xref: /freebsd/sys/geom/raid/md_nvidia.c (revision 734e82fe33aa764367791a7d603b383996c6b40b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include <sys/param.h>
32 #include <sys/bio.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/kobj.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/systm.h>
41 #include <sys/taskqueue.h>
42 #include <geom/geom.h>
43 #include <geom/geom_dbg.h>
44 #include "geom/raid/g_raid.h"
45 #include "g_raid_md_if.h"
46 
47 static MALLOC_DEFINE(M_MD_NVIDIA, "md_nvidia_data", "GEOM_RAID NVIDIA metadata");
48 
49 struct nvidia_raid_conf {
50 	uint8_t		nvidia_id[8];
51 #define NVIDIA_MAGIC                "NVIDIA  "
52 
53 	uint32_t	config_size;
54 	uint32_t	checksum;
55 	uint16_t	version;
56 	uint8_t		disk_number;
57 	uint8_t		dummy_0;
58 	uint32_t	total_sectors;
59 	uint32_t	sector_size;
60 	uint8_t		name[16];
61 	uint8_t		revision[4];
62 	uint32_t	disk_status;
63 
64 	uint32_t	magic_0;
65 #define NVIDIA_MAGIC0		0x00640044
66 
67 	uint64_t	volume_id[2];
68 	uint8_t		state;
69 #define NVIDIA_S_IDLE		0
70 #define NVIDIA_S_INIT		2
71 #define NVIDIA_S_REBUILD	3
72 #define NVIDIA_S_UPGRADE	4
73 #define NVIDIA_S_SYNC		5
74 	uint8_t		array_width;
75 	uint8_t		total_disks;
76 	uint8_t		orig_array_width;
77 	uint16_t	type;
78 #define NVIDIA_T_RAID0		0x0080
79 #define NVIDIA_T_RAID1		0x0081
80 #define NVIDIA_T_RAID3		0x0083
81 #define NVIDIA_T_RAID5		0x0085	/* RLQ = 00/02? */
82 #define NVIDIA_T_RAID5_SYM	0x0095	/* RLQ = 03 */
83 #define NVIDIA_T_RAID10		0x008a
84 #define NVIDIA_T_RAID01		0x8180
85 #define NVIDIA_T_CONCAT		0x00ff
86 
87 	uint16_t	dummy_3;
88 	uint32_t	strip_sectors;
89 	uint32_t	strip_bytes;
90 	uint32_t	strip_shift;
91 	uint32_t	strip_mask;
92 	uint32_t	stripe_sectors;
93 	uint32_t	stripe_bytes;
94 	uint32_t	rebuild_lba;
95 	uint32_t	orig_type;
96 	uint32_t	orig_total_sectors;
97 	uint32_t	status;
98 #define NVIDIA_S_BOOTABLE	0x00000001
99 #define NVIDIA_S_DEGRADED	0x00000002
100 
101 	uint32_t	filler[98];
102 } __packed;
103 
104 struct g_raid_md_nvidia_perdisk {
105 	struct nvidia_raid_conf	*pd_meta;
106 	int			 pd_disk_pos;
107 	off_t			 pd_disk_size;
108 };
109 
110 struct g_raid_md_nvidia_object {
111 	struct g_raid_md_object	 mdio_base;
112 	uint64_t		 mdio_volume_id[2];
113 	struct nvidia_raid_conf	*mdio_meta;
114 	struct callout		 mdio_start_co;	/* STARTING state timer. */
115 	int			 mdio_total_disks;
116 	int			 mdio_disks_present;
117 	int			 mdio_started;
118 	int			 mdio_incomplete;
119 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
120 };
121 
122 static g_raid_md_create_t g_raid_md_create_nvidia;
123 static g_raid_md_taste_t g_raid_md_taste_nvidia;
124 static g_raid_md_event_t g_raid_md_event_nvidia;
125 static g_raid_md_ctl_t g_raid_md_ctl_nvidia;
126 static g_raid_md_write_t g_raid_md_write_nvidia;
127 static g_raid_md_fail_disk_t g_raid_md_fail_disk_nvidia;
128 static g_raid_md_free_disk_t g_raid_md_free_disk_nvidia;
129 static g_raid_md_free_t g_raid_md_free_nvidia;
130 
131 static kobj_method_t g_raid_md_nvidia_methods[] = {
132 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_nvidia),
133 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_nvidia),
134 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_nvidia),
135 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_nvidia),
136 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_nvidia),
137 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_nvidia),
138 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_nvidia),
139 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_nvidia),
140 	{ 0, 0 }
141 };
142 
143 static struct g_raid_md_class g_raid_md_nvidia_class = {
144 	"NVIDIA",
145 	g_raid_md_nvidia_methods,
146 	sizeof(struct g_raid_md_nvidia_object),
147 	.mdc_enable = 1,
148 	.mdc_priority = 100
149 };
150 
151 static int NVIDIANodeID = 1;
152 
153 static void
154 g_raid_md_nvidia_print(struct nvidia_raid_conf *meta)
155 {
156 
157 	if (g_raid_debug < 1)
158 		return;
159 
160 	printf("********* ATA NVIDIA RAID Metadata *********\n");
161 	printf("nvidia_id           <%.8s>\n", meta->nvidia_id);
162 	printf("config_size         %u\n", meta->config_size);
163 	printf("checksum            0x%08x\n", meta->checksum);
164 	printf("version             0x%04x\n", meta->version);
165 	printf("disk_number         %d\n", meta->disk_number);
166 	printf("dummy_0             0x%02x\n", meta->dummy_0);
167 	printf("total_sectors       %u\n", meta->total_sectors);
168 	printf("sector_size         %u\n", meta->sector_size);
169 	printf("name                <%.16s>\n", meta->name);
170 	printf("revision            0x%02x%02x%02x%02x\n",
171 	    meta->revision[0], meta->revision[1],
172 	    meta->revision[2], meta->revision[3]);
173 	printf("disk_status         0x%08x\n", meta->disk_status);
174 	printf("magic_0             0x%08x\n", meta->magic_0);
175 	printf("volume_id           0x%016jx%016jx\n",
176 	    meta->volume_id[1], meta->volume_id[0]);
177 	printf("state               0x%02x\n", meta->state);
178 	printf("array_width         %u\n", meta->array_width);
179 	printf("total_disks         %u\n", meta->total_disks);
180 	printf("orig_array_width    %u\n", meta->orig_array_width);
181 	printf("type                0x%04x\n", meta->type);
182 	printf("dummy_3             0x%04x\n", meta->dummy_3);
183 	printf("strip_sectors       %u\n", meta->strip_sectors);
184 	printf("strip_bytes         %u\n", meta->strip_bytes);
185 	printf("strip_shift         %u\n", meta->strip_shift);
186 	printf("strip_mask          0x%08x\n", meta->strip_mask);
187 	printf("stripe_sectors      %u\n", meta->stripe_sectors);
188 	printf("stripe_bytes        %u\n", meta->stripe_bytes);
189 	printf("rebuild_lba         %u\n", meta->rebuild_lba);
190 	printf("orig_type           0x%04x\n", meta->orig_type);
191 	printf("orig_total_sectors  %u\n", meta->orig_total_sectors);
192 	printf("status              0x%08x\n", meta->status);
193 	printf("=================================================\n");
194 }
195 
196 static struct nvidia_raid_conf *
197 nvidia_meta_copy(struct nvidia_raid_conf *meta)
198 {
199 	struct nvidia_raid_conf *nmeta;
200 
201 	nmeta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
202 	memcpy(nmeta, meta, sizeof(*meta));
203 	return (nmeta);
204 }
205 
206 static int
207 nvidia_meta_translate_disk(struct nvidia_raid_conf *meta, int md_disk_pos)
208 {
209 	int disk_pos;
210 
211 	if (md_disk_pos >= 0 && meta->type == NVIDIA_T_RAID01) {
212 		disk_pos = (md_disk_pos / meta->array_width) +
213 		    (md_disk_pos % meta->array_width) * meta->array_width;
214 	} else
215 		disk_pos = md_disk_pos;
216 	return (disk_pos);
217 }
218 
219 static void
220 nvidia_meta_get_name(struct nvidia_raid_conf *meta, char *buf)
221 {
222 	int i;
223 
224 	strncpy(buf, meta->name, 16);
225 	buf[16] = 0;
226 	for (i = 15; i >= 0; i--) {
227 		if (buf[i] > 0x20)
228 			break;
229 		buf[i] = 0;
230 	}
231 }
232 
233 static void
234 nvidia_meta_put_name(struct nvidia_raid_conf *meta, char *buf)
235 {
236 
237 	memset(meta->name, 0x20, 16);
238 	memcpy(meta->name, buf, MIN(strlen(buf), 16));
239 }
240 
241 static struct nvidia_raid_conf *
242 nvidia_meta_read(struct g_consumer *cp)
243 {
244 	struct g_provider *pp;
245 	struct nvidia_raid_conf *meta;
246 	char *buf;
247 	int error, i;
248 	uint32_t checksum, *ptr;
249 
250 	pp = cp->provider;
251 	if (pp->sectorsize < sizeof(*meta))
252 		return (NULL);
253 	/* Read the anchor sector. */
254 	buf = g_read_data(cp,
255 	    pp->mediasize - 2 * pp->sectorsize, pp->sectorsize, &error);
256 	if (buf == NULL) {
257 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
258 		    pp->name, error);
259 		return (NULL);
260 	}
261 	meta = (struct nvidia_raid_conf *)buf;
262 
263 	/* Check if this is an NVIDIA RAID struct */
264 	if (strncmp(meta->nvidia_id, NVIDIA_MAGIC, strlen(NVIDIA_MAGIC))) {
265 		G_RAID_DEBUG(1, "NVIDIA signature check failed on %s", pp->name);
266 		g_free(buf);
267 		return (NULL);
268 	}
269 	if (meta->config_size > 128 ||
270 	    meta->config_size < 30) {
271 		G_RAID_DEBUG(1, "NVIDIA metadata size looks wrong: %d",
272 		    meta->config_size);
273 		g_free(buf);
274 		return (NULL);
275 	}
276 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
277 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
278 	g_free(buf);
279 
280 	/* Check metadata checksum. */
281 	for (checksum = 0, ptr = (uint32_t *)meta,
282 	    i = 0; i < meta->config_size; i++)
283 		checksum += *ptr++;
284 	if (checksum != 0) {
285 		G_RAID_DEBUG(1, "NVIDIA checksum check failed on %s", pp->name);
286 		free(meta, M_MD_NVIDIA);
287 		return (NULL);
288 	}
289 
290 	/* Check volume state. */
291 	if (meta->state != NVIDIA_S_IDLE && meta->state != NVIDIA_S_INIT &&
292 	    meta->state != NVIDIA_S_REBUILD && meta->state != NVIDIA_S_SYNC) {
293 		G_RAID_DEBUG(1, "NVIDIA unknown state on %s (0x%02x)",
294 		    pp->name, meta->state);
295 		free(meta, M_MD_NVIDIA);
296 		return (NULL);
297 	}
298 
299 	/* Check raid type. */
300 	if (meta->type != NVIDIA_T_RAID0 && meta->type != NVIDIA_T_RAID1 &&
301 	    meta->type != NVIDIA_T_RAID3 && meta->type != NVIDIA_T_RAID5 &&
302 	    meta->type != NVIDIA_T_RAID5_SYM &&
303 	    meta->type != NVIDIA_T_RAID01 && meta->type != NVIDIA_T_CONCAT) {
304 		G_RAID_DEBUG(1, "NVIDIA unknown RAID level on %s (0x%02x)",
305 		    pp->name, meta->type);
306 		free(meta, M_MD_NVIDIA);
307 		return (NULL);
308 	}
309 
310 	return (meta);
311 }
312 
313 static int
314 nvidia_meta_write(struct g_consumer *cp, struct nvidia_raid_conf *meta)
315 {
316 	struct g_provider *pp;
317 	char *buf;
318 	int error, i;
319 	uint32_t checksum, *ptr;
320 
321 	pp = cp->provider;
322 
323 	/* Recalculate checksum for case if metadata were changed. */
324 	meta->checksum = 0;
325 	for (checksum = 0, ptr = (uint32_t *)meta,
326 	    i = 0; i < meta->config_size; i++)
327 		checksum += *ptr++;
328 	meta->checksum -= checksum;
329 
330 	/* Create and fill buffer. */
331 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
332 	memcpy(buf, meta, sizeof(*meta));
333 
334 	/* Write metadata. */
335 	error = g_write_data(cp,
336 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
337 	if (error != 0) {
338 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
339 		    pp->name, error);
340 	}
341 
342 	free(buf, M_MD_NVIDIA);
343 	return (error);
344 }
345 
346 static int
347 nvidia_meta_erase(struct g_consumer *cp)
348 {
349 	struct g_provider *pp;
350 	char *buf;
351 	int error;
352 
353 	pp = cp->provider;
354 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
355 	error = g_write_data(cp,
356 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
357 	if (error != 0) {
358 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
359 		    pp->name, error);
360 	}
361 	free(buf, M_MD_NVIDIA);
362 	return (error);
363 }
364 
365 static struct g_raid_disk *
366 g_raid_md_nvidia_get_disk(struct g_raid_softc *sc, int id)
367 {
368 	struct g_raid_disk	*disk;
369 	struct g_raid_md_nvidia_perdisk *pd;
370 
371 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
372 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
373 		if (pd->pd_disk_pos == id)
374 			break;
375 	}
376 	return (disk);
377 }
378 
379 static int
380 g_raid_md_nvidia_supported(int level, int qual, int disks, int force)
381 {
382 
383 	switch (level) {
384 	case G_RAID_VOLUME_RL_RAID0:
385 		if (disks < 1)
386 			return (0);
387 		if (!force && (disks < 2 || disks > 6))
388 			return (0);
389 		break;
390 	case G_RAID_VOLUME_RL_RAID1:
391 		if (disks < 1)
392 			return (0);
393 		if (!force && (disks != 2))
394 			return (0);
395 		break;
396 	case G_RAID_VOLUME_RL_RAID1E:
397 		if (disks < 2)
398 			return (0);
399 		if (disks % 2 != 0)
400 			return (0);
401 		if (!force && (disks < 4))
402 			return (0);
403 		break;
404 	case G_RAID_VOLUME_RL_SINGLE:
405 		if (disks != 1)
406 			return (0);
407 		break;
408 	case G_RAID_VOLUME_RL_CONCAT:
409 		if (disks < 2)
410 			return (0);
411 		break;
412 	case G_RAID_VOLUME_RL_RAID5:
413 		if (disks < 3)
414 			return (0);
415 		if (qual != G_RAID_VOLUME_RLQ_R5LA &&
416 		    qual != G_RAID_VOLUME_RLQ_R5LS)
417 			return (0);
418 		break;
419 	default:
420 		return (0);
421 	}
422 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
423 		return (0);
424 	return (1);
425 }
426 
427 static int
428 g_raid_md_nvidia_start_disk(struct g_raid_disk *disk)
429 {
430 	struct g_raid_softc *sc;
431 	struct g_raid_subdisk *sd, *tmpsd;
432 	struct g_raid_disk *olddisk, *tmpdisk;
433 	struct g_raid_md_object *md;
434 	struct g_raid_md_nvidia_object *mdi;
435 	struct g_raid_md_nvidia_perdisk *pd, *oldpd;
436 	struct nvidia_raid_conf *meta;
437 	int disk_pos, resurrection = 0;
438 
439 	sc = disk->d_softc;
440 	md = sc->sc_md;
441 	mdi = (struct g_raid_md_nvidia_object *)md;
442 	meta = mdi->mdio_meta;
443 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
444 	olddisk = NULL;
445 
446 	/* Find disk position in metadata by its serial. */
447 	if (pd->pd_meta != NULL) {
448 		disk_pos = pd->pd_meta->disk_number;
449 		if (disk_pos >= meta->total_disks || mdi->mdio_started)
450 			disk_pos = -3;
451 	} else
452 		disk_pos = -3;
453 	/* For RAID0+1 we need to translate order. */
454 	disk_pos = nvidia_meta_translate_disk(meta, disk_pos);
455 	if (disk_pos < 0) {
456 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
457 		/* If we are in the start process, that's all for now. */
458 		if (!mdi->mdio_started)
459 			goto nofit;
460 		/*
461 		 * If we have already started - try to get use of the disk.
462 		 * Try to replace OFFLINE disks first, then FAILED.
463 		 */
464 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
465 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
466 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
467 				continue;
468 			/* Make sure this disk is big enough. */
469 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
470 				if (sd->sd_offset + sd->sd_size + 2 * 512 >
471 				    pd->pd_disk_size) {
472 					G_RAID_DEBUG1(1, sc,
473 					    "Disk too small (%ju < %ju)",
474 					    pd->pd_disk_size,
475 					    sd->sd_offset + sd->sd_size + 512);
476 					break;
477 				}
478 			}
479 			if (sd != NULL)
480 				continue;
481 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
482 				olddisk = tmpdisk;
483 				break;
484 			} else if (olddisk == NULL)
485 				olddisk = tmpdisk;
486 		}
487 		if (olddisk == NULL) {
488 nofit:
489 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
490 			return (1);
491 		}
492 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
493 		disk_pos = oldpd->pd_disk_pos;
494 		resurrection = 1;
495 	}
496 
497 	if (olddisk == NULL) {
498 		/* Find placeholder by position. */
499 		olddisk = g_raid_md_nvidia_get_disk(sc, disk_pos);
500 		if (olddisk == NULL)
501 			panic("No disk at position %d!", disk_pos);
502 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
503 			G_RAID_DEBUG1(1, sc, "More than one disk for pos %d",
504 			    disk_pos);
505 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
506 			return (0);
507 		}
508 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
509 	}
510 
511 	/* Replace failed disk or placeholder with new disk. */
512 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
513 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
514 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
515 		sd->sd_disk = disk;
516 	}
517 	oldpd->pd_disk_pos = -2;
518 	pd->pd_disk_pos = disk_pos;
519 
520 	/* If it was placeholder -- destroy it. */
521 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
522 		g_raid_destroy_disk(olddisk);
523 	} else {
524 		/* Otherwise, make it STALE_FAILED. */
525 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
526 	}
527 
528 	/* Welcome the new disk. */
529 	if (resurrection)
530 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
531 	else// if (pd->pd_meta->disk_status == NVIDIA_S_CURRENT ||
532 	    //pd->pd_meta->disk_status == NVIDIA_S_REBUILD)
533 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
534 //	else
535 //		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
536 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
537 		/*
538 		 * Different disks may have different sizes,
539 		 * in concat mode. Update from real disk size.
540 		 */
541 		if (meta->type == NVIDIA_T_CONCAT)
542 			sd->sd_size = pd->pd_disk_size - 0x800 * 512;
543 
544 		if (resurrection) {
545 			/* New or ex-spare disk. */
546 			g_raid_change_subdisk_state(sd,
547 			    G_RAID_SUBDISK_S_NEW);
548 		} else if (meta->state == NVIDIA_S_REBUILD &&
549 		    (pd->pd_meta->disk_status & 0x100)) {
550 			/* Rebuilding disk. */
551 			g_raid_change_subdisk_state(sd,
552 			    G_RAID_SUBDISK_S_REBUILD);
553 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
554 			    meta->array_width * pd->pd_meta->sector_size;
555 		} else if (meta->state == NVIDIA_S_SYNC) {
556 			/* Resyncing/dirty disk. */
557 			g_raid_change_subdisk_state(sd,
558 			    G_RAID_SUBDISK_S_RESYNC);
559 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
560 			    meta->array_width * pd->pd_meta->sector_size;
561 		} else {
562 			/* Up to date disk. */
563 			g_raid_change_subdisk_state(sd,
564 			    G_RAID_SUBDISK_S_ACTIVE);
565 		}
566 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
567 		    G_RAID_EVENT_SUBDISK);
568 	}
569 
570 	/* Update status of our need for spare. */
571 	if (mdi->mdio_started) {
572 		mdi->mdio_incomplete =
573 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
574 		     mdi->mdio_total_disks);
575 	}
576 
577 	return (resurrection);
578 }
579 
580 static void
581 g_disk_md_nvidia_retaste(void *arg, int pending)
582 {
583 
584 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
585 	g_retaste(&g_raid_class);
586 	free(arg, M_MD_NVIDIA);
587 }
588 
589 static void
590 g_raid_md_nvidia_refill(struct g_raid_softc *sc)
591 {
592 	struct g_raid_md_object *md;
593 	struct g_raid_md_nvidia_object *mdi;
594 	struct g_raid_disk *disk;
595 	struct task *task;
596 	int update, na;
597 
598 	md = sc->sc_md;
599 	mdi = (struct g_raid_md_nvidia_object *)md;
600 	update = 0;
601 	do {
602 		/* Make sure we miss anything. */
603 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
604 		if (na == mdi->mdio_total_disks)
605 			break;
606 
607 		G_RAID_DEBUG1(1, md->mdo_softc,
608 		    "Array is not complete (%d of %d), "
609 		    "trying to refill.", na, mdi->mdio_total_disks);
610 
611 		/* Try to get use some of STALE disks. */
612 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
613 			if (disk->d_state == G_RAID_DISK_S_STALE) {
614 				update += g_raid_md_nvidia_start_disk(disk);
615 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
616 					break;
617 			}
618 		}
619 		if (disk != NULL)
620 			continue;
621 
622 		/* Try to get use some of SPARE disks. */
623 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
624 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
625 				update += g_raid_md_nvidia_start_disk(disk);
626 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
627 					break;
628 			}
629 		}
630 	} while (disk != NULL);
631 
632 	/* Write new metadata if we changed something. */
633 	if (update)
634 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
635 
636 	/* Update status of our need for spare. */
637 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
638 	    mdi->mdio_total_disks);
639 
640 	/* Request retaste hoping to find spare. */
641 	if (mdi->mdio_incomplete) {
642 		task = malloc(sizeof(struct task),
643 		    M_MD_NVIDIA, M_WAITOK | M_ZERO);
644 		TASK_INIT(task, 0, g_disk_md_nvidia_retaste, task);
645 		taskqueue_enqueue(taskqueue_swi, task);
646 	}
647 }
648 
649 static void
650 g_raid_md_nvidia_start(struct g_raid_softc *sc)
651 {
652 	struct g_raid_md_object *md;
653 	struct g_raid_md_nvidia_object *mdi;
654 	struct g_raid_md_nvidia_perdisk *pd;
655 	struct nvidia_raid_conf *meta;
656 	struct g_raid_volume *vol;
657 	struct g_raid_subdisk *sd;
658 	struct g_raid_disk *disk;
659 	off_t size;
660 	int j, disk_pos;
661 	char buf[17];
662 
663 	md = sc->sc_md;
664 	mdi = (struct g_raid_md_nvidia_object *)md;
665 	meta = mdi->mdio_meta;
666 
667 	/* Create volumes and subdisks. */
668 	nvidia_meta_get_name(meta, buf);
669 	vol = g_raid_create_volume(sc, buf, -1);
670 	vol->v_mediasize = (off_t)meta->total_sectors * 512;
671 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
672 	if (meta->type == NVIDIA_T_RAID0) {
673 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
674 		size = vol->v_mediasize / mdi->mdio_total_disks;
675 	} else if (meta->type == NVIDIA_T_RAID1) {
676 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
677 		size = vol->v_mediasize;
678 	} else if (meta->type == NVIDIA_T_RAID01) {
679 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
680 		size = vol->v_mediasize / (mdi->mdio_total_disks / 2);
681 	} else if (meta->type == NVIDIA_T_CONCAT) {
682 		if (mdi->mdio_total_disks == 1)
683 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
684 		else
685 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
686 		size = 0;
687 	} else if (meta->type == NVIDIA_T_RAID5) {
688 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
689 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
690 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
691 	} else if (meta->type == NVIDIA_T_RAID5_SYM) {
692 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
693 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LS;
694 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
695 	} else {
696 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
697 		size = 0;
698 	}
699 	vol->v_strip_size = meta->strip_sectors * 512; //ZZZ
700 	vol->v_disks_count = mdi->mdio_total_disks;
701 	vol->v_sectorsize = 512; //ZZZ
702 	for (j = 0; j < vol->v_disks_count; j++) {
703 		sd = &vol->v_subdisks[j];
704 		sd->sd_offset = 0;
705 		sd->sd_size = size;
706 	}
707 	g_raid_start_volume(vol);
708 
709 	/* Create disk placeholders to store data for later writing. */
710 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
711 		pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
712 		pd->pd_disk_pos = disk_pos;
713 		disk = g_raid_create_disk(sc);
714 		disk->d_md_data = (void *)pd;
715 		disk->d_state = G_RAID_DISK_S_OFFLINE;
716 		sd = &vol->v_subdisks[disk_pos];
717 		sd->sd_disk = disk;
718 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
719 	}
720 
721 	/* Make all disks found till the moment take their places. */
722 	do {
723 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
724 			if (disk->d_state == G_RAID_DISK_S_NONE) {
725 				g_raid_md_nvidia_start_disk(disk);
726 				break;
727 			}
728 		}
729 	} while (disk != NULL);
730 
731 	mdi->mdio_started = 1;
732 	G_RAID_DEBUG1(0, sc, "Array started.");
733 	g_raid_md_write_nvidia(md, NULL, NULL, NULL);
734 
735 	/* Pickup any STALE/SPARE disks to refill array if needed. */
736 	g_raid_md_nvidia_refill(sc);
737 
738 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
739 
740 	callout_stop(&mdi->mdio_start_co);
741 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
742 	root_mount_rel(mdi->mdio_rootmount);
743 	mdi->mdio_rootmount = NULL;
744 }
745 
746 static void
747 g_raid_md_nvidia_new_disk(struct g_raid_disk *disk)
748 {
749 	struct g_raid_softc *sc;
750 	struct g_raid_md_object *md;
751 	struct g_raid_md_nvidia_object *mdi;
752 	struct nvidia_raid_conf *pdmeta;
753 	struct g_raid_md_nvidia_perdisk *pd;
754 
755 	sc = disk->d_softc;
756 	md = sc->sc_md;
757 	mdi = (struct g_raid_md_nvidia_object *)md;
758 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
759 	pdmeta = pd->pd_meta;
760 
761 	if (mdi->mdio_started) {
762 		if (g_raid_md_nvidia_start_disk(disk))
763 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
764 	} else {
765 		if (mdi->mdio_meta == NULL ||
766 		    mdi->mdio_meta->disk_number >= mdi->mdio_meta->total_disks) {
767 			G_RAID_DEBUG1(1, sc, "Newer disk");
768 			if (mdi->mdio_meta != NULL)
769 				free(mdi->mdio_meta, M_MD_NVIDIA);
770 			mdi->mdio_meta = nvidia_meta_copy(pdmeta);
771 			mdi->mdio_total_disks = pdmeta->total_disks;
772 			mdi->mdio_disks_present = 1;
773 		} else if (pdmeta->disk_number < mdi->mdio_meta->total_disks) {
774 			mdi->mdio_disks_present++;
775 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
776 			    mdi->mdio_disks_present,
777 			    mdi->mdio_total_disks);
778 		} else
779 			G_RAID_DEBUG1(1, sc, "Spare disk");
780 
781 		/* If we collected all needed disks - start array. */
782 		if (mdi->mdio_disks_present == mdi->mdio_total_disks)
783 			g_raid_md_nvidia_start(sc);
784 	}
785 }
786 
787 static void
788 g_raid_nvidia_go(void *arg)
789 {
790 	struct g_raid_softc *sc;
791 	struct g_raid_md_object *md;
792 	struct g_raid_md_nvidia_object *mdi;
793 
794 	sc = arg;
795 	md = sc->sc_md;
796 	mdi = (struct g_raid_md_nvidia_object *)md;
797 	if (!mdi->mdio_started) {
798 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
799 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
800 	}
801 }
802 
803 static int
804 g_raid_md_create_nvidia(struct g_raid_md_object *md, struct g_class *mp,
805     struct g_geom **gp)
806 {
807 	struct g_raid_softc *sc;
808 	struct g_raid_md_nvidia_object *mdi;
809 	char name[32];
810 
811 	mdi = (struct g_raid_md_nvidia_object *)md;
812 	arc4rand(&mdi->mdio_volume_id, 16, 0);
813 	snprintf(name, sizeof(name), "NVIDIA-%d",
814 	    atomic_fetchadd_int(&NVIDIANodeID, 1));
815 	sc = g_raid_create_node(mp, name, md);
816 	if (sc == NULL)
817 		return (G_RAID_MD_TASTE_FAIL);
818 	md->mdo_softc = sc;
819 	*gp = sc->sc_geom;
820 	return (G_RAID_MD_TASTE_NEW);
821 }
822 
823 static int
824 g_raid_md_taste_nvidia(struct g_raid_md_object *md, struct g_class *mp,
825                               struct g_consumer *cp, struct g_geom **gp)
826 {
827 	struct g_consumer *rcp;
828 	struct g_provider *pp;
829 	struct g_raid_md_nvidia_object *mdi, *mdi1;
830 	struct g_raid_softc *sc;
831 	struct g_raid_disk *disk;
832 	struct nvidia_raid_conf *meta;
833 	struct g_raid_md_nvidia_perdisk *pd;
834 	struct g_geom *geom;
835 	int result, spare, len;
836 	char name[32];
837 	uint16_t vendor;
838 
839 	G_RAID_DEBUG(1, "Tasting NVIDIA on %s", cp->provider->name);
840 	mdi = (struct g_raid_md_nvidia_object *)md;
841 	pp = cp->provider;
842 
843 	/* Read metadata from device. */
844 	meta = NULL;
845 	g_topology_unlock();
846 	vendor = 0xffff;
847 	len = sizeof(vendor);
848 	if (pp->geom->rank == 1)
849 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
850 	meta = nvidia_meta_read(cp);
851 	g_topology_lock();
852 	if (meta == NULL) {
853 		if (g_raid_aggressive_spare) {
854 			if (vendor == 0x10de) {
855 				G_RAID_DEBUG(1,
856 				    "No NVIDIA metadata, forcing spare.");
857 				spare = 2;
858 				goto search;
859 			} else {
860 				G_RAID_DEBUG(1,
861 				    "NVIDIA vendor mismatch 0x%04x != 0x10de",
862 				    vendor);
863 			}
864 		}
865 		return (G_RAID_MD_TASTE_FAIL);
866 	}
867 
868 	/* Metadata valid. Print it. */
869 	g_raid_md_nvidia_print(meta);
870 	G_RAID_DEBUG(1, "NVIDIA disk position %d", meta->disk_number);
871 	spare = 0;//(meta->type == NVIDIA_T_SPARE) ? 1 : 0;
872 
873 search:
874 	/* Search for matching node. */
875 	sc = NULL;
876 	mdi1 = NULL;
877 	LIST_FOREACH(geom, &mp->geom, geom) {
878 		sc = geom->softc;
879 		if (sc == NULL)
880 			continue;
881 		if (sc->sc_stopping != 0)
882 			continue;
883 		if (sc->sc_md->mdo_class != md->mdo_class)
884 			continue;
885 		mdi1 = (struct g_raid_md_nvidia_object *)sc->sc_md;
886 		if (spare) {
887 			if (mdi1->mdio_incomplete)
888 				break;
889 		} else {
890 			if (memcmp(&mdi1->mdio_volume_id,
891 			     &meta->volume_id, 16) == 0)
892 				break;
893 		}
894 	}
895 
896 	/* Found matching node. */
897 	if (geom != NULL) {
898 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
899 		result = G_RAID_MD_TASTE_EXISTING;
900 
901 	} else if (spare) { /* Not found needy node -- left for later. */
902 		G_RAID_DEBUG(1, "Spare is not needed at this time");
903 		goto fail1;
904 
905 	} else { /* Not found matching node -- create one. */
906 		result = G_RAID_MD_TASTE_NEW;
907 		memcpy(&mdi->mdio_volume_id, &meta->volume_id, 16);
908 		snprintf(name, sizeof(name), "NVIDIA-%d",
909 		    atomic_fetchadd_int(&NVIDIANodeID, 1));
910 		sc = g_raid_create_node(mp, name, md);
911 		md->mdo_softc = sc;
912 		geom = sc->sc_geom;
913 		callout_init(&mdi->mdio_start_co, 1);
914 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
915 		    g_raid_nvidia_go, sc);
916 		mdi->mdio_rootmount = root_mount_hold("GRAID-NVIDIA");
917 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
918 	}
919 
920 	/* There is no return after this point, so we close passed consumer. */
921 	g_access(cp, -1, 0, 0);
922 
923 	rcp = g_new_consumer(geom);
924 	rcp->flags |= G_CF_DIRECT_RECEIVE;
925 	g_attach(rcp, pp);
926 	if (g_access(rcp, 1, 1, 1) != 0)
927 		; //goto fail1;
928 
929 	g_topology_unlock();
930 	sx_xlock(&sc->sc_lock);
931 
932 	pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
933 	pd->pd_meta = meta;
934 	if (spare == 2) {
935 		pd->pd_disk_pos = -3;
936 	} else {
937 		pd->pd_disk_pos = -1;
938 	}
939 	pd->pd_disk_size = pp->mediasize;
940 	disk = g_raid_create_disk(sc);
941 	disk->d_md_data = (void *)pd;
942 	disk->d_consumer = rcp;
943 	rcp->private = disk;
944 
945 	g_raid_get_disk_info(disk);
946 
947 	g_raid_md_nvidia_new_disk(disk);
948 
949 	sx_xunlock(&sc->sc_lock);
950 	g_topology_lock();
951 	*gp = geom;
952 	return (result);
953 fail1:
954 	free(meta, M_MD_NVIDIA);
955 	return (G_RAID_MD_TASTE_FAIL);
956 }
957 
958 static int
959 g_raid_md_event_nvidia(struct g_raid_md_object *md,
960     struct g_raid_disk *disk, u_int event)
961 {
962 	struct g_raid_softc *sc;
963 	struct g_raid_subdisk *sd;
964 	struct g_raid_md_nvidia_object *mdi;
965 	struct g_raid_md_nvidia_perdisk *pd;
966 
967 	sc = md->mdo_softc;
968 	mdi = (struct g_raid_md_nvidia_object *)md;
969 	if (disk == NULL) {
970 		switch (event) {
971 		case G_RAID_NODE_E_START:
972 			if (!mdi->mdio_started) {
973 				/* Bump volume ID to drop missing disks. */
974 				arc4rand(&mdi->mdio_volume_id, 16, 0);
975 				g_raid_md_nvidia_start(sc);
976 			}
977 			return (0);
978 		}
979 		return (-1);
980 	}
981 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
982 	switch (event) {
983 	case G_RAID_DISK_E_DISCONNECTED:
984 		/* If disk was assigned, just update statuses. */
985 		if (pd->pd_disk_pos >= 0) {
986 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
987 			if (disk->d_consumer) {
988 				g_raid_kill_consumer(sc, disk->d_consumer);
989 				disk->d_consumer = NULL;
990 			}
991 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
992 				g_raid_change_subdisk_state(sd,
993 				    G_RAID_SUBDISK_S_NONE);
994 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
995 				    G_RAID_EVENT_SUBDISK);
996 			}
997 		} else {
998 			/* Otherwise -- delete. */
999 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1000 			g_raid_destroy_disk(disk);
1001 		}
1002 
1003 		if (mdi->mdio_started) {
1004 			/* Bump volume ID to prevent disk resurrection. */
1005 			if (pd->pd_disk_pos >= 0)
1006 				arc4rand(&mdi->mdio_volume_id, 16, 0);
1007 
1008 			/* Write updated metadata to all disks. */
1009 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1010 		}
1011 
1012 		/* Check if anything left except placeholders. */
1013 		if (g_raid_ndisks(sc, -1) ==
1014 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1015 			g_raid_destroy_node(sc, 0);
1016 		else
1017 			g_raid_md_nvidia_refill(sc);
1018 		return (0);
1019 	}
1020 	return (-2);
1021 }
1022 
1023 static int
1024 g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
1025     struct gctl_req *req)
1026 {
1027 	struct g_raid_softc *sc;
1028 	struct g_raid_volume *vol;
1029 	struct g_raid_subdisk *sd;
1030 	struct g_raid_disk *disk;
1031 	struct g_raid_md_nvidia_object *mdi;
1032 	struct g_raid_md_nvidia_perdisk *pd;
1033 	struct g_consumer *cp;
1034 	struct g_provider *pp;
1035 	char arg[16];
1036 	const char *verb, *volname, *levelname, *diskname;
1037 	int *nargs, *force;
1038 	off_t size, sectorsize, strip, volsize;
1039 	intmax_t *sizearg, *striparg;
1040 	int numdisks, i, len, level, qual, update;
1041 	int error;
1042 
1043 	sc = md->mdo_softc;
1044 	mdi = (struct g_raid_md_nvidia_object *)md;
1045 	verb = gctl_get_param(req, "verb", NULL);
1046 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1047 	error = 0;
1048 	if (strcmp(verb, "label") == 0) {
1049 		if (*nargs < 4) {
1050 			gctl_error(req, "Invalid number of arguments.");
1051 			return (-1);
1052 		}
1053 		volname = gctl_get_asciiparam(req, "arg1");
1054 		if (volname == NULL) {
1055 			gctl_error(req, "No volume name.");
1056 			return (-2);
1057 		}
1058 		levelname = gctl_get_asciiparam(req, "arg2");
1059 		if (levelname == NULL) {
1060 			gctl_error(req, "No RAID level.");
1061 			return (-3);
1062 		}
1063 		if (strcasecmp(levelname, "RAID5") == 0)
1064 			levelname = "RAID5-LS";
1065 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1066 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1067 			return (-4);
1068 		}
1069 		numdisks = *nargs - 3;
1070 		force = gctl_get_paraml(req, "force", sizeof(*force));
1071 		if (!g_raid_md_nvidia_supported(level, qual, numdisks,
1072 		    force ? *force : 0)) {
1073 			gctl_error(req, "Unsupported RAID level "
1074 			    "(0x%02x/0x%02x), or number of disks (%d).",
1075 			    level, qual, numdisks);
1076 			return (-5);
1077 		}
1078 
1079 		/* Search for disks, connect them and probe. */
1080 		size = 0x7fffffffffffffffllu;
1081 		sectorsize = 0;
1082 		for (i = 0; i < numdisks; i++) {
1083 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1084 			diskname = gctl_get_asciiparam(req, arg);
1085 			if (diskname == NULL) {
1086 				gctl_error(req, "No disk name (%s).", arg);
1087 				error = -6;
1088 				break;
1089 			}
1090 			if (strcmp(diskname, "NONE") == 0) {
1091 				cp = NULL;
1092 				pp = NULL;
1093 			} else {
1094 				g_topology_lock();
1095 				cp = g_raid_open_consumer(sc, diskname);
1096 				if (cp == NULL) {
1097 					gctl_error(req, "Can't open '%s'.",
1098 					    diskname);
1099 					g_topology_unlock();
1100 					error = -7;
1101 					break;
1102 				}
1103 				pp = cp->provider;
1104 			}
1105 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1106 			pd->pd_disk_pos = i;
1107 			disk = g_raid_create_disk(sc);
1108 			disk->d_md_data = (void *)pd;
1109 			disk->d_consumer = cp;
1110 			if (cp == NULL)
1111 				continue;
1112 			cp->private = disk;
1113 			g_topology_unlock();
1114 
1115 			g_raid_get_disk_info(disk);
1116 
1117 			pd->pd_disk_size = pp->mediasize;
1118 			if (size > pp->mediasize)
1119 				size = pp->mediasize;
1120 			if (sectorsize < pp->sectorsize)
1121 				sectorsize = pp->sectorsize;
1122 		}
1123 		if (error != 0)
1124 			return (error);
1125 
1126 		if (sectorsize <= 0) {
1127 			gctl_error(req, "Can't get sector size.");
1128 			return (-8);
1129 		}
1130 
1131 		/* Reserve space for metadata. */
1132 		size -= 2 * sectorsize;
1133 
1134 		/* Handle size argument. */
1135 		len = sizeof(*sizearg);
1136 		sizearg = gctl_get_param(req, "size", &len);
1137 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1138 		    *sizearg > 0) {
1139 			if (*sizearg > size) {
1140 				gctl_error(req, "Size too big %lld > %lld.",
1141 				    (long long)*sizearg, (long long)size);
1142 				return (-9);
1143 			}
1144 			size = *sizearg;
1145 		}
1146 
1147 		/* Handle strip argument. */
1148 		strip = 131072;
1149 		len = sizeof(*striparg);
1150 		striparg = gctl_get_param(req, "strip", &len);
1151 		if (striparg != NULL && len == sizeof(*striparg) &&
1152 		    *striparg > 0) {
1153 			if (*striparg < sectorsize) {
1154 				gctl_error(req, "Strip size too small.");
1155 				return (-10);
1156 			}
1157 			if (*striparg % sectorsize != 0) {
1158 				gctl_error(req, "Incorrect strip size.");
1159 				return (-11);
1160 			}
1161 			if (strip > 65535 * sectorsize) {
1162 				gctl_error(req, "Strip size too big.");
1163 				return (-12);
1164 			}
1165 			strip = *striparg;
1166 		}
1167 
1168 		/* Round size down to strip or sector. */
1169 		if (level == G_RAID_VOLUME_RL_RAID1)
1170 			size -= (size % sectorsize);
1171 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1172 		    (numdisks & 1) != 0)
1173 			size -= (size % (2 * strip));
1174 		else
1175 			size -= (size % strip);
1176 		if (size <= 0) {
1177 			gctl_error(req, "Size too small.");
1178 			return (-13);
1179 		}
1180 
1181 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1182 		    level == G_RAID_VOLUME_RL_CONCAT ||
1183 		    level == G_RAID_VOLUME_RL_SINGLE)
1184 			volsize = size * numdisks;
1185 		else if (level == G_RAID_VOLUME_RL_RAID1)
1186 			volsize = size;
1187 		else if (level == G_RAID_VOLUME_RL_RAID5)
1188 			volsize = size * (numdisks - 1);
1189 		else { /* RAID1E */
1190 			volsize = ((size * numdisks) / strip / 2) *
1191 			    strip;
1192 		}
1193 		if (volsize > 0xffffffffllu * sectorsize) {
1194 			gctl_error(req, "Size too big.");
1195 			return (-14);
1196 		}
1197 
1198 		/* We have all we need, create things: volume, ... */
1199 		mdi->mdio_total_disks = numdisks;
1200 		mdi->mdio_started = 1;
1201 		vol = g_raid_create_volume(sc, volname, -1);
1202 		vol->v_md_data = (void *)(intptr_t)0;
1203 		vol->v_raid_level = level;
1204 		vol->v_raid_level_qualifier = qual;
1205 		vol->v_strip_size = strip;
1206 		vol->v_disks_count = numdisks;
1207 		vol->v_mediasize = volsize;
1208 		vol->v_sectorsize = sectorsize;
1209 		g_raid_start_volume(vol);
1210 
1211 		/* , and subdisks. */
1212 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1213 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1214 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1215 			sd->sd_disk = disk;
1216 			sd->sd_offset = 0;
1217 			sd->sd_size = size;
1218 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1219 			if (sd->sd_disk->d_consumer != NULL) {
1220 				g_raid_change_disk_state(disk,
1221 				    G_RAID_DISK_S_ACTIVE);
1222 				g_raid_change_subdisk_state(sd,
1223 				    G_RAID_SUBDISK_S_ACTIVE);
1224 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1225 				    G_RAID_EVENT_SUBDISK);
1226 			} else {
1227 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1228 			}
1229 		}
1230 
1231 		/* Write metadata based on created entities. */
1232 		G_RAID_DEBUG1(0, sc, "Array started.");
1233 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1234 
1235 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1236 		g_raid_md_nvidia_refill(sc);
1237 
1238 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1239 		    G_RAID_EVENT_VOLUME);
1240 		return (0);
1241 	}
1242 	if (strcmp(verb, "delete") == 0) {
1243 		/* Check if some volume is still open. */
1244 		force = gctl_get_paraml(req, "force", sizeof(*force));
1245 		if (force != NULL && *force == 0 &&
1246 		    g_raid_nopens(sc) != 0) {
1247 			gctl_error(req, "Some volume is still open.");
1248 			return (-4);
1249 		}
1250 
1251 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1252 			if (disk->d_consumer)
1253 				nvidia_meta_erase(disk->d_consumer);
1254 		}
1255 		g_raid_destroy_node(sc, 0);
1256 		return (0);
1257 	}
1258 	if (strcmp(verb, "remove") == 0 ||
1259 	    strcmp(verb, "fail") == 0) {
1260 		if (*nargs < 2) {
1261 			gctl_error(req, "Invalid number of arguments.");
1262 			return (-1);
1263 		}
1264 		for (i = 1; i < *nargs; i++) {
1265 			snprintf(arg, sizeof(arg), "arg%d", i);
1266 			diskname = gctl_get_asciiparam(req, arg);
1267 			if (diskname == NULL) {
1268 				gctl_error(req, "No disk name (%s).", arg);
1269 				error = -2;
1270 				break;
1271 			}
1272 			if (strncmp(diskname, _PATH_DEV, 5) == 0)
1273 				diskname += 5;
1274 
1275 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1276 				if (disk->d_consumer != NULL &&
1277 				    disk->d_consumer->provider != NULL &&
1278 				    strcmp(disk->d_consumer->provider->name,
1279 				     diskname) == 0)
1280 					break;
1281 			}
1282 			if (disk == NULL) {
1283 				gctl_error(req, "Disk '%s' not found.",
1284 				    diskname);
1285 				error = -3;
1286 				break;
1287 			}
1288 
1289 			if (strcmp(verb, "fail") == 0) {
1290 				g_raid_md_fail_disk_nvidia(md, NULL, disk);
1291 				continue;
1292 			}
1293 
1294 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1295 
1296 			/* Erase metadata on deleting disk. */
1297 			nvidia_meta_erase(disk->d_consumer);
1298 
1299 			/* If disk was assigned, just update statuses. */
1300 			if (pd->pd_disk_pos >= 0) {
1301 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1302 				g_raid_kill_consumer(sc, disk->d_consumer);
1303 				disk->d_consumer = NULL;
1304 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1305 					g_raid_change_subdisk_state(sd,
1306 					    G_RAID_SUBDISK_S_NONE);
1307 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1308 					    G_RAID_EVENT_SUBDISK);
1309 				}
1310 			} else {
1311 				/* Otherwise -- delete. */
1312 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1313 				g_raid_destroy_disk(disk);
1314 			}
1315 		}
1316 
1317 		/* Write updated metadata to remaining disks. */
1318 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1319 
1320 		/* Check if anything left except placeholders. */
1321 		if (g_raid_ndisks(sc, -1) ==
1322 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1323 			g_raid_destroy_node(sc, 0);
1324 		else
1325 			g_raid_md_nvidia_refill(sc);
1326 		return (error);
1327 	}
1328 	if (strcmp(verb, "insert") == 0) {
1329 		if (*nargs < 2) {
1330 			gctl_error(req, "Invalid number of arguments.");
1331 			return (-1);
1332 		}
1333 		update = 0;
1334 		for (i = 1; i < *nargs; i++) {
1335 			/* Get disk name. */
1336 			snprintf(arg, sizeof(arg), "arg%d", i);
1337 			diskname = gctl_get_asciiparam(req, arg);
1338 			if (diskname == NULL) {
1339 				gctl_error(req, "No disk name (%s).", arg);
1340 				error = -3;
1341 				break;
1342 			}
1343 
1344 			/* Try to find provider with specified name. */
1345 			g_topology_lock();
1346 			cp = g_raid_open_consumer(sc, diskname);
1347 			if (cp == NULL) {
1348 				gctl_error(req, "Can't open disk '%s'.",
1349 				    diskname);
1350 				g_topology_unlock();
1351 				error = -4;
1352 				break;
1353 			}
1354 			pp = cp->provider;
1355 
1356 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1357 			pd->pd_disk_pos = -3;
1358 			pd->pd_disk_size = pp->mediasize;
1359 
1360 			disk = g_raid_create_disk(sc);
1361 			disk->d_consumer = cp;
1362 			disk->d_md_data = (void *)pd;
1363 			cp->private = disk;
1364 			g_topology_unlock();
1365 
1366 			g_raid_get_disk_info(disk);
1367 
1368 			/* Welcome the "new" disk. */
1369 			update += g_raid_md_nvidia_start_disk(disk);
1370 			if (disk->d_state != G_RAID_DISK_S_SPARE &&
1371 			    disk->d_state != G_RAID_DISK_S_ACTIVE) {
1372 				gctl_error(req, "Disk '%s' doesn't fit.",
1373 				    diskname);
1374 				g_raid_destroy_disk(disk);
1375 				error = -8;
1376 				break;
1377 			}
1378 		}
1379 
1380 		/* Write new metadata if we changed something. */
1381 		if (update)
1382 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1383 		return (error);
1384 	}
1385 	gctl_error(req, "Command '%s' is not supported.", verb);
1386 	return (-100);
1387 }
1388 
1389 static int
1390 g_raid_md_write_nvidia(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1391     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1392 {
1393 	struct g_raid_softc *sc;
1394 	struct g_raid_volume *vol;
1395 	struct g_raid_subdisk *sd;
1396 	struct g_raid_disk *disk;
1397 	struct g_raid_md_nvidia_object *mdi;
1398 	struct g_raid_md_nvidia_perdisk *pd;
1399 	struct nvidia_raid_conf *meta;
1400 	int i, spares;
1401 
1402 	sc = md->mdo_softc;
1403 	mdi = (struct g_raid_md_nvidia_object *)md;
1404 
1405 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1406 		return (0);
1407 
1408 	/* There is only one volume. */
1409 	vol = TAILQ_FIRST(&sc->sc_volumes);
1410 
1411 	/* Fill global fields. */
1412 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1413 	if (mdi->mdio_meta)
1414 		memcpy(meta, mdi->mdio_meta, sizeof(*meta));
1415 	memcpy(meta->nvidia_id, NVIDIA_MAGIC, sizeof(NVIDIA_MAGIC) - 1);
1416 	meta->config_size = 30;
1417 	meta->version = 0x0064;
1418 	meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1419 	meta->sector_size = vol->v_sectorsize;
1420 	nvidia_meta_put_name(meta, vol->v_name);
1421 	meta->magic_0 = NVIDIA_MAGIC0;
1422 	memcpy(&meta->volume_id, &mdi->mdio_volume_id, 16);
1423 	meta->state = NVIDIA_S_IDLE;
1424 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1425 		meta->array_width = 1;
1426 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1427 		meta->array_width = vol->v_disks_count / 2;
1428 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1429 		meta->array_width = vol->v_disks_count - 1;
1430 	else
1431 		meta->array_width = vol->v_disks_count;
1432 	meta->total_disks = vol->v_disks_count;
1433 	meta->orig_array_width = meta->array_width;
1434 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
1435 		meta->type = NVIDIA_T_RAID0;
1436 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1437 		meta->type = NVIDIA_T_RAID1;
1438 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1439 		meta->type = NVIDIA_T_RAID01;
1440 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1441 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1442 		meta->type = NVIDIA_T_CONCAT;
1443 	else if (vol->v_raid_level_qualifier == G_RAID_VOLUME_RLQ_R5LA)
1444 		meta->type = NVIDIA_T_RAID5;
1445 	else
1446 		meta->type = NVIDIA_T_RAID5_SYM;
1447 	meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize;
1448 	meta->strip_bytes = vol->v_strip_size;
1449 	meta->strip_shift = ffs(meta->strip_sectors) - 1;
1450 	meta->strip_mask = meta->strip_sectors - 1;
1451 	meta->stripe_sectors = meta->strip_sectors * meta->orig_array_width;
1452 	meta->stripe_bytes = meta->stripe_sectors * vol->v_sectorsize;
1453 	meta->rebuild_lba = 0;
1454 	meta->orig_type = meta->type;
1455 	meta->orig_total_sectors = meta->total_sectors;
1456 	meta->status = 0;
1457 
1458 	for (i = 0; i < vol->v_disks_count; i++) {
1459 		sd = &vol->v_subdisks[i];
1460 		if ((sd->sd_state == G_RAID_SUBDISK_S_STALE ||
1461 		     sd->sd_state == G_RAID_SUBDISK_S_RESYNC ||
1462 		     vol->v_dirty) &&
1463 		     meta->state != NVIDIA_S_REBUILD)
1464 			meta->state = NVIDIA_S_SYNC;
1465 		else if (sd->sd_state == G_RAID_SUBDISK_S_NEW ||
1466 		     sd->sd_state == G_RAID_SUBDISK_S_REBUILD)
1467 			meta->state = NVIDIA_S_REBUILD;
1468 	}
1469 
1470 	/* We are done. Print meta data and store them to disks. */
1471 	if (mdi->mdio_meta != NULL)
1472 		free(mdi->mdio_meta, M_MD_NVIDIA);
1473 	mdi->mdio_meta = meta;
1474 	spares = 0;
1475 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1476 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1477 		if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1478 		    disk->d_state != G_RAID_DISK_S_SPARE)
1479 			continue;
1480 		if (pd->pd_meta != NULL) {
1481 			free(pd->pd_meta, M_MD_NVIDIA);
1482 			pd->pd_meta = NULL;
1483 		}
1484 		pd->pd_meta = nvidia_meta_copy(meta);
1485 		if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1486 			/* For RAID0+1 we need to translate order. */
1487 			pd->pd_meta->disk_number =
1488 			    nvidia_meta_translate_disk(meta, sd->sd_pos);
1489 			if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE) {
1490 				pd->pd_meta->disk_status = 0x100;
1491 				pd->pd_meta->rebuild_lba =
1492 				    sd->sd_rebuild_pos / vol->v_sectorsize *
1493 				    meta->array_width;
1494 			}
1495 		} else
1496 			pd->pd_meta->disk_number = meta->total_disks + spares++;
1497 		G_RAID_DEBUG(1, "Writing NVIDIA metadata to %s",
1498 		    g_raid_get_diskname(disk));
1499 		g_raid_md_nvidia_print(pd->pd_meta);
1500 		nvidia_meta_write(disk->d_consumer, pd->pd_meta);
1501 	}
1502 	return (0);
1503 }
1504 
1505 static int
1506 g_raid_md_fail_disk_nvidia(struct g_raid_md_object *md,
1507     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1508 {
1509 	struct g_raid_softc *sc;
1510 	struct g_raid_md_nvidia_perdisk *pd;
1511 	struct g_raid_subdisk *sd;
1512 
1513 	sc = md->mdo_softc;
1514 	pd = (struct g_raid_md_nvidia_perdisk *)tdisk->d_md_data;
1515 
1516 	/* We can't fail disk that is not a part of array now. */
1517 	if (pd->pd_disk_pos < 0)
1518 		return (-1);
1519 
1520 	/* Erase metadata to prevent disks's later resurrection. */
1521 	if (tdisk->d_consumer)
1522 		nvidia_meta_erase(tdisk->d_consumer);
1523 
1524 	/* Change states. */
1525 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1526 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1527 		g_raid_change_subdisk_state(sd,
1528 		    G_RAID_SUBDISK_S_FAILED);
1529 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1530 		    G_RAID_EVENT_SUBDISK);
1531 	}
1532 
1533 	/* Write updated metadata to remaining disks. */
1534 	g_raid_md_write_nvidia(md, NULL, NULL, tdisk);
1535 
1536 	/* Check if anything left except placeholders. */
1537 	if (g_raid_ndisks(sc, -1) ==
1538 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1539 		g_raid_destroy_node(sc, 0);
1540 	else
1541 		g_raid_md_nvidia_refill(sc);
1542 	return (0);
1543 }
1544 
1545 static int
1546 g_raid_md_free_disk_nvidia(struct g_raid_md_object *md,
1547     struct g_raid_disk *disk)
1548 {
1549 	struct g_raid_md_nvidia_perdisk *pd;
1550 
1551 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1552 	if (pd->pd_meta != NULL) {
1553 		free(pd->pd_meta, M_MD_NVIDIA);
1554 		pd->pd_meta = NULL;
1555 	}
1556 	free(pd, M_MD_NVIDIA);
1557 	disk->d_md_data = NULL;
1558 	return (0);
1559 }
1560 
1561 static int
1562 g_raid_md_free_nvidia(struct g_raid_md_object *md)
1563 {
1564 	struct g_raid_md_nvidia_object *mdi;
1565 
1566 	mdi = (struct g_raid_md_nvidia_object *)md;
1567 	if (!mdi->mdio_started) {
1568 		mdi->mdio_started = 0;
1569 		callout_stop(&mdi->mdio_start_co);
1570 		G_RAID_DEBUG1(1, md->mdo_softc,
1571 		    "root_mount_rel %p", mdi->mdio_rootmount);
1572 		root_mount_rel(mdi->mdio_rootmount);
1573 		mdi->mdio_rootmount = NULL;
1574 	}
1575 	if (mdi->mdio_meta != NULL) {
1576 		free(mdi->mdio_meta, M_MD_NVIDIA);
1577 		mdi->mdio_meta = NULL;
1578 	}
1579 	return (0);
1580 }
1581 
1582 G_RAID_MD_DECLARE(nvidia, "NVIDIA");
1583