 			disk->d_consumer = cp;
1363 			disk->d_md_data = (void *)pd;
1364 			cp->private = disk;
1365 			g_topology_unlock();
1366 
1367 			g_raid_get_disk_info(disk);
1368 
1369 			/* Welcome the "new" disk. */
1370 			update += g_raid_md_nvidia_start_disk(disk);
1371 			if (disk->d_state != G_RAID_DISK_S_SPARE &&
1372 			    disk->d_state != G_RAID_DISK_S_ACTIVE) {
1373 				gctl_error(req, "Disk '%s' doesn't fit.",
1374 				    diskname);
1375 				g_raid_destroy_disk(disk);
1376 				error = -8;
1377 				break;
1378 			}
1379 		}
1380 
1381 		/* Write new metadata if we changed something. */
1382 		if (update)
1383 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1384 		return (error);
1385 	}
1386 	gctl_error(req, "Command '%s' is not supported.", verb);
1387 	return (-100);
1388 }
1389 
1390 static int
1391 g_raid_md_write_nvidia(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1392     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1393 {
1394 	struct g_raid_softc *sc;
1395 	struct g_raid_volume *vol;
1396 	struct g_raid_subdisk *sd;
1397 	struct g_raid_disk *disk;
1398 	struct g_raid_md_nvidia_object *mdi;
1399 	struct g_raid_md_nvidia_perdisk *pd;
1400 	struct nvidia_raid_conf *meta;
1401 	int i, spares;
1402 
1403 	sc = md->mdo_softc;
1404 	mdi = (struct g_raid_md_nvidia_object *)md;
1405 
1406 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1407 		return (0);
1408 
1409 	/* There is only one volume. */
1410 	vol = TAILQ_FIRST(&sc->sc_volumes);
1411 
1412 	/* Fill global fields. */
1413 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1414 	if (mdi->mdio_meta)
1415 		memcpy(meta, mdi->mdio_meta, sizeof(*meta));
1416 	memcpy(meta->nvidia_id, NVIDIA_MAGIC, sizeof(NVIDIA_MAGIC) - 1);
1417 	meta->config_size = 30;
1418 	meta->version = 0x0064;
1419 	meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1420 	meta->sector_size = vol->v_sectorsize;
1421 	nvidia_meta_put_name(meta, vol->v_name);
1422 	meta->magic_0 = NVIDIA_MAGIC0;
1423 	memcpy(&meta->volume_id, &mdi->mdio_volume_id, 16);
1424 	meta->state = NVIDIA_S_IDLE;
1425 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1426 		meta->array_width = 1;
1427 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1428 		meta->array_width = vol->v_disks_count / 2;
1429 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1430 		meta->array_width = vol->v_disks_count - 1;
1431 	else
1432 		meta->array_width = vol->v_disks_count;
1433 	meta->total_disks = vol->v_disks_count;
1434 	meta->orig_array_width = meta->array_width;
1435 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
1436 		meta->type = NVIDIA_T_RAID0;
1437 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1438 		meta->type = NVIDIA_T_RAID1;
1439 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1440 		meta->type = NVIDIA_T_RAID01;
1441 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1442 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1443 		meta->type = NVIDIA_T_CONCAT;
1444 	else if (vol->v_raid_level_qualifier == G_RAID_VOLUME_RLQ_R5LA)
1445 		meta->type = NVIDIA_T_RAID5;
1446 	else
1447 		meta->type = NVIDIA_T_RAID5_SYM;
1448 	meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize;
1449 	meta->strip_bytes = vol->v_strip_size;
1450 	meta->strip_shift = ffs(meta->strip_sectors) - 1;
1451 	meta->strip_mask = meta->strip_sectors - 1;
1452 	meta->stripe_sectors = meta->strip_sectors * meta->orig_array_width;
1453 	meta->stripe_bytes = meta->stripe_sectors * vol->v_sectorsize;
1454 	meta->rebuild_lba = 0;
1455 	meta->orig_type = meta->type;
1456 	meta->orig_total_sectors = meta->total_sectors;
1457 	meta->status = 0;
1458 
1459 	for (i = 0; i < vol->v_disks_count; i++) {
1460 		sd = &vol->v_subdisks[i];
1461 		if ((sd->sd_state == G_RAID_SUBDISK_S_STALE ||
1462 		     sd->sd_state == G_RAID_SUBDISK_S_RESYNC ||
1463 		     vol->v_dirty) &&
1464 		     meta->state != NVIDIA_S_REBUILD)
1465 			meta->state = NVIDIA_S_SYNC;
1466 		else if (sd->sd_state == G_RAID_SUBDISK_S_NEW ||
1467 		     sd->sd_state == G_RAID_SUBDISK_S_REBUILD)
1468 			meta->state = NVIDIA_S_REBUILD;
1469 	}
1470 
1471 	/* We are done. Print meta data and store them to disks. */
1472 	if (mdi->mdio_meta != NULL)
1473 		free(mdi->mdio_meta, M_MD_NVIDIA);
1474 	mdi->mdio_meta = meta;
1475 	spares = 0;
1476 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1477 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1478 		if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1479 		    disk->d_state != G_RAID_DISK_S_SPARE)
1480 			continue;
1481 		if (pd->pd_meta != NULL) {
1482 			free(pd->pd_meta, M_MD_NVIDIA);
1483 			pd->pd_meta = NULL;
1484 		}
1485 		pd->pd_meta = nvidia_meta_copy(meta);
1486 		if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1487 			/* For RAID0+1 we need to translate order. */
1488 			pd->pd_meta->disk_number =
1489 			    nvidia_meta_translate_disk(meta, sd->sd_pos);
1490 			if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE) {
1491 				pd->pd_meta->disk_status = 0x100;
1492 				pd->pd_meta->rebuild_lba =
1493 				    sd->sd_rebuild_pos / vol->v_sectorsize *
1494 				    meta->array_width;
1495 			}
1496 		} else
1497 			pd->pd_meta->disk_number = meta->total_disks + spares++;
1498 		G_RAID_DEBUG(1, "Writing NVIDIA metadata to %s",
1499 		    g_raid_get_diskname(disk));
1500 		g_raid_md_nvidia_print(pd->pd_meta);
1501 		nvidia_meta_write(disk->d_consumer, pd->pd_meta);
1502 	}
1503 	return (0);
1504 }
1505 
1506 static int
1507 g_raid_md_fail_disk_nvidia(struct g_raid_md_object *md,
1508     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1509 {
1510 	struct g_raid_softc *sc;
1511 	struct g_raid_md_nvidia_perdisk *pd;
1512 	struct g_raid_subdisk *sd;
1513 
1514 	sc = md->mdo_softc;
1515 	pd = (struct g_raid_md_nvidia_perdisk *)tdisk->d_md_data;
1516 
1517 	/* We can't fail disk that is not a part of array now. */
1518 	if (pd->pd_disk_pos < 0)
1519 		return (-1);
1520 
1521 	/* Erase metadata to prevent disks's later resurrection. */
1522 	if (tdisk->d_consumer)
1523 		nvidia_meta_erase(tdisk->d_consumer);
1524 
1525 	/* Change states. */
1526 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1527 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1528 		g_raid_change_subdisk_state(sd,
1529 		    G_RAID_SUBDISK_S_FAILED);
1530 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1531 		    G_RAID_EVENT_SUBDISK);
1532 	}
1533 
1534 	/* Write updated metadata to remaining disks. */
1535 	g_raid_md_write_nvidia(md, NULL, NULL, tdisk);
1536 
1537 	/* Check if anything left except placeholders. */
1538 	if (g_raid_ndisks(sc, -1) ==
1539 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1540 		g_raid_destroy_node(sc, 0);
1541 	else
1542 		g_raid_md_nvidia_refill(sc);
1543 	return (0);
1544 }
1545 
1546 static int
1547 g_raid_md_free_disk_nvidia(struct g_raid_md_object *md,
1548     struct g_raid_disk *disk)
1549 {
1550 	struct g_raid_md_nvidia_perdisk *pd;
1551 
1552 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1553 	if (pd->pd_meta != NULL) {
1554 		free(pd->pd_meta, M_MD_NVIDIA);
1555 		pd->pd_meta = NULL;
1556 	}
1557 	free(pd, M_MD_NVIDIA);
1558 	disk->d_md_data = NULL;
1559 	return (0);
1560 }
1561 
1562 static int
1563 g_raid_md_free_nvidia(struct g_raid_md_object *md)
1564 {
1565 	struct g_raid_md_nvidia_object *mdi;
1566 
1567 	mdi = (struct g_raid_md_nvidia_object *)md;
1568 	if (!mdi->mdio_started) {
1569 		mdi->mdio_started = 0;
1570 		callout_stop(&mdi->mdio_start_co);
1571 		G_RAID_DEBUG1(1, md->mdo_softc,
1572 		    "root_mount_rel %p", mdi->mdio_rootmount);
1573 		root_mount_rel(mdi->mdio_rootmount);
1574 		mdi->mdio_rootmount = NULL;
1575 	}
1576 	if (mdi->mdio_meta != NULL) {
1577 		free(mdi->mdio_meta, M_MD_NVIDIA);
1578 		mdi->mdio_meta = NULL;
1579 	}
1580 	return (0);
1581 }
1582 
1583 G_RAID_MD_DECLARE(nvidia, "NVIDIA");
1584