xref: /freebsd/sys/geom/raid/md_nvidia.c (revision 2e1417489338b971e5fd599ff48b5f65df9e8d3b)
1 /*-
2  * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/endian.h>
33 #include <sys/kernel.h>
34 #include <sys/kobj.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/systm.h>
40 #include <sys/taskqueue.h>
41 #include <geom/geom.h>
42 #include "geom/raid/g_raid.h"
43 #include "g_raid_md_if.h"
44 
45 static MALLOC_DEFINE(M_MD_NVIDIA, "md_nvidia_data", "GEOM_RAID NVIDIA metadata");
46 
47 struct nvidia_raid_conf {
48 	uint8_t		nvidia_id[8];
49 #define NVIDIA_MAGIC                "NVIDIA  "
50 
51 	uint32_t	config_size;
52 	uint32_t	checksum;
53 	uint16_t	version;
54 	uint8_t		disk_number;
55 	uint8_t		dummy_0;
56 	uint32_t	total_sectors;
57 	uint32_t	sector_size;
58 	uint8_t		name[16];
59 	uint8_t		revision[4];
60 	uint32_t	disk_status;
61 
62 	uint32_t	magic_0;
63 #define NVIDIA_MAGIC0		0x00640044
64 
65 	uint64_t	volume_id[2];
66 	uint8_t		state;
67 #define NVIDIA_S_IDLE		0
68 #define NVIDIA_S_INIT		2
69 #define NVIDIA_S_REBUILD	3
70 #define NVIDIA_S_UPGRADE	4
71 #define NVIDIA_S_SYNC		5
72 	uint8_t		array_width;
73 	uint8_t		total_disks;
74 	uint8_t		orig_array_width;
75 	uint16_t	type;
76 #define NVIDIA_T_RAID0		0x0080
77 #define NVIDIA_T_RAID1		0x0081
78 #define NVIDIA_T_RAID3		0x0083
79 #define NVIDIA_T_RAID5		0x0085	/* RLQ = 00/02? */
80 #define NVIDIA_T_RAID5_SYM	0x0095	/* RLQ = 03 */
81 #define NVIDIA_T_RAID10		0x008a
82 #define NVIDIA_T_RAID01		0x8180
83 #define NVIDIA_T_CONCAT		0x00ff
84 
85 	uint16_t	dummy_3;
86 	uint32_t	strip_sectors;
87 	uint32_t	strip_bytes;
88 	uint32_t	strip_shift;
89 	uint32_t	strip_mask;
90 	uint32_t	stripe_sectors;
91 	uint32_t	stripe_bytes;
92 	uint32_t	rebuild_lba;
93 	uint32_t	orig_type;
94 	uint32_t	orig_total_sectors;
95 	uint32_t	status;
96 #define NVIDIA_S_BOOTABLE	0x00000001
97 #define NVIDIA_S_DEGRADED	0x00000002
98 
99 	uint32_t	filler[98];
100 } __packed;
101 
102 struct g_raid_md_nvidia_perdisk {
103 	struct nvidia_raid_conf	*pd_meta;
104 	int			 pd_disk_pos;
105 	off_t			 pd_disk_size;
106 };
107 
108 struct g_raid_md_nvidia_object {
109 	struct g_raid_md_object	 mdio_base;
110 	uint64_t		 mdio_volume_id[2];
111 	struct nvidia_raid_conf	*mdio_meta;
112 	struct callout		 mdio_start_co;	/* STARTING state timer. */
113 	int			 mdio_total_disks;
114 	int			 mdio_disks_present;
115 	int			 mdio_started;
116 	int			 mdio_incomplete;
117 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
118 };
119 
120 static g_raid_md_create_t g_raid_md_create_nvidia;
121 static g_raid_md_taste_t g_raid_md_taste_nvidia;
122 static g_raid_md_event_t g_raid_md_event_nvidia;
123 static g_raid_md_ctl_t g_raid_md_ctl_nvidia;
124 static g_raid_md_write_t g_raid_md_write_nvidia;
125 static g_raid_md_fail_disk_t g_raid_md_fail_disk_nvidia;
126 static g_raid_md_free_disk_t g_raid_md_free_disk_nvidia;
127 static g_raid_md_free_t g_raid_md_free_nvidia;
128 
129 static kobj_method_t g_raid_md_nvidia_methods[] = {
130 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_nvidia),
131 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_nvidia),
132 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_nvidia),
133 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_nvidia),
134 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_nvidia),
135 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_nvidia),
136 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_nvidia),
137 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_nvidia),
138 	{ 0, 0 }
139 };
140 
141 static struct g_raid_md_class g_raid_md_nvidia_class = {
142 	"NVIDIA",
143 	g_raid_md_nvidia_methods,
144 	sizeof(struct g_raid_md_nvidia_object),
145 	.mdc_priority = 100
146 };
147 
148 static int NVIDIANodeID = 1;
149 
150 static void
151 g_raid_md_nvidia_print(struct nvidia_raid_conf *meta)
152 {
153 
154 	if (g_raid_debug < 1)
155 		return;
156 
157 	printf("********* ATA NVIDIA RAID Metadata *********\n");
158 	printf("nvidia_id           <%.8s>\n", meta->nvidia_id);
159 	printf("config_size         %u\n", meta->config_size);
160 	printf("checksum            0x%08x\n", meta->checksum);
161 	printf("version             0x%04x\n", meta->version);
162 	printf("disk_number         %d\n", meta->disk_number);
163 	printf("dummy_0             0x%02x\n", meta->dummy_0);
164 	printf("total_sectors       %u\n", meta->total_sectors);
165 	printf("sector_size         %u\n", meta->sector_size);
166 	printf("name                <%.16s>\n", meta->name);
167 	printf("revision            0x%02x%02x%02x%02x\n",
168 	    meta->revision[0], meta->revision[1],
169 	    meta->revision[2], meta->revision[3]);
170 	printf("disk_status         0x%08x\n", meta->disk_status);
171 	printf("magic_0             0x%08x\n", meta->magic_0);
172 	printf("volume_id           0x%016jx%016jx\n",
173 	    meta->volume_id[1], meta->volume_id[0]);
174 	printf("state               0x%02x\n", meta->state);
175 	printf("array_width         %u\n", meta->array_width);
176 	printf("total_disks         %u\n", meta->total_disks);
177 	printf("orig_array_width    %u\n", meta->orig_array_width);
178 	printf("type                0x%04x\n", meta->type);
179 	printf("dummy_3             0x%04x\n", meta->dummy_3);
180 	printf("strip_sectors       %u\n", meta->strip_sectors);
181 	printf("strip_bytes         %u\n", meta->strip_bytes);
182 	printf("strip_shift         %u\n", meta->strip_shift);
183 	printf("strip_mask          0x%08x\n", meta->strip_mask);
184 	printf("stripe_sectors      %u\n", meta->stripe_sectors);
185 	printf("stripe_bytes        %u\n", meta->stripe_bytes);
186 	printf("rebuild_lba         %u\n", meta->rebuild_lba);
187 	printf("orig_type           0x%04x\n", meta->orig_type);
188 	printf("orig_total_sectors  %u\n", meta->orig_total_sectors);
189 	printf("status              0x%08x\n", meta->status);
190 	printf("=================================================\n");
191 }
192 
193 static struct nvidia_raid_conf *
194 nvidia_meta_copy(struct nvidia_raid_conf *meta)
195 {
196 	struct nvidia_raid_conf *nmeta;
197 
198 	nmeta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
199 	memcpy(nmeta, meta, sizeof(*meta));
200 	return (nmeta);
201 }
202 
203 static int
204 nvidia_meta_translate_disk(struct nvidia_raid_conf *meta, int md_disk_pos)
205 {
206 	int disk_pos;
207 
208 	if (md_disk_pos >= 0 && meta->type == NVIDIA_T_RAID01) {
209 		disk_pos = (md_disk_pos / meta->array_width) +
210 		    (md_disk_pos % meta->array_width) * meta->array_width;
211 	} else
212 		disk_pos = md_disk_pos;
213 	return (disk_pos);
214 }
215 
216 static void
217 nvidia_meta_get_name(struct nvidia_raid_conf *meta, char *buf)
218 {
219 	int i;
220 
221 	strncpy(buf, meta->name, 16);
222 	buf[16] = 0;
223 	for (i = 15; i >= 0; i--) {
224 		if (buf[i] > 0x20)
225 			break;
226 		buf[i] = 0;
227 	}
228 }
229 
230 static void
231 nvidia_meta_put_name(struct nvidia_raid_conf *meta, char *buf)
232 {
233 
234 	memset(meta->name, 0x20, 16);
235 	memcpy(meta->name, buf, MIN(strlen(buf), 16));
236 }
237 
238 static struct nvidia_raid_conf *
239 nvidia_meta_read(struct g_consumer *cp)
240 {
241 	struct g_provider *pp;
242 	struct nvidia_raid_conf *meta;
243 	char *buf;
244 	int error, i;
245 	uint32_t checksum, *ptr;
246 
247 	pp = cp->provider;
248 
249 	/* Read the anchor sector. */
250 	buf = g_read_data(cp,
251 	    pp->mediasize - 2 * pp->sectorsize, pp->sectorsize, &error);
252 	if (buf == NULL) {
253 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
254 		    pp->name, error);
255 		return (NULL);
256 	}
257 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
258 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
259 	g_free(buf);
260 
261 	/* Check if this is an NVIDIA RAID struct */
262 	if (strncmp(meta->nvidia_id, NVIDIA_MAGIC, strlen(NVIDIA_MAGIC))) {
263 		G_RAID_DEBUG(1, "NVIDIA signature check failed on %s", pp->name);
264 		free(meta, M_MD_NVIDIA);
265 		return (NULL);
266 	}
267 	if (meta->config_size > 128 ||
268 	    meta->config_size < 30) {
269 		G_RAID_DEBUG(1, "NVIDIA metadata size looks wrong: %d",
270 		    meta->config_size);
271 		free(meta, M_MD_NVIDIA);
272 		return (NULL);
273 	}
274 
275 	/* Check metadata checksum. */
276 	for (checksum = 0, ptr = (uint32_t *)meta,
277 	    i = 0; i < meta->config_size; i++)
278 		checksum += *ptr++;
279 	if (checksum != 0) {
280 		G_RAID_DEBUG(1, "NVIDIA checksum check failed on %s", pp->name);
281 		free(meta, M_MD_NVIDIA);
282 		return (NULL);
283 	}
284 
285 	/* Check volume state. */
286 	if (meta->state != NVIDIA_S_IDLE && meta->state != NVIDIA_S_INIT &&
287 	    meta->state != NVIDIA_S_REBUILD && meta->state != NVIDIA_S_SYNC) {
288 		G_RAID_DEBUG(1, "NVIDIA unknown state on %s (0x%02x)",
289 		    pp->name, meta->state);
290 		free(meta, M_MD_NVIDIA);
291 		return (NULL);
292 	}
293 
294 	/* Check raid type. */
295 	if (meta->type != NVIDIA_T_RAID0 && meta->type != NVIDIA_T_RAID1 &&
296 	    meta->type != NVIDIA_T_RAID3 && meta->type != NVIDIA_T_RAID5 &&
297 	    meta->type != NVIDIA_T_RAID5_SYM &&
298 	    meta->type != NVIDIA_T_RAID01 && meta->type != NVIDIA_T_CONCAT) {
299 		G_RAID_DEBUG(1, "NVIDIA unknown RAID level on %s (0x%02x)",
300 		    pp->name, meta->type);
301 		free(meta, M_MD_NVIDIA);
302 		return (NULL);
303 	}
304 
305 	return (meta);
306 }
307 
308 static int
309 nvidia_meta_write(struct g_consumer *cp, struct nvidia_raid_conf *meta)
310 {
311 	struct g_provider *pp;
312 	char *buf;
313 	int error, i;
314 	uint32_t checksum, *ptr;
315 
316 	pp = cp->provider;
317 
318 	/* Recalculate checksum for case if metadata were changed. */
319 	meta->checksum = 0;
320 	for (checksum = 0, ptr = (uint32_t *)meta,
321 	    i = 0; i < meta->config_size; i++)
322 		checksum += *ptr++;
323 	meta->checksum -= checksum;
324 
325 	/* Create and fill buffer. */
326 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
327 	memcpy(buf, meta, sizeof(*meta));
328 
329 	/* Write metadata. */
330 	error = g_write_data(cp,
331 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
332 	if (error != 0) {
333 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
334 		    pp->name, error);
335 	}
336 
337 	free(buf, M_MD_NVIDIA);
338 	return (error);
339 }
340 
341 static int
342 nvidia_meta_erase(struct g_consumer *cp)
343 {
344 	struct g_provider *pp;
345 	char *buf;
346 	int error;
347 
348 	pp = cp->provider;
349 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
350 	error = g_write_data(cp,
351 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
352 	if (error != 0) {
353 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
354 		    pp->name, error);
355 	}
356 	free(buf, M_MD_NVIDIA);
357 	return (error);
358 }
359 
360 static struct g_raid_disk *
361 g_raid_md_nvidia_get_disk(struct g_raid_softc *sc, int id)
362 {
363 	struct g_raid_disk	*disk;
364 	struct g_raid_md_nvidia_perdisk *pd;
365 
366 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
367 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
368 		if (pd->pd_disk_pos == id)
369 			break;
370 	}
371 	return (disk);
372 }
373 
374 static int
375 g_raid_md_nvidia_supported(int level, int qual, int disks, int force)
376 {
377 
378 	switch (level) {
379 	case G_RAID_VOLUME_RL_RAID0:
380 		if (disks < 1)
381 			return (0);
382 		if (!force && (disks < 2 || disks > 6))
383 			return (0);
384 		break;
385 	case G_RAID_VOLUME_RL_RAID1:
386 		if (disks < 1)
387 			return (0);
388 		if (!force && (disks != 2))
389 			return (0);
390 		break;
391 	case G_RAID_VOLUME_RL_RAID1E:
392 		if (disks < 2)
393 			return (0);
394 		if (disks % 2 != 0)
395 			return (0);
396 		if (!force && (disks < 4))
397 			return (0);
398 		break;
399 	case G_RAID_VOLUME_RL_SINGLE:
400 		if (disks != 1)
401 			return (0);
402 		break;
403 	case G_RAID_VOLUME_RL_CONCAT:
404 		if (disks < 2)
405 			return (0);
406 		break;
407 	case G_RAID_VOLUME_RL_RAID5:
408 		if (disks < 3)
409 			return (0);
410 		break;
411 	default:
412 		return (0);
413 	}
414 	if (qual != G_RAID_VOLUME_RLQ_NONE)
415 		return (0);
416 	return (1);
417 }
418 
419 static int
420 g_raid_md_nvidia_start_disk(struct g_raid_disk *disk)
421 {
422 	struct g_raid_softc *sc;
423 	struct g_raid_subdisk *sd, *tmpsd;
424 	struct g_raid_disk *olddisk, *tmpdisk;
425 	struct g_raid_md_object *md;
426 	struct g_raid_md_nvidia_object *mdi;
427 	struct g_raid_md_nvidia_perdisk *pd, *oldpd;
428 	struct nvidia_raid_conf *meta;
429 	int disk_pos, resurrection = 0;
430 
431 	sc = disk->d_softc;
432 	md = sc->sc_md;
433 	mdi = (struct g_raid_md_nvidia_object *)md;
434 	meta = mdi->mdio_meta;
435 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
436 	olddisk = NULL;
437 
438 	/* Find disk position in metadata by it's serial. */
439 	if (pd->pd_meta != NULL) {
440 		disk_pos = pd->pd_meta->disk_number;
441 		if (disk_pos >= meta->total_disks || mdi->mdio_started)
442 			disk_pos = -3;
443 	} else
444 		disk_pos = -3;
445 	/* For RAID0+1 we need to translate order. */
446 	disk_pos = nvidia_meta_translate_disk(meta, disk_pos);
447 	if (disk_pos < 0) {
448 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
449 		/* If we are in the start process, that's all for now. */
450 		if (!mdi->mdio_started)
451 			goto nofit;
452 		/*
453 		 * If we have already started - try to get use of the disk.
454 		 * Try to replace OFFLINE disks first, then FAILED.
455 		 */
456 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
457 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
458 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
459 				continue;
460 			/* Make sure this disk is big enough. */
461 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
462 				if (sd->sd_offset + sd->sd_size + 2 * 512 >
463 				    pd->pd_disk_size) {
464 					G_RAID_DEBUG1(1, sc,
465 					    "Disk too small (%ju < %ju)",
466 					    pd->pd_disk_size,
467 					    sd->sd_offset + sd->sd_size + 512);
468 					break;
469 				}
470 			}
471 			if (sd != NULL)
472 				continue;
473 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
474 				olddisk = tmpdisk;
475 				break;
476 			} else if (olddisk == NULL)
477 				olddisk = tmpdisk;
478 		}
479 		if (olddisk == NULL) {
480 nofit:
481 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
482 			return (1);
483 		}
484 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
485 		disk_pos = oldpd->pd_disk_pos;
486 		resurrection = 1;
487 	}
488 
489 	if (olddisk == NULL) {
490 		/* Find placeholder by position. */
491 		olddisk = g_raid_md_nvidia_get_disk(sc, disk_pos);
492 		if (olddisk == NULL)
493 			panic("No disk at position %d!", disk_pos);
494 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
495 			G_RAID_DEBUG1(1, sc, "More then one disk for pos %d",
496 			    disk_pos);
497 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
498 			return (0);
499 		}
500 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
501 	}
502 
503 	/* Replace failed disk or placeholder with new disk. */
504 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
505 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
506 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
507 		sd->sd_disk = disk;
508 	}
509 	oldpd->pd_disk_pos = -2;
510 	pd->pd_disk_pos = disk_pos;
511 
512 	/* If it was placeholder -- destroy it. */
513 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
514 		g_raid_destroy_disk(olddisk);
515 	} else {
516 		/* Otherwise, make it STALE_FAILED. */
517 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
518 	}
519 
520 	/* Welcome the new disk. */
521 	if (resurrection)
522 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
523 	else// if (pd->pd_meta->disk_status == NVIDIA_S_CURRENT ||
524 	    //pd->pd_meta->disk_status == NVIDIA_S_REBUILD)
525 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
526 //	else
527 //		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
528 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
529 
530 		/*
531 		 * Different disks may have different sizes,
532 		 * in concat mode. Update from real disk size.
533 		 */
534 		if (meta->type == NVIDIA_T_CONCAT)
535 			sd->sd_size = pd->pd_disk_size - 0x800 * 512;
536 
537 		if (resurrection) {
538 			/* New or ex-spare disk. */
539 			g_raid_change_subdisk_state(sd,
540 			    G_RAID_SUBDISK_S_NEW);
541 		} else if (meta->state == NVIDIA_S_REBUILD &&
542 		    (pd->pd_meta->disk_status & 0x100)) {
543 			/* Rebuilding disk. */
544 			g_raid_change_subdisk_state(sd,
545 			    G_RAID_SUBDISK_S_REBUILD);
546 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
547 			    meta->array_width * pd->pd_meta->sector_size;
548 		} else if (meta->state == NVIDIA_S_SYNC) {
549 			/* Resyncing/dirty disk. */
550 			g_raid_change_subdisk_state(sd,
551 			    G_RAID_SUBDISK_S_RESYNC);
552 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
553 			    meta->array_width * pd->pd_meta->sector_size;
554 		} else {
555 			/* Up to date disk. */
556 			g_raid_change_subdisk_state(sd,
557 			    G_RAID_SUBDISK_S_ACTIVE);
558 		}
559 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
560 		    G_RAID_EVENT_SUBDISK);
561 	}
562 
563 	/* Update status of our need for spare. */
564 	if (mdi->mdio_started) {
565 		mdi->mdio_incomplete =
566 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
567 		     mdi->mdio_total_disks);
568 	}
569 
570 	return (resurrection);
571 }
572 
573 static void
574 g_disk_md_nvidia_retaste(void *arg, int pending)
575 {
576 
577 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
578 	g_retaste(&g_raid_class);
579 	free(arg, M_MD_NVIDIA);
580 }
581 
582 static void
583 g_raid_md_nvidia_refill(struct g_raid_softc *sc)
584 {
585 	struct g_raid_md_object *md;
586 	struct g_raid_md_nvidia_object *mdi;
587 	struct g_raid_disk *disk;
588 	struct task *task;
589 	int update, na;
590 
591 	md = sc->sc_md;
592 	mdi = (struct g_raid_md_nvidia_object *)md;
593 	update = 0;
594 	do {
595 		/* Make sure we miss anything. */
596 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
597 		if (na == mdi->mdio_total_disks)
598 			break;
599 
600 		G_RAID_DEBUG1(1, md->mdo_softc,
601 		    "Array is not complete (%d of %d), "
602 		    "trying to refill.", na, mdi->mdio_total_disks);
603 
604 		/* Try to get use some of STALE disks. */
605 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
606 			if (disk->d_state == G_RAID_DISK_S_STALE) {
607 				update += g_raid_md_nvidia_start_disk(disk);
608 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
609 					break;
610 			}
611 		}
612 		if (disk != NULL)
613 			continue;
614 
615 		/* Try to get use some of SPARE disks. */
616 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
617 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
618 				update += g_raid_md_nvidia_start_disk(disk);
619 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
620 					break;
621 			}
622 		}
623 	} while (disk != NULL);
624 
625 	/* Write new metadata if we changed something. */
626 	if (update)
627 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
628 
629 	/* Update status of our need for spare. */
630 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
631 	    mdi->mdio_total_disks);
632 
633 	/* Request retaste hoping to find spare. */
634 	if (mdi->mdio_incomplete) {
635 		task = malloc(sizeof(struct task),
636 		    M_MD_NVIDIA, M_WAITOK | M_ZERO);
637 		TASK_INIT(task, 0, g_disk_md_nvidia_retaste, task);
638 		taskqueue_enqueue(taskqueue_swi, task);
639 	}
640 }
641 
642 static void
643 g_raid_md_nvidia_start(struct g_raid_softc *sc)
644 {
645 	struct g_raid_md_object *md;
646 	struct g_raid_md_nvidia_object *mdi;
647 	struct g_raid_md_nvidia_perdisk *pd;
648 	struct nvidia_raid_conf *meta;
649 	struct g_raid_volume *vol;
650 	struct g_raid_subdisk *sd;
651 	struct g_raid_disk *disk;
652 	off_t size;
653 	int j, disk_pos;
654 	char buf[17];
655 
656 	md = sc->sc_md;
657 	mdi = (struct g_raid_md_nvidia_object *)md;
658 	meta = mdi->mdio_meta;
659 
660 	/* Create volumes and subdisks. */
661 	nvidia_meta_get_name(meta, buf);
662 	vol = g_raid_create_volume(sc, buf, -1);
663 	vol->v_mediasize = (off_t)meta->total_sectors * 512;
664 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
665 	if (meta->type == NVIDIA_T_RAID0) {
666 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
667 		size = vol->v_mediasize / mdi->mdio_total_disks;
668 	} else if (meta->type == NVIDIA_T_RAID1) {
669 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
670 		size = vol->v_mediasize;
671 	} else if (meta->type == NVIDIA_T_RAID01) {
672 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
673 		size = vol->v_mediasize / (mdi->mdio_total_disks / 2);
674 	} else if (meta->type == NVIDIA_T_CONCAT) {
675 		if (mdi->mdio_total_disks == 1)
676 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
677 		else
678 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
679 		size = 0;
680 	} else if (meta->type == NVIDIA_T_RAID5) {
681 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
682 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
683 	} else if (meta->type == NVIDIA_T_RAID5_SYM) {
684 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
685 //		vol->v_raid_level_qualifier = 0x03;
686 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
687 	} else {
688 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
689 		size = 0;
690 	}
691 	vol->v_strip_size = meta->strip_sectors * 512; //ZZZ
692 	vol->v_disks_count = mdi->mdio_total_disks;
693 	vol->v_sectorsize = 512; //ZZZ
694 	for (j = 0; j < vol->v_disks_count; j++) {
695 		sd = &vol->v_subdisks[j];
696 		sd->sd_offset = 0;
697 		sd->sd_size = size;
698 	}
699 	g_raid_start_volume(vol);
700 
701 	/* Create disk placeholders to store data for later writing. */
702 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
703 		pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
704 		pd->pd_disk_pos = disk_pos;
705 		disk = g_raid_create_disk(sc);
706 		disk->d_md_data = (void *)pd;
707 		disk->d_state = G_RAID_DISK_S_OFFLINE;
708 		sd = &vol->v_subdisks[disk_pos];
709 		sd->sd_disk = disk;
710 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
711 	}
712 
713 	/* Make all disks found till the moment take their places. */
714 	do {
715 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
716 			if (disk->d_state == G_RAID_DISK_S_NONE) {
717 				g_raid_md_nvidia_start_disk(disk);
718 				break;
719 			}
720 		}
721 	} while (disk != NULL);
722 
723 	mdi->mdio_started = 1;
724 	G_RAID_DEBUG1(0, sc, "Array started.");
725 	g_raid_md_write_nvidia(md, NULL, NULL, NULL);
726 
727 	/* Pickup any STALE/SPARE disks to refill array if needed. */
728 	g_raid_md_nvidia_refill(sc);
729 
730 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
731 
732 	callout_stop(&mdi->mdio_start_co);
733 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
734 	root_mount_rel(mdi->mdio_rootmount);
735 	mdi->mdio_rootmount = NULL;
736 }
737 
738 static void
739 g_raid_md_nvidia_new_disk(struct g_raid_disk *disk)
740 {
741 	struct g_raid_softc *sc;
742 	struct g_raid_md_object *md;
743 	struct g_raid_md_nvidia_object *mdi;
744 	struct nvidia_raid_conf *pdmeta;
745 	struct g_raid_md_nvidia_perdisk *pd;
746 
747 	sc = disk->d_softc;
748 	md = sc->sc_md;
749 	mdi = (struct g_raid_md_nvidia_object *)md;
750 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
751 	pdmeta = pd->pd_meta;
752 
753 	if (mdi->mdio_started) {
754 		if (g_raid_md_nvidia_start_disk(disk))
755 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
756 	} else {
757 		if (mdi->mdio_meta == NULL ||
758 		    mdi->mdio_meta->disk_number >= mdi->mdio_meta->total_disks) {
759 			G_RAID_DEBUG1(1, sc, "Newer disk");
760 			if (mdi->mdio_meta != NULL)
761 				free(mdi->mdio_meta, M_MD_NVIDIA);
762 			mdi->mdio_meta = nvidia_meta_copy(pdmeta);
763 			mdi->mdio_total_disks = pdmeta->total_disks;
764 			mdi->mdio_disks_present = 1;
765 		} else if (pdmeta->disk_number < mdi->mdio_meta->total_disks) {
766 			mdi->mdio_disks_present++;
767 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
768 			    mdi->mdio_disks_present,
769 			    mdi->mdio_total_disks);
770 		} else
771 			G_RAID_DEBUG1(1, sc, "Spare disk");
772 
773 		/* If we collected all needed disks - start array. */
774 		if (mdi->mdio_disks_present == mdi->mdio_total_disks)
775 			g_raid_md_nvidia_start(sc);
776 	}
777 }
778 
779 static void
780 g_raid_nvidia_go(void *arg)
781 {
782 	struct g_raid_softc *sc;
783 	struct g_raid_md_object *md;
784 	struct g_raid_md_nvidia_object *mdi;
785 
786 	sc = arg;
787 	md = sc->sc_md;
788 	mdi = (struct g_raid_md_nvidia_object *)md;
789 	if (!mdi->mdio_started) {
790 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
791 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
792 	}
793 }
794 
795 static int
796 g_raid_md_create_nvidia(struct g_raid_md_object *md, struct g_class *mp,
797     struct g_geom **gp)
798 {
799 	struct g_raid_softc *sc;
800 	struct g_raid_md_nvidia_object *mdi;
801 	char name[32];
802 
803 	mdi = (struct g_raid_md_nvidia_object *)md;
804 	arc4rand(&mdi->mdio_volume_id, 16, 0);
805 	snprintf(name, sizeof(name), "NVIDIA-%d",
806 	    atomic_fetchadd_int(&NVIDIANodeID, 1));
807 	sc = g_raid_create_node(mp, name, md);
808 	if (sc == NULL)
809 		return (G_RAID_MD_TASTE_FAIL);
810 	md->mdo_softc = sc;
811 	*gp = sc->sc_geom;
812 	return (G_RAID_MD_TASTE_NEW);
813 }
814 
815 static int
816 g_raid_md_taste_nvidia(struct g_raid_md_object *md, struct g_class *mp,
817                               struct g_consumer *cp, struct g_geom **gp)
818 {
819 	struct g_consumer *rcp;
820 	struct g_provider *pp;
821 	struct g_raid_md_nvidia_object *mdi, *mdi1;
822 	struct g_raid_softc *sc;
823 	struct g_raid_disk *disk;
824 	struct nvidia_raid_conf *meta;
825 	struct g_raid_md_nvidia_perdisk *pd;
826 	struct g_geom *geom;
827 	int error, result, spare, len;
828 	char name[32];
829 	uint16_t vendor;
830 
831 	G_RAID_DEBUG(1, "Tasting NVIDIA on %s", cp->provider->name);
832 	mdi = (struct g_raid_md_nvidia_object *)md;
833 	pp = cp->provider;
834 
835 	/* Read metadata from device. */
836 	meta = NULL;
837 	vendor = 0xffff;
838 	if (g_access(cp, 1, 0, 0) != 0)
839 		return (G_RAID_MD_TASTE_FAIL);
840 	g_topology_unlock();
841 	len = 2;
842 	if (pp->geom->rank == 1)
843 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
844 	meta = nvidia_meta_read(cp);
845 	g_topology_lock();
846 	g_access(cp, -1, 0, 0);
847 	if (meta == NULL) {
848 		if (g_raid_aggressive_spare) {
849 			if (vendor == 0x10de) {
850 				G_RAID_DEBUG(1,
851 				    "No NVIDIA metadata, forcing spare.");
852 				spare = 2;
853 				goto search;
854 			} else {
855 				G_RAID_DEBUG(1,
856 				    "NVIDIA vendor mismatch 0x%04x != 0x10de",
857 				    vendor);
858 			}
859 		}
860 		return (G_RAID_MD_TASTE_FAIL);
861 	}
862 
863 	/* Metadata valid. Print it. */
864 	g_raid_md_nvidia_print(meta);
865 	G_RAID_DEBUG(1, "NVIDIA disk position %d", meta->disk_number);
866 	spare = 0;//(meta->type == NVIDIA_T_SPARE) ? 1 : 0;
867 
868 search:
869 	/* Search for matching node. */
870 	sc = NULL;
871 	mdi1 = NULL;
872 	LIST_FOREACH(geom, &mp->geom, geom) {
873 		sc = geom->softc;
874 		if (sc == NULL)
875 			continue;
876 		if (sc->sc_stopping != 0)
877 			continue;
878 		if (sc->sc_md->mdo_class != md->mdo_class)
879 			continue;
880 		mdi1 = (struct g_raid_md_nvidia_object *)sc->sc_md;
881 		if (spare) {
882 			if (mdi1->mdio_incomplete)
883 				break;
884 		} else {
885 			if (memcmp(&mdi1->mdio_volume_id,
886 			     &meta->volume_id, 16) == 0)
887 				break;
888 		}
889 	}
890 
891 	/* Found matching node. */
892 	if (geom != NULL) {
893 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
894 		result = G_RAID_MD_TASTE_EXISTING;
895 
896 	} else if (spare) { /* Not found needy node -- left for later. */
897 		G_RAID_DEBUG(1, "Spare is not needed at this time");
898 		goto fail1;
899 
900 	} else { /* Not found matching node -- create one. */
901 		result = G_RAID_MD_TASTE_NEW;
902 		memcpy(&mdi->mdio_volume_id, &meta->volume_id, 16);
903 		snprintf(name, sizeof(name), "NVIDIA-%d",
904 		    atomic_fetchadd_int(&NVIDIANodeID, 1));
905 		sc = g_raid_create_node(mp, name, md);
906 		md->mdo_softc = sc;
907 		geom = sc->sc_geom;
908 		callout_init(&mdi->mdio_start_co, 1);
909 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
910 		    g_raid_nvidia_go, sc);
911 		mdi->mdio_rootmount = root_mount_hold("GRAID-NVIDIA");
912 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
913 	}
914 
915 	rcp = g_new_consumer(geom);
916 	g_attach(rcp, pp);
917 	if (g_access(rcp, 1, 1, 1) != 0)
918 		; //goto fail1;
919 
920 	g_topology_unlock();
921 	sx_xlock(&sc->sc_lock);
922 
923 	pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
924 	pd->pd_meta = meta;
925 	if (spare == 2) {
926 		pd->pd_disk_pos = -3;
927 	} else {
928 		pd->pd_disk_pos = -1;
929 	}
930 	pd->pd_disk_size = pp->mediasize;
931 	disk = g_raid_create_disk(sc);
932 	disk->d_md_data = (void *)pd;
933 	disk->d_consumer = rcp;
934 	rcp->private = disk;
935 
936 	/* Read kernel dumping information. */
937 	disk->d_kd.offset = 0;
938 	disk->d_kd.length = OFF_MAX;
939 	len = sizeof(disk->d_kd);
940 	error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd);
941 	if (disk->d_kd.di.dumper == NULL)
942 		G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.",
943 		    rcp->provider->name, error);
944 
945 	g_raid_md_nvidia_new_disk(disk);
946 
947 	sx_xunlock(&sc->sc_lock);
948 	g_topology_lock();
949 	*gp = geom;
950 	return (result);
951 fail1:
952 	free(meta, M_MD_NVIDIA);
953 	return (G_RAID_MD_TASTE_FAIL);
954 }
955 
956 static int
957 g_raid_md_event_nvidia(struct g_raid_md_object *md,
958     struct g_raid_disk *disk, u_int event)
959 {
960 	struct g_raid_softc *sc;
961 	struct g_raid_subdisk *sd;
962 	struct g_raid_md_nvidia_object *mdi;
963 	struct g_raid_md_nvidia_perdisk *pd;
964 
965 	sc = md->mdo_softc;
966 	mdi = (struct g_raid_md_nvidia_object *)md;
967 	if (disk == NULL) {
968 		switch (event) {
969 		case G_RAID_NODE_E_START:
970 			if (!mdi->mdio_started) {
971 				/* Bump volume ID to drop missing disks. */
972 				arc4rand(&mdi->mdio_volume_id, 16, 0);
973 				g_raid_md_nvidia_start(sc);
974 			}
975 			return (0);
976 		}
977 		return (-1);
978 	}
979 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
980 	switch (event) {
981 	case G_RAID_DISK_E_DISCONNECTED:
982 		/* If disk was assigned, just update statuses. */
983 		if (pd->pd_disk_pos >= 0) {
984 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
985 			if (disk->d_consumer) {
986 				g_raid_kill_consumer(sc, disk->d_consumer);
987 				disk->d_consumer = NULL;
988 			}
989 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
990 				g_raid_change_subdisk_state(sd,
991 				    G_RAID_SUBDISK_S_NONE);
992 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
993 				    G_RAID_EVENT_SUBDISK);
994 			}
995 		} else {
996 			/* Otherwise -- delete. */
997 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
998 			g_raid_destroy_disk(disk);
999 		}
1000 
1001 		if (mdi->mdio_started) {
1002 			/* Bump volume ID to prevent disk resurrection. */
1003 			if (pd->pd_disk_pos >= 0)
1004 				arc4rand(&mdi->mdio_volume_id, 16, 0);
1005 
1006 			/* Write updated metadata to all disks. */
1007 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1008 		}
1009 
1010 		/* Check if anything left except placeholders. */
1011 		if (g_raid_ndisks(sc, -1) ==
1012 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1013 			g_raid_destroy_node(sc, 0);
1014 		else
1015 			g_raid_md_nvidia_refill(sc);
1016 		return (0);
1017 	}
1018 	return (-2);
1019 }
1020 
1021 static int
1022 g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
1023     struct gctl_req *req)
1024 {
1025 	struct g_raid_softc *sc;
1026 	struct g_raid_volume *vol;
1027 	struct g_raid_subdisk *sd;
1028 	struct g_raid_disk *disk;
1029 	struct g_raid_md_nvidia_object *mdi;
1030 	struct g_raid_md_nvidia_perdisk *pd;
1031 	struct g_consumer *cp;
1032 	struct g_provider *pp;
1033 	char arg[16];
1034 	const char *verb, *volname, *levelname, *diskname;
1035 	int *nargs, *force;
1036 	off_t size, sectorsize, strip, volsize;
1037 	intmax_t *sizearg, *striparg;
1038 	int numdisks, i, len, level, qual, update;
1039 	int error;
1040 
1041 	sc = md->mdo_softc;
1042 	mdi = (struct g_raid_md_nvidia_object *)md;
1043 	verb = gctl_get_param(req, "verb", NULL);
1044 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1045 	error = 0;
1046 	if (strcmp(verb, "label") == 0) {
1047 
1048 		if (*nargs < 4) {
1049 			gctl_error(req, "Invalid number of arguments.");
1050 			return (-1);
1051 		}
1052 		volname = gctl_get_asciiparam(req, "arg1");
1053 		if (volname == NULL) {
1054 			gctl_error(req, "No volume name.");
1055 			return (-2);
1056 		}
1057 		levelname = gctl_get_asciiparam(req, "arg2");
1058 		if (levelname == NULL) {
1059 			gctl_error(req, "No RAID level.");
1060 			return (-3);
1061 		}
1062 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1063 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1064 			return (-4);
1065 		}
1066 		numdisks = *nargs - 3;
1067 		force = gctl_get_paraml(req, "force", sizeof(*force));
1068 		if (!g_raid_md_nvidia_supported(level, qual, numdisks,
1069 		    force ? *force : 0)) {
1070 			gctl_error(req, "Unsupported RAID level "
1071 			    "(0x%02x/0x%02x), or number of disks (%d).",
1072 			    level, qual, numdisks);
1073 			return (-5);
1074 		}
1075 
1076 		/* Search for disks, connect them and probe. */
1077 		size = 0x7fffffffffffffffllu;
1078 		sectorsize = 0;
1079 		for (i = 0; i < numdisks; i++) {
1080 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1081 			diskname = gctl_get_asciiparam(req, arg);
1082 			if (diskname == NULL) {
1083 				gctl_error(req, "No disk name (%s).", arg);
1084 				error = -6;
1085 				break;
1086 			}
1087 			if (strcmp(diskname, "NONE") == 0) {
1088 				cp = NULL;
1089 				pp = NULL;
1090 			} else {
1091 				g_topology_lock();
1092 				cp = g_raid_open_consumer(sc, diskname);
1093 				if (cp == NULL) {
1094 					gctl_error(req, "Can't open '%s'.",
1095 					    diskname);
1096 					g_topology_unlock();
1097 					error = -7;
1098 					break;
1099 				}
1100 				pp = cp->provider;
1101 			}
1102 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1103 			pd->pd_disk_pos = i;
1104 			disk = g_raid_create_disk(sc);
1105 			disk->d_md_data = (void *)pd;
1106 			disk->d_consumer = cp;
1107 			if (cp == NULL)
1108 				continue;
1109 			cp->private = disk;
1110 			g_topology_unlock();
1111 
1112 			/* Read kernel dumping information. */
1113 			disk->d_kd.offset = 0;
1114 			disk->d_kd.length = OFF_MAX;
1115 			len = sizeof(disk->d_kd);
1116 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1117 			if (disk->d_kd.di.dumper == NULL)
1118 				G_RAID_DEBUG1(2, sc,
1119 				    "Dumping not supported by %s.",
1120 				    cp->provider->name);
1121 
1122 			pd->pd_disk_size = pp->mediasize;
1123 			if (size > pp->mediasize)
1124 				size = pp->mediasize;
1125 			if (sectorsize < pp->sectorsize)
1126 				sectorsize = pp->sectorsize;
1127 		}
1128 		if (error != 0)
1129 			return (error);
1130 
1131 		if (sectorsize <= 0) {
1132 			gctl_error(req, "Can't get sector size.");
1133 			return (-8);
1134 		}
1135 
1136 		/* Reserve space for metadata. */
1137 		size -= 2 * sectorsize;
1138 
1139 		/* Handle size argument. */
1140 		len = sizeof(*sizearg);
1141 		sizearg = gctl_get_param(req, "size", &len);
1142 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1143 		    *sizearg > 0) {
1144 			if (*sizearg > size) {
1145 				gctl_error(req, "Size too big %lld > %lld.",
1146 				    (long long)*sizearg, (long long)size);
1147 				return (-9);
1148 			}
1149 			size = *sizearg;
1150 		}
1151 
1152 		/* Handle strip argument. */
1153 		strip = 131072;
1154 		len = sizeof(*striparg);
1155 		striparg = gctl_get_param(req, "strip", &len);
1156 		if (striparg != NULL && len == sizeof(*striparg) &&
1157 		    *striparg > 0) {
1158 			if (*striparg < sectorsize) {
1159 				gctl_error(req, "Strip size too small.");
1160 				return (-10);
1161 			}
1162 			if (*striparg % sectorsize != 0) {
1163 				gctl_error(req, "Incorrect strip size.");
1164 				return (-11);
1165 			}
1166 			if (strip > 65535 * sectorsize) {
1167 				gctl_error(req, "Strip size too big.");
1168 				return (-12);
1169 			}
1170 			strip = *striparg;
1171 		}
1172 
1173 		/* Round size down to strip or sector. */
1174 		if (level == G_RAID_VOLUME_RL_RAID1)
1175 			size -= (size % sectorsize);
1176 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1177 		    (numdisks & 1) != 0)
1178 			size -= (size % (2 * strip));
1179 		else
1180 			size -= (size % strip);
1181 		if (size <= 0) {
1182 			gctl_error(req, "Size too small.");
1183 			return (-13);
1184 		}
1185 
1186 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1187 		    level == G_RAID_VOLUME_RL_CONCAT ||
1188 		    level == G_RAID_VOLUME_RL_SINGLE)
1189 			volsize = size * numdisks;
1190 		else if (level == G_RAID_VOLUME_RL_RAID1)
1191 			volsize = size;
1192 		else if (level == G_RAID_VOLUME_RL_RAID5)
1193 			volsize = size * (numdisks - 1);
1194 		else { /* RAID1E */
1195 			volsize = ((size * numdisks) / strip / 2) *
1196 			    strip;
1197 		}
1198 		if (volsize > 0xffffffffllu * sectorsize) {
1199 			gctl_error(req, "Size too big.");
1200 			return (-14);
1201 		}
1202 
1203 		/* We have all we need, create things: volume, ... */
1204 		mdi->mdio_total_disks = numdisks;
1205 		mdi->mdio_started = 1;
1206 		vol = g_raid_create_volume(sc, volname, -1);
1207 		vol->v_md_data = (void *)(intptr_t)0;
1208 		vol->v_raid_level = level;
1209 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
1210 		vol->v_strip_size = strip;
1211 		vol->v_disks_count = numdisks;
1212 		vol->v_mediasize = volsize;
1213 		vol->v_sectorsize = sectorsize;
1214 		g_raid_start_volume(vol);
1215 
1216 		/* , and subdisks. */
1217 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1218 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1219 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1220 			sd->sd_disk = disk;
1221 			sd->sd_offset = 0;
1222 			sd->sd_size = size;
1223 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1224 			if (sd->sd_disk->d_consumer != NULL) {
1225 				g_raid_change_disk_state(disk,
1226 				    G_RAID_DISK_S_ACTIVE);
1227 				g_raid_change_subdisk_state(sd,
1228 				    G_RAID_SUBDISK_S_ACTIVE);
1229 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1230 				    G_RAID_EVENT_SUBDISK);
1231 			} else {
1232 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1233 			}
1234 		}
1235 
1236 		/* Write metadata based on created entities. */
1237 		G_RAID_DEBUG1(0, sc, "Array started.");
1238 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1239 
1240 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1241 		g_raid_md_nvidia_refill(sc);
1242 
1243 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1244 		    G_RAID_EVENT_VOLUME);
1245 		return (0);
1246 	}
1247 	if (strcmp(verb, "delete") == 0) {
1248 
1249 		/* Check if some volume is still open. */
1250 		force = gctl_get_paraml(req, "force", sizeof(*force));
1251 		if (force != NULL && *force == 0 &&
1252 		    g_raid_nopens(sc) != 0) {
1253 			gctl_error(req, "Some volume is still open.");
1254 			return (-4);
1255 		}
1256 
1257 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1258 			if (disk->d_consumer)
1259 				nvidia_meta_erase(disk->d_consumer);
1260 		}
1261 		g_raid_destroy_node(sc, 0);
1262 		return (0);
1263 	}
1264 	if (strcmp(verb, "remove") == 0 ||
1265 	    strcmp(verb, "fail") == 0) {
1266 		if (*nargs < 2) {
1267 			gctl_error(req, "Invalid number of arguments.");
1268 			return (-1);
1269 		}
1270 		for (i = 1; i < *nargs; i++) {
1271 			snprintf(arg, sizeof(arg), "arg%d", i);
1272 			diskname = gctl_get_asciiparam(req, arg);
1273 			if (diskname == NULL) {
1274 				gctl_error(req, "No disk name (%s).", arg);
1275 				error = -2;
1276 				break;
1277 			}
1278 			if (strncmp(diskname, "/dev/", 5) == 0)
1279 				diskname += 5;
1280 
1281 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1282 				if (disk->d_consumer != NULL &&
1283 				    disk->d_consumer->provider != NULL &&
1284 				    strcmp(disk->d_consumer->provider->name,
1285 				     diskname) == 0)
1286 					break;
1287 			}
1288 			if (disk == NULL) {
1289 				gctl_error(req, "Disk '%s' not found.",
1290 				    diskname);
1291 				error = -3;
1292 				break;
1293 			}
1294 
1295 			if (strcmp(verb, "fail") == 0) {
1296 				g_raid_md_fail_disk_nvidia(md, NULL, disk);
1297 				continue;
1298 			}
1299 
1300 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1301 
1302 			/* Erase metadata on deleting disk. */
1303 			nvidia_meta_erase(disk->d_consumer);
1304 
1305 			/* If disk was assigned, just update statuses. */
1306 			if (pd->pd_disk_pos >= 0) {
1307 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1308 				g_raid_kill_consumer(sc, disk->d_consumer);
1309 				disk->d_consumer = NULL;
1310 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1311 					g_raid_change_subdisk_state(sd,
1312 					    G_RAID_SUBDISK_S_NONE);
1313 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1314 					    G_RAID_EVENT_SUBDISK);
1315 				}
1316 			} else {
1317 				/* Otherwise -- delete. */
1318 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1319 				g_raid_destroy_disk(disk);
1320 			}
1321 		}
1322 
1323 		/* Write updated metadata to remaining disks. */
1324 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1325 
1326 		/* Check if anything left except placeholders. */
1327 		if (g_raid_ndisks(sc, -1) ==
1328 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1329 			g_raid_destroy_node(sc, 0);
1330 		else
1331 			g_raid_md_nvidia_refill(sc);
1332 		return (error);
1333 	}
1334 	if (strcmp(verb, "insert") == 0) {
1335 		if (*nargs < 2) {
1336 			gctl_error(req, "Invalid number of arguments.");
1337 			return (-1);
1338 		}
1339 		update = 0;
1340 		for (i = 1; i < *nargs; i++) {
1341 			/* Get disk name. */
1342 			snprintf(arg, sizeof(arg), "arg%d", i);
1343 			diskname = gctl_get_asciiparam(req, arg);
1344 			if (diskname == NULL) {
1345 				gctl_error(req, "No disk name (%s).", arg);
1346 				error = -3;
1347 				break;
1348 			}
1349 
1350 			/* Try to find provider with specified name. */
1351 			g_topology_lock();
1352 			cp = g_raid_open_consumer(sc, diskname);
1353 			if (cp == NULL) {
1354 				gctl_error(req, "Can't open disk '%s'.",
1355 				    diskname);
1356 				g_topology_unlock();
1357 				error = -4;
1358 				break;
1359 			}
1360 			pp = cp->provider;
1361 
1362 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1363 			pd->pd_disk_pos = -3;
1364 			pd->pd_disk_size = pp->mediasize;
1365 
1366 			disk = g_raid_create_disk(sc);
1367 			disk->d_consumer = cp;
1368 			disk->d_md_data = (void *)pd;
1369 			cp->private = disk;
1370 			g_topology_unlock();
1371 
1372 			/* Read kernel dumping information. */
1373 			disk->d_kd.offset = 0;
1374 			disk->d_kd.length = OFF_MAX;
1375 			len = sizeof(disk->d_kd);
1376 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1377 			if (disk->d_kd.di.dumper == NULL)
1378 				G_RAID_DEBUG1(2, sc,
1379 				    "Dumping not supported by %s.",
1380 				    cp->provider->name);
1381 
1382 			/* Welcome the "new" disk. */
1383 			update += g_raid_md_nvidia_start_disk(disk);
1384 			if (disk->d_state != G_RAID_DISK_S_SPARE &&
1385 			    disk->d_state != G_RAID_DISK_S_ACTIVE) {
1386 				gctl_error(req, "Disk '%s' doesn't fit.",
1387 				    diskname);
1388 				g_raid_destroy_disk(disk);
1389 				error = -8;
1390 				break;
1391 			}
1392 		}
1393 
1394 		/* Write new metadata if we changed something. */
1395 		if (update)
1396 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1397 		return (error);
1398 	}
1399 	gctl_error(req, "Command '%s' is not supported.", verb);
1400 	return (-100);
1401 }
1402 
1403 static int
1404 g_raid_md_write_nvidia(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1405     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1406 {
1407 	struct g_raid_softc *sc;
1408 	struct g_raid_volume *vol;
1409 	struct g_raid_subdisk *sd;
1410 	struct g_raid_disk *disk;
1411 	struct g_raid_md_nvidia_object *mdi;
1412 	struct g_raid_md_nvidia_perdisk *pd;
1413 	struct nvidia_raid_conf *meta;
1414 	int i, spares;
1415 
1416 	sc = md->mdo_softc;
1417 	mdi = (struct g_raid_md_nvidia_object *)md;
1418 
1419 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1420 		return (0);
1421 
1422 	/* There is only one volume. */
1423 	vol = TAILQ_FIRST(&sc->sc_volumes);
1424 
1425 	/* Fill global fields. */
1426 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1427 	if (mdi->mdio_meta)
1428 		memcpy(meta, mdi->mdio_meta, sizeof(*meta));
1429 	memcpy(meta->nvidia_id, NVIDIA_MAGIC, sizeof(NVIDIA_MAGIC) - 1);
1430 	meta->config_size = 30;
1431 	meta->version = 0x0064;
1432 	meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1433 	meta->sector_size = vol->v_sectorsize;
1434 	nvidia_meta_put_name(meta, vol->v_name);
1435 	meta->magic_0 = NVIDIA_MAGIC0;
1436 	memcpy(&meta->volume_id, &mdi->mdio_volume_id, 16);
1437 	meta->state = NVIDIA_S_IDLE;
1438 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1439 		meta->array_width = 1;
1440 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1441 		meta->array_width = vol->v_disks_count / 2;
1442 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1443 		meta->array_width = vol->v_disks_count - 1;
1444 	else
1445 		meta->array_width = vol->v_disks_count;
1446 	meta->total_disks = vol->v_disks_count;
1447 	meta->orig_array_width = meta->array_width;
1448 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
1449 		meta->type = NVIDIA_T_RAID0;
1450 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1451 		meta->type = NVIDIA_T_RAID1;
1452 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1453 		meta->type = NVIDIA_T_RAID01;
1454 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1455 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1456 		meta->type = NVIDIA_T_CONCAT;
1457 //	else if (vol->v_raid_level_qualifier == 0)
1458 //		meta->type = NVIDIA_T_RAID5;
1459 	else
1460 		meta->type = NVIDIA_T_RAID5_SYM;
1461 	meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize;
1462 	meta->strip_bytes = vol->v_strip_size;
1463 	meta->strip_shift = ffs(meta->strip_sectors) - 1;
1464 	meta->strip_mask = meta->strip_sectors - 1;
1465 	meta->stripe_sectors = meta->strip_sectors * meta->orig_array_width;
1466 	meta->stripe_bytes = meta->stripe_sectors * vol->v_sectorsize;
1467 	meta->rebuild_lba = 0;
1468 	meta->orig_type = meta->type;
1469 	meta->orig_total_sectors = meta->total_sectors;
1470 	meta->status = 0;
1471 
1472 	for (i = 0; i < vol->v_disks_count; i++) {
1473 		sd = &vol->v_subdisks[i];
1474 		if ((sd->sd_state == G_RAID_SUBDISK_S_STALE ||
1475 		     sd->sd_state == G_RAID_SUBDISK_S_RESYNC ||
1476 		     vol->v_dirty) &&
1477 		     meta->state != NVIDIA_S_REBUILD)
1478 			meta->state = NVIDIA_S_SYNC;
1479 		else if (sd->sd_state == G_RAID_SUBDISK_S_NEW ||
1480 		     sd->sd_state == G_RAID_SUBDISK_S_REBUILD)
1481 			meta->state = NVIDIA_S_REBUILD;
1482 	}
1483 
1484 	/* We are done. Print meta data and store them to disks. */
1485 	if (mdi->mdio_meta != NULL)
1486 		free(mdi->mdio_meta, M_MD_NVIDIA);
1487 	mdi->mdio_meta = meta;
1488 	spares = 0;
1489 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1490 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1491 		if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1492 		    disk->d_state != G_RAID_DISK_S_SPARE)
1493 			continue;
1494 		if (pd->pd_meta != NULL) {
1495 			free(pd->pd_meta, M_MD_NVIDIA);
1496 			pd->pd_meta = NULL;
1497 		}
1498 		pd->pd_meta = nvidia_meta_copy(meta);
1499 		if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1500 			/* For RAID0+1 we need to translate order. */
1501 			pd->pd_meta->disk_number =
1502 			    nvidia_meta_translate_disk(meta, sd->sd_pos);
1503 			if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE) {
1504 				pd->pd_meta->disk_status = 0x100;
1505 				pd->pd_meta->rebuild_lba =
1506 				    sd->sd_rebuild_pos / vol->v_sectorsize *
1507 				    meta->array_width;
1508 			}
1509 		} else
1510 			pd->pd_meta->disk_number = meta->total_disks + spares++;
1511 		G_RAID_DEBUG(1, "Writing NVIDIA metadata to %s",
1512 		    g_raid_get_diskname(disk));
1513 		g_raid_md_nvidia_print(pd->pd_meta);
1514 		nvidia_meta_write(disk->d_consumer, pd->pd_meta);
1515 	}
1516 	return (0);
1517 }
1518 
1519 static int
1520 g_raid_md_fail_disk_nvidia(struct g_raid_md_object *md,
1521     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1522 {
1523 	struct g_raid_softc *sc;
1524 	struct g_raid_md_nvidia_perdisk *pd;
1525 	struct g_raid_subdisk *sd;
1526 
1527 	sc = md->mdo_softc;
1528 	pd = (struct g_raid_md_nvidia_perdisk *)tdisk->d_md_data;
1529 
1530 	/* We can't fail disk that is not a part of array now. */
1531 	if (pd->pd_disk_pos < 0)
1532 		return (-1);
1533 
1534 	/* Erase metadata to prevent disks's later resurrection. */
1535 	if (tdisk->d_consumer)
1536 		nvidia_meta_erase(tdisk->d_consumer);
1537 
1538 	/* Change states. */
1539 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1540 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1541 		g_raid_change_subdisk_state(sd,
1542 		    G_RAID_SUBDISK_S_FAILED);
1543 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1544 		    G_RAID_EVENT_SUBDISK);
1545 	}
1546 
1547 	/* Write updated metadata to remaining disks. */
1548 	g_raid_md_write_nvidia(md, NULL, NULL, tdisk);
1549 
1550 	/* Check if anything left except placeholders. */
1551 	if (g_raid_ndisks(sc, -1) ==
1552 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1553 		g_raid_destroy_node(sc, 0);
1554 	else
1555 		g_raid_md_nvidia_refill(sc);
1556 	return (0);
1557 }
1558 
1559 static int
1560 g_raid_md_free_disk_nvidia(struct g_raid_md_object *md,
1561     struct g_raid_disk *disk)
1562 {
1563 	struct g_raid_md_nvidia_perdisk *pd;
1564 
1565 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1566 	if (pd->pd_meta != NULL) {
1567 		free(pd->pd_meta, M_MD_NVIDIA);
1568 		pd->pd_meta = NULL;
1569 	}
1570 	free(pd, M_MD_NVIDIA);
1571 	disk->d_md_data = NULL;
1572 	return (0);
1573 }
1574 
1575 static int
1576 g_raid_md_free_nvidia(struct g_raid_md_object *md)
1577 {
1578 	struct g_raid_md_nvidia_object *mdi;
1579 
1580 	mdi = (struct g_raid_md_nvidia_object *)md;
1581 	if (!mdi->mdio_started) {
1582 		mdi->mdio_started = 0;
1583 		callout_stop(&mdi->mdio_start_co);
1584 		G_RAID_DEBUG1(1, md->mdo_softc,
1585 		    "root_mount_rel %p", mdi->mdio_rootmount);
1586 		root_mount_rel(mdi->mdio_rootmount);
1587 		mdi->mdio_rootmount = NULL;
1588 	}
1589 	if (mdi->mdio_meta != NULL) {
1590 		free(mdi->mdio_meta, M_MD_NVIDIA);
1591 		mdi->mdio_meta = NULL;
1592 	}
1593 	return (0);
1594 }
1595 
1596 G_RAID_MD_DECLARE(g_raid_md_nvidia);
1597