xref: /freebsd/sys/geom/raid/md_nvidia.c (revision 09d325677d53a12c79a43664ff29871e92247629)
1 /*-
2  * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
3  * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include <sys/param.h>
32 #include <sys/bio.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/kobj.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/systm.h>
41 #include <sys/taskqueue.h>
42 #include <geom/geom.h>
43 #include "geom/raid/g_raid.h"
44 #include "g_raid_md_if.h"
45 
46 static MALLOC_DEFINE(M_MD_NVIDIA, "md_nvidia_data", "GEOM_RAID NVIDIA metadata");
47 
48 struct nvidia_raid_conf {
49 	uint8_t		nvidia_id[8];
50 #define NVIDIA_MAGIC                "NVIDIA  "
51 
52 	uint32_t	config_size;
53 	uint32_t	checksum;
54 	uint16_t	version;
55 	uint8_t		disk_number;
56 	uint8_t		dummy_0;
57 	uint32_t	total_sectors;
58 	uint32_t	sector_size;
59 	uint8_t		name[16];
60 	uint8_t		revision[4];
61 	uint32_t	disk_status;
62 
63 	uint32_t	magic_0;
64 #define NVIDIA_MAGIC0		0x00640044
65 
66 	uint64_t	volume_id[2];
67 	uint8_t		state;
68 #define NVIDIA_S_IDLE		0
69 #define NVIDIA_S_INIT		2
70 #define NVIDIA_S_REBUILD	3
71 #define NVIDIA_S_UPGRADE	4
72 #define NVIDIA_S_SYNC		5
73 	uint8_t		array_width;
74 	uint8_t		total_disks;
75 	uint8_t		orig_array_width;
76 	uint16_t	type;
77 #define NVIDIA_T_RAID0		0x0080
78 #define NVIDIA_T_RAID1		0x0081
79 #define NVIDIA_T_RAID3		0x0083
80 #define NVIDIA_T_RAID5		0x0085	/* RLQ = 00/02? */
81 #define NVIDIA_T_RAID5_SYM	0x0095	/* RLQ = 03 */
82 #define NVIDIA_T_RAID10		0x008a
83 #define NVIDIA_T_RAID01		0x8180
84 #define NVIDIA_T_CONCAT		0x00ff
85 
86 	uint16_t	dummy_3;
87 	uint32_t	strip_sectors;
88 	uint32_t	strip_bytes;
89 	uint32_t	strip_shift;
90 	uint32_t	strip_mask;
91 	uint32_t	stripe_sectors;
92 	uint32_t	stripe_bytes;
93 	uint32_t	rebuild_lba;
94 	uint32_t	orig_type;
95 	uint32_t	orig_total_sectors;
96 	uint32_t	status;
97 #define NVIDIA_S_BOOTABLE	0x00000001
98 #define NVIDIA_S_DEGRADED	0x00000002
99 
100 	uint32_t	filler[98];
101 } __packed;
102 
103 struct g_raid_md_nvidia_perdisk {
104 	struct nvidia_raid_conf	*pd_meta;
105 	int			 pd_disk_pos;
106 	off_t			 pd_disk_size;
107 };
108 
109 struct g_raid_md_nvidia_object {
110 	struct g_raid_md_object	 mdio_base;
111 	uint64_t		 mdio_volume_id[2];
112 	struct nvidia_raid_conf	*mdio_meta;
113 	struct callout		 mdio_start_co;	/* STARTING state timer. */
114 	int			 mdio_total_disks;
115 	int			 mdio_disks_present;
116 	int			 mdio_started;
117 	int			 mdio_incomplete;
118 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
119 };
120 
121 static g_raid_md_create_t g_raid_md_create_nvidia;
122 static g_raid_md_taste_t g_raid_md_taste_nvidia;
123 static g_raid_md_event_t g_raid_md_event_nvidia;
124 static g_raid_md_ctl_t g_raid_md_ctl_nvidia;
125 static g_raid_md_write_t g_raid_md_write_nvidia;
126 static g_raid_md_fail_disk_t g_raid_md_fail_disk_nvidia;
127 static g_raid_md_free_disk_t g_raid_md_free_disk_nvidia;
128 static g_raid_md_free_t g_raid_md_free_nvidia;
129 
130 static kobj_method_t g_raid_md_nvidia_methods[] = {
131 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_nvidia),
132 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_nvidia),
133 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_nvidia),
134 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_nvidia),
135 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_nvidia),
136 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_nvidia),
137 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_nvidia),
138 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_nvidia),
139 	{ 0, 0 }
140 };
141 
142 static struct g_raid_md_class g_raid_md_nvidia_class = {
143 	"NVIDIA",
144 	g_raid_md_nvidia_methods,
145 	sizeof(struct g_raid_md_nvidia_object),
146 	.mdc_enable = 1,
147 	.mdc_priority = 100
148 };
149 
150 static int NVIDIANodeID = 1;
151 
152 static void
153 g_raid_md_nvidia_print(struct nvidia_raid_conf *meta)
154 {
155 
156 	if (g_raid_debug < 1)
157 		return;
158 
159 	printf("********* ATA NVIDIA RAID Metadata *********\n");
160 	printf("nvidia_id           <%.8s>\n", meta->nvidia_id);
161 	printf("config_size         %u\n", meta->config_size);
162 	printf("checksum            0x%08x\n", meta->checksum);
163 	printf("version             0x%04x\n", meta->version);
164 	printf("disk_number         %d\n", meta->disk_number);
165 	printf("dummy_0             0x%02x\n", meta->dummy_0);
166 	printf("total_sectors       %u\n", meta->total_sectors);
167 	printf("sector_size         %u\n", meta->sector_size);
168 	printf("name                <%.16s>\n", meta->name);
169 	printf("revision            0x%02x%02x%02x%02x\n",
170 	    meta->revision[0], meta->revision[1],
171 	    meta->revision[2], meta->revision[3]);
172 	printf("disk_status         0x%08x\n", meta->disk_status);
173 	printf("magic_0             0x%08x\n", meta->magic_0);
174 	printf("volume_id           0x%016jx%016jx\n",
175 	    meta->volume_id[1], meta->volume_id[0]);
176 	printf("state               0x%02x\n", meta->state);
177 	printf("array_width         %u\n", meta->array_width);
178 	printf("total_disks         %u\n", meta->total_disks);
179 	printf("orig_array_width    %u\n", meta->orig_array_width);
180 	printf("type                0x%04x\n", meta->type);
181 	printf("dummy_3             0x%04x\n", meta->dummy_3);
182 	printf("strip_sectors       %u\n", meta->strip_sectors);
183 	printf("strip_bytes         %u\n", meta->strip_bytes);
184 	printf("strip_shift         %u\n", meta->strip_shift);
185 	printf("strip_mask          0x%08x\n", meta->strip_mask);
186 	printf("stripe_sectors      %u\n", meta->stripe_sectors);
187 	printf("stripe_bytes        %u\n", meta->stripe_bytes);
188 	printf("rebuild_lba         %u\n", meta->rebuild_lba);
189 	printf("orig_type           0x%04x\n", meta->orig_type);
190 	printf("orig_total_sectors  %u\n", meta->orig_total_sectors);
191 	printf("status              0x%08x\n", meta->status);
192 	printf("=================================================\n");
193 }
194 
195 static struct nvidia_raid_conf *
196 nvidia_meta_copy(struct nvidia_raid_conf *meta)
197 {
198 	struct nvidia_raid_conf *nmeta;
199 
200 	nmeta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
201 	memcpy(nmeta, meta, sizeof(*meta));
202 	return (nmeta);
203 }
204 
205 static int
206 nvidia_meta_translate_disk(struct nvidia_raid_conf *meta, int md_disk_pos)
207 {
208 	int disk_pos;
209 
210 	if (md_disk_pos >= 0 && meta->type == NVIDIA_T_RAID01) {
211 		disk_pos = (md_disk_pos / meta->array_width) +
212 		    (md_disk_pos % meta->array_width) * meta->array_width;
213 	} else
214 		disk_pos = md_disk_pos;
215 	return (disk_pos);
216 }
217 
218 static void
219 nvidia_meta_get_name(struct nvidia_raid_conf *meta, char *buf)
220 {
221 	int i;
222 
223 	strncpy(buf, meta->name, 16);
224 	buf[16] = 0;
225 	for (i = 15; i >= 0; i--) {
226 		if (buf[i] > 0x20)
227 			break;
228 		buf[i] = 0;
229 	}
230 }
231 
232 static void
233 nvidia_meta_put_name(struct nvidia_raid_conf *meta, char *buf)
234 {
235 
236 	memset(meta->name, 0x20, 16);
237 	memcpy(meta->name, buf, MIN(strlen(buf), 16));
238 }
239 
240 static struct nvidia_raid_conf *
241 nvidia_meta_read(struct g_consumer *cp)
242 {
243 	struct g_provider *pp;
244 	struct nvidia_raid_conf *meta;
245 	char *buf;
246 	int error, i;
247 	uint32_t checksum, *ptr;
248 
249 	pp = cp->provider;
250 
251 	/* Read the anchor sector. */
252 	buf = g_read_data(cp,
253 	    pp->mediasize - 2 * pp->sectorsize, pp->sectorsize, &error);
254 	if (buf == NULL) {
255 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
256 		    pp->name, error);
257 		return (NULL);
258 	}
259 	meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
260 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
261 	g_free(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 		free(meta, M_MD_NVIDIA);
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 		free(meta, M_MD_NVIDIA);
274 		return (NULL);
275 	}
276 
277 	/* Check metadata checksum. */
278 	for (checksum = 0, ptr = (uint32_t *)meta,
279 	    i = 0; i < meta->config_size; i++)
280 		checksum += *ptr++;
281 	if (checksum != 0) {
282 		G_RAID_DEBUG(1, "NVIDIA checksum check failed on %s", pp->name);
283 		free(meta, M_MD_NVIDIA);
284 		return (NULL);
285 	}
286 
287 	/* Check volume state. */
288 	if (meta->state != NVIDIA_S_IDLE && meta->state != NVIDIA_S_INIT &&
289 	    meta->state != NVIDIA_S_REBUILD && meta->state != NVIDIA_S_SYNC) {
290 		G_RAID_DEBUG(1, "NVIDIA unknown state on %s (0x%02x)",
291 		    pp->name, meta->state);
292 		free(meta, M_MD_NVIDIA);
293 		return (NULL);
294 	}
295 
296 	/* Check raid type. */
297 	if (meta->type != NVIDIA_T_RAID0 && meta->type != NVIDIA_T_RAID1 &&
298 	    meta->type != NVIDIA_T_RAID3 && meta->type != NVIDIA_T_RAID5 &&
299 	    meta->type != NVIDIA_T_RAID5_SYM &&
300 	    meta->type != NVIDIA_T_RAID01 && meta->type != NVIDIA_T_CONCAT) {
301 		G_RAID_DEBUG(1, "NVIDIA unknown RAID level on %s (0x%02x)",
302 		    pp->name, meta->type);
303 		free(meta, M_MD_NVIDIA);
304 		return (NULL);
305 	}
306 
307 	return (meta);
308 }
309 
310 static int
311 nvidia_meta_write(struct g_consumer *cp, struct nvidia_raid_conf *meta)
312 {
313 	struct g_provider *pp;
314 	char *buf;
315 	int error, i;
316 	uint32_t checksum, *ptr;
317 
318 	pp = cp->provider;
319 
320 	/* Recalculate checksum for case if metadata were changed. */
321 	meta->checksum = 0;
322 	for (checksum = 0, ptr = (uint32_t *)meta,
323 	    i = 0; i < meta->config_size; i++)
324 		checksum += *ptr++;
325 	meta->checksum -= checksum;
326 
327 	/* Create and fill buffer. */
328 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
329 	memcpy(buf, meta, sizeof(*meta));
330 
331 	/* Write metadata. */
332 	error = g_write_data(cp,
333 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
334 	if (error != 0) {
335 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
336 		    pp->name, error);
337 	}
338 
339 	free(buf, M_MD_NVIDIA);
340 	return (error);
341 }
342 
343 static int
344 nvidia_meta_erase(struct g_consumer *cp)
345 {
346 	struct g_provider *pp;
347 	char *buf;
348 	int error;
349 
350 	pp = cp->provider;
351 	buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
352 	error = g_write_data(cp,
353 	    pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
354 	if (error != 0) {
355 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
356 		    pp->name, error);
357 	}
358 	free(buf, M_MD_NVIDIA);
359 	return (error);
360 }
361 
362 static struct g_raid_disk *
363 g_raid_md_nvidia_get_disk(struct g_raid_softc *sc, int id)
364 {
365 	struct g_raid_disk	*disk;
366 	struct g_raid_md_nvidia_perdisk *pd;
367 
368 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
369 		pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
370 		if (pd->pd_disk_pos == id)
371 			break;
372 	}
373 	return (disk);
374 }
375 
376 static int
377 g_raid_md_nvidia_supported(int level, int qual, int disks, int force)
378 {
379 
380 	switch (level) {
381 	case G_RAID_VOLUME_RL_RAID0:
382 		if (disks < 1)
383 			return (0);
384 		if (!force && (disks < 2 || disks > 6))
385 			return (0);
386 		break;
387 	case G_RAID_VOLUME_RL_RAID1:
388 		if (disks < 1)
389 			return (0);
390 		if (!force && (disks != 2))
391 			return (0);
392 		break;
393 	case G_RAID_VOLUME_RL_RAID1E:
394 		if (disks < 2)
395 			return (0);
396 		if (disks % 2 != 0)
397 			return (0);
398 		if (!force && (disks < 4))
399 			return (0);
400 		break;
401 	case G_RAID_VOLUME_RL_SINGLE:
402 		if (disks != 1)
403 			return (0);
404 		break;
405 	case G_RAID_VOLUME_RL_CONCAT:
406 		if (disks < 2)
407 			return (0);
408 		break;
409 	case G_RAID_VOLUME_RL_RAID5:
410 		if (disks < 3)
411 			return (0);
412 		if (qual != G_RAID_VOLUME_RLQ_R5LA &&
413 		    qual != G_RAID_VOLUME_RLQ_R5LS)
414 			return (0);
415 		break;
416 	default:
417 		return (0);
418 	}
419 	if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
420 		return (0);
421 	return (1);
422 }
423 
424 static int
425 g_raid_md_nvidia_start_disk(struct g_raid_disk *disk)
426 {
427 	struct g_raid_softc *sc;
428 	struct g_raid_subdisk *sd, *tmpsd;
429 	struct g_raid_disk *olddisk, *tmpdisk;
430 	struct g_raid_md_object *md;
431 	struct g_raid_md_nvidia_object *mdi;
432 	struct g_raid_md_nvidia_perdisk *pd, *oldpd;
433 	struct nvidia_raid_conf *meta;
434 	int disk_pos, resurrection = 0;
435 
436 	sc = disk->d_softc;
437 	md = sc->sc_md;
438 	mdi = (struct g_raid_md_nvidia_object *)md;
439 	meta = mdi->mdio_meta;
440 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
441 	olddisk = NULL;
442 
443 	/* Find disk position in metadata by it's serial. */
444 	if (pd->pd_meta != NULL) {
445 		disk_pos = pd->pd_meta->disk_number;
446 		if (disk_pos >= meta->total_disks || mdi->mdio_started)
447 			disk_pos = -3;
448 	} else
449 		disk_pos = -3;
450 	/* For RAID0+1 we need to translate order. */
451 	disk_pos = nvidia_meta_translate_disk(meta, disk_pos);
452 	if (disk_pos < 0) {
453 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
454 		/* If we are in the start process, that's all for now. */
455 		if (!mdi->mdio_started)
456 			goto nofit;
457 		/*
458 		 * If we have already started - try to get use of the disk.
459 		 * Try to replace OFFLINE disks first, then FAILED.
460 		 */
461 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
462 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
463 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
464 				continue;
465 			/* Make sure this disk is big enough. */
466 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
467 				if (sd->sd_offset + sd->sd_size + 2 * 512 >
468 				    pd->pd_disk_size) {
469 					G_RAID_DEBUG1(1, sc,
470 					    "Disk too small (%ju < %ju)",
471 					    pd->pd_disk_size,
472 					    sd->sd_offset + sd->sd_size + 512);
473 					break;
474 				}
475 			}
476 			if (sd != NULL)
477 				continue;
478 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
479 				olddisk = tmpdisk;
480 				break;
481 			} else if (olddisk == NULL)
482 				olddisk = tmpdisk;
483 		}
484 		if (olddisk == NULL) {
485 nofit:
486 			g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
487 			return (1);
488 		}
489 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
490 		disk_pos = oldpd->pd_disk_pos;
491 		resurrection = 1;
492 	}
493 
494 	if (olddisk == NULL) {
495 		/* Find placeholder by position. */
496 		olddisk = g_raid_md_nvidia_get_disk(sc, disk_pos);
497 		if (olddisk == NULL)
498 			panic("No disk at position %d!", disk_pos);
499 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
500 			G_RAID_DEBUG1(1, sc, "More then one disk for pos %d",
501 			    disk_pos);
502 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
503 			return (0);
504 		}
505 		oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
506 	}
507 
508 	/* Replace failed disk or placeholder with new disk. */
509 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
510 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
511 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
512 		sd->sd_disk = disk;
513 	}
514 	oldpd->pd_disk_pos = -2;
515 	pd->pd_disk_pos = disk_pos;
516 
517 	/* If it was placeholder -- destroy it. */
518 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
519 		g_raid_destroy_disk(olddisk);
520 	} else {
521 		/* Otherwise, make it STALE_FAILED. */
522 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
523 	}
524 
525 	/* Welcome the new disk. */
526 	if (resurrection)
527 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
528 	else// if (pd->pd_meta->disk_status == NVIDIA_S_CURRENT ||
529 	    //pd->pd_meta->disk_status == NVIDIA_S_REBUILD)
530 		g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
531 //	else
532 //		g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
533 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
534 
535 		/*
536 		 * Different disks may have different sizes,
537 		 * in concat mode. Update from real disk size.
538 		 */
539 		if (meta->type == NVIDIA_T_CONCAT)
540 			sd->sd_size = pd->pd_disk_size - 0x800 * 512;
541 
542 		if (resurrection) {
543 			/* New or ex-spare disk. */
544 			g_raid_change_subdisk_state(sd,
545 			    G_RAID_SUBDISK_S_NEW);
546 		} else if (meta->state == NVIDIA_S_REBUILD &&
547 		    (pd->pd_meta->disk_status & 0x100)) {
548 			/* Rebuilding disk. */
549 			g_raid_change_subdisk_state(sd,
550 			    G_RAID_SUBDISK_S_REBUILD);
551 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
552 			    meta->array_width * pd->pd_meta->sector_size;
553 		} else if (meta->state == NVIDIA_S_SYNC) {
554 			/* Resyncing/dirty disk. */
555 			g_raid_change_subdisk_state(sd,
556 			    G_RAID_SUBDISK_S_RESYNC);
557 			sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
558 			    meta->array_width * pd->pd_meta->sector_size;
559 		} else {
560 			/* Up to date disk. */
561 			g_raid_change_subdisk_state(sd,
562 			    G_RAID_SUBDISK_S_ACTIVE);
563 		}
564 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
565 		    G_RAID_EVENT_SUBDISK);
566 	}
567 
568 	/* Update status of our need for spare. */
569 	if (mdi->mdio_started) {
570 		mdi->mdio_incomplete =
571 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
572 		     mdi->mdio_total_disks);
573 	}
574 
575 	return (resurrection);
576 }
577 
578 static void
579 g_disk_md_nvidia_retaste(void *arg, int pending)
580 {
581 
582 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
583 	g_retaste(&g_raid_class);
584 	free(arg, M_MD_NVIDIA);
585 }
586 
587 static void
588 g_raid_md_nvidia_refill(struct g_raid_softc *sc)
589 {
590 	struct g_raid_md_object *md;
591 	struct g_raid_md_nvidia_object *mdi;
592 	struct g_raid_disk *disk;
593 	struct task *task;
594 	int update, na;
595 
596 	md = sc->sc_md;
597 	mdi = (struct g_raid_md_nvidia_object *)md;
598 	update = 0;
599 	do {
600 		/* Make sure we miss anything. */
601 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
602 		if (na == mdi->mdio_total_disks)
603 			break;
604 
605 		G_RAID_DEBUG1(1, md->mdo_softc,
606 		    "Array is not complete (%d of %d), "
607 		    "trying to refill.", na, mdi->mdio_total_disks);
608 
609 		/* Try to get use some of STALE disks. */
610 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
611 			if (disk->d_state == G_RAID_DISK_S_STALE) {
612 				update += g_raid_md_nvidia_start_disk(disk);
613 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
614 					break;
615 			}
616 		}
617 		if (disk != NULL)
618 			continue;
619 
620 		/* Try to get use some of SPARE disks. */
621 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
622 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
623 				update += g_raid_md_nvidia_start_disk(disk);
624 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
625 					break;
626 			}
627 		}
628 	} while (disk != NULL);
629 
630 	/* Write new metadata if we changed something. */
631 	if (update)
632 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
633 
634 	/* Update status of our need for spare. */
635 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
636 	    mdi->mdio_total_disks);
637 
638 	/* Request retaste hoping to find spare. */
639 	if (mdi->mdio_incomplete) {
640 		task = malloc(sizeof(struct task),
641 		    M_MD_NVIDIA, M_WAITOK | M_ZERO);
642 		TASK_INIT(task, 0, g_disk_md_nvidia_retaste, task);
643 		taskqueue_enqueue(taskqueue_swi, task);
644 	}
645 }
646 
647 static void
648 g_raid_md_nvidia_start(struct g_raid_softc *sc)
649 {
650 	struct g_raid_md_object *md;
651 	struct g_raid_md_nvidia_object *mdi;
652 	struct g_raid_md_nvidia_perdisk *pd;
653 	struct nvidia_raid_conf *meta;
654 	struct g_raid_volume *vol;
655 	struct g_raid_subdisk *sd;
656 	struct g_raid_disk *disk;
657 	off_t size;
658 	int j, disk_pos;
659 	char buf[17];
660 
661 	md = sc->sc_md;
662 	mdi = (struct g_raid_md_nvidia_object *)md;
663 	meta = mdi->mdio_meta;
664 
665 	/* Create volumes and subdisks. */
666 	nvidia_meta_get_name(meta, buf);
667 	vol = g_raid_create_volume(sc, buf, -1);
668 	vol->v_mediasize = (off_t)meta->total_sectors * 512;
669 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
670 	if (meta->type == NVIDIA_T_RAID0) {
671 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
672 		size = vol->v_mediasize / mdi->mdio_total_disks;
673 	} else if (meta->type == NVIDIA_T_RAID1) {
674 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
675 		size = vol->v_mediasize;
676 	} else if (meta->type == NVIDIA_T_RAID01) {
677 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
678 		size = vol->v_mediasize / (mdi->mdio_total_disks / 2);
679 	} else if (meta->type == NVIDIA_T_CONCAT) {
680 		if (mdi->mdio_total_disks == 1)
681 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
682 		else
683 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
684 		size = 0;
685 	} else if (meta->type == NVIDIA_T_RAID5) {
686 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
687 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
688 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
689 	} else if (meta->type == NVIDIA_T_RAID5_SYM) {
690 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
691 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LS;
692 		size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
693 	} else {
694 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
695 		size = 0;
696 	}
697 	vol->v_strip_size = meta->strip_sectors * 512; //ZZZ
698 	vol->v_disks_count = mdi->mdio_total_disks;
699 	vol->v_sectorsize = 512; //ZZZ
700 	for (j = 0; j < vol->v_disks_count; j++) {
701 		sd = &vol->v_subdisks[j];
702 		sd->sd_offset = 0;
703 		sd->sd_size = size;
704 	}
705 	g_raid_start_volume(vol);
706 
707 	/* Create disk placeholders to store data for later writing. */
708 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
709 		pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
710 		pd->pd_disk_pos = disk_pos;
711 		disk = g_raid_create_disk(sc);
712 		disk->d_md_data = (void *)pd;
713 		disk->d_state = G_RAID_DISK_S_OFFLINE;
714 		sd = &vol->v_subdisks[disk_pos];
715 		sd->sd_disk = disk;
716 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
717 	}
718 
719 	/* Make all disks found till the moment take their places. */
720 	do {
721 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
722 			if (disk->d_state == G_RAID_DISK_S_NONE) {
723 				g_raid_md_nvidia_start_disk(disk);
724 				break;
725 			}
726 		}
727 	} while (disk != NULL);
728 
729 	mdi->mdio_started = 1;
730 	G_RAID_DEBUG1(0, sc, "Array started.");
731 	g_raid_md_write_nvidia(md, NULL, NULL, NULL);
732 
733 	/* Pickup any STALE/SPARE disks to refill array if needed. */
734 	g_raid_md_nvidia_refill(sc);
735 
736 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
737 
738 	callout_stop(&mdi->mdio_start_co);
739 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
740 	root_mount_rel(mdi->mdio_rootmount);
741 	mdi->mdio_rootmount = NULL;
742 }
743 
744 static void
745 g_raid_md_nvidia_new_disk(struct g_raid_disk *disk)
746 {
747 	struct g_raid_softc *sc;
748 	struct g_raid_md_object *md;
749 	struct g_raid_md_nvidia_object *mdi;
750 	struct nvidia_raid_conf *pdmeta;
751 	struct g_raid_md_nvidia_perdisk *pd;
752 
753 	sc = disk->d_softc;
754 	md = sc->sc_md;
755 	mdi = (struct g_raid_md_nvidia_object *)md;
756 	pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
757 	pdmeta = pd->pd_meta;
758 
759 	if (mdi->mdio_started) {
760 		if (g_raid_md_nvidia_start_disk(disk))
761 			g_raid_md_write_nvidia(md, NULL, NULL, NULL);
762 	} else {
763 		if (mdi->mdio_meta == NULL ||
764 		    mdi->mdio_meta->disk_number >= mdi->mdio_meta->total_disks) {
765 			G_RAID_DEBUG1(1, sc, "Newer disk");
766 			if (mdi->mdio_meta != NULL)
767 				free(mdi->mdio_meta, M_MD_NVIDIA);
768 			mdi->mdio_meta = nvidia_meta_copy(pdmeta);
769 			mdi->mdio_total_disks = pdmeta->total_disks;
770 			mdi->mdio_disks_present = 1;
771 		} else if (pdmeta->disk_number < mdi->mdio_meta->total_disks) {
772 			mdi->mdio_disks_present++;
773 			G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
774 			    mdi->mdio_disks_present,
775 			    mdi->mdio_total_disks);
776 		} else
777 			G_RAID_DEBUG1(1, sc, "Spare disk");
778 
779 		/* If we collected all needed disks - start array. */
780 		if (mdi->mdio_disks_present == mdi->mdio_total_disks)
781 			g_raid_md_nvidia_start(sc);
782 	}
783 }
784 
785 static void
786 g_raid_nvidia_go(void *arg)
787 {
788 	struct g_raid_softc *sc;
789 	struct g_raid_md_object *md;
790 	struct g_raid_md_nvidia_object *mdi;
791 
792 	sc = arg;
793 	md = sc->sc_md;
794 	mdi = (struct g_raid_md_nvidia_object *)md;
795 	if (!mdi->mdio_started) {
796 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
797 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
798 	}
799 }
800 
801 static int
802 g_raid_md_create_nvidia(struct g_raid_md_object *md, struct g_class *mp,
803     struct g_geom **gp)
804 {
805 	struct g_raid_softc *sc;
806 	struct g_raid_md_nvidia_object *mdi;
807 	char name[32];
808 
809 	mdi = (struct g_raid_md_nvidia_object *)md;
810 	arc4rand(&mdi->mdio_volume_id, 16, 0);
811 	snprintf(name, sizeof(name), "NVIDIA-%d",
812 	    atomic_fetchadd_int(&NVIDIANodeID, 1));
813 	sc = g_raid_create_node(mp, name, md);
814 	if (sc == NULL)
815 		return (G_RAID_MD_TASTE_FAIL);
816 	md->mdo_softc = sc;
817 	*gp = sc->sc_geom;
818 	return (G_RAID_MD_TASTE_NEW);
819 }
820 
821 static int
822 g_raid_md_taste_nvidia(struct g_raid_md_object *md, struct g_class *mp,
823                               struct g_consumer *cp, struct g_geom **gp)
824 {
825 	struct g_consumer *rcp;
826 	struct g_provider *pp;
827 	struct g_raid_md_nvidia_object *mdi, *mdi1;
828 	struct g_raid_softc *sc;
829 	struct g_raid_disk *disk;
830 	struct nvidia_raid_conf *meta;
831 	struct g_raid_md_nvidia_perdisk *pd;
832 	struct g_geom *geom;
833 	int result, spare, len;
834 	char name[32];
835 	uint16_t vendor;
836 
837 	G_RAID_DEBUG(1, "Tasting NVIDIA on %s", cp->provider->name);
838 	mdi = (struct g_raid_md_nvidia_object *)md;
839 	pp = cp->provider;
840 
841 	/* Read metadata from device. */
842 	meta = NULL;
843 	vendor = 0xffff;
844 	if (g_access(cp, 1, 0, 0) != 0)
845 		return (G_RAID_MD_TASTE_FAIL);
846 	g_topology_unlock();
847 	len = 2;
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 	g_access(cp, -1, 0, 0);
853 	if (meta == NULL) {
854 		if (g_raid_aggressive_spare) {
855 			if (vendor == 0x10de) {
856 				G_RAID_DEBUG(1,
857 				    "No NVIDIA metadata, forcing spare.");
858 				spare = 2;
859 				goto search;
860 			} else {
861 				G_RAID_DEBUG(1,
862 				    "NVIDIA vendor mismatch 0x%04x != 0x10de",
863 				    vendor);
864 			}
865 		}
866 		return (G_RAID_MD_TASTE_FAIL);
867 	}
868 
869 	/* Metadata valid. Print it. */
870 	g_raid_md_nvidia_print(meta);
871 	G_RAID_DEBUG(1, "NVIDIA disk position %d", meta->disk_number);
872 	spare = 0;//(meta->type == NVIDIA_T_SPARE) ? 1 : 0;
873 
874 search:
875 	/* Search for matching node. */
876 	sc = NULL;
877 	mdi1 = NULL;
878 	LIST_FOREACH(geom, &mp->geom, geom) {
879 		sc = geom->softc;
880 		if (sc == NULL)
881 			continue;
882 		if (sc->sc_stopping != 0)
883 			continue;
884 		if (sc->sc_md->mdo_class != md->mdo_class)
885 			continue;
886 		mdi1 = (struct g_raid_md_nvidia_object *)sc->sc_md;
887 		if (spare) {
888 			if (mdi1->mdio_incomplete)
889 				break;
890 		} else {
891 			if (memcmp(&mdi1->mdio_volume_id,
892 			     &meta->volume_id, 16) == 0)
893 				break;
894 		}
895 	}
896 
897 	/* Found matching node. */
898 	if (geom != NULL) {
899 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
900 		result = G_RAID_MD_TASTE_EXISTING;
901 
902 	} else if (spare) { /* Not found needy node -- left for later. */
903 		G_RAID_DEBUG(1, "Spare is not needed at this time");
904 		goto fail1;
905 
906 	} else { /* Not found matching node -- create one. */
907 		result = G_RAID_MD_TASTE_NEW;
908 		memcpy(&mdi->mdio_volume_id, &meta->volume_id, 16);
909 		snprintf(name, sizeof(name), "NVIDIA-%d",
910 		    atomic_fetchadd_int(&NVIDIANodeID, 1));
911 		sc = g_raid_create_node(mp, name, md);
912 		md->mdo_softc = sc;
913 		geom = sc->sc_geom;
914 		callout_init(&mdi->mdio_start_co, 1);
915 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
916 		    g_raid_nvidia_go, sc);
917 		mdi->mdio_rootmount = root_mount_hold("GRAID-NVIDIA");
918 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
919 	}
920 
921 	rcp = g_new_consumer(geom);
922 	rcp->flags |= G_CF_DIRECT_RECEIVE;
923 	g_attach(rcp, pp);
924 	if (g_access(rcp, 1, 1, 1) != 0)
925 		; //goto fail1;
926 
927 	g_topology_unlock();
928 	sx_xlock(&sc->sc_lock);
929 
930 	pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
931 	pd->pd_meta = meta;
932 	if (spare == 2) {
933 		pd->pd_disk_pos = -3;
934 	} else {
935 		pd->pd_disk_pos = -1;
936 	}
937 	pd->pd_disk_size = pp->mediasize;
938 	disk = g_raid_create_disk(sc);
939 	disk->d_md_data = (void *)pd;
940 	disk->d_consumer = rcp;
941 	rcp->private = disk;
942 
943 	g_raid_get_disk_info(disk);
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 (strcasecmp(levelname, "RAID5") == 0)
1063 			levelname = "RAID5-LS";
1064 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1065 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1066 			return (-4);
1067 		}
1068 		numdisks = *nargs - 3;
1069 		force = gctl_get_paraml(req, "force", sizeof(*force));
1070 		if (!g_raid_md_nvidia_supported(level, qual, numdisks,
1071 		    force ? *force : 0)) {
1072 			gctl_error(req, "Unsupported RAID level "
1073 			    "(0x%02x/0x%02x), or number of disks (%d).",
1074 			    level, qual, numdisks);
1075 			return (-5);
1076 		}
1077 
1078 		/* Search for disks, connect them and probe. */
1079 		size = 0x7fffffffffffffffllu;
1080 		sectorsize = 0;
1081 		for (i = 0; i < numdisks; i++) {
1082 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1083 			diskname = gctl_get_asciiparam(req, arg);
1084 			if (diskname == NULL) {
1085 				gctl_error(req, "No disk name (%s).", arg);
1086 				error = -6;
1087 				break;
1088 			}
1089 			if (strcmp(diskname, "NONE") == 0) {
1090 				cp = NULL;
1091 				pp = NULL;
1092 			} else {
1093 				g_topology_lock();
1094 				cp = g_raid_open_consumer(sc, diskname);
1095 				if (cp == NULL) {
1096 					gctl_error(req, "Can't open '%s'.",
1097 					    diskname);
1098 					g_topology_unlock();
1099 					error = -7;
1100 					break;
1101 				}
1102 				pp = cp->provider;
1103 			}
1104 			pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1105 			pd->pd_disk_pos = i;
1106 			disk = g_raid_create_disk(sc);
1107 			disk->d_md_data = (void *)pd;
1108 			disk->d_consumer = cp;
1109 			if (cp == NULL)
1110 				continue;
1111 			cp->private = disk;
1112 			g_topology_unlock();
1113 
1114 			g_raid_get_disk_info(disk);
1115 
1116 			pd->pd_disk_size = pp->mediasize;
1117 			if (size > pp->mediasize)
1118 				size = pp->mediasize;
1119 			if (sectorsize < pp->sectorsize)
1120 				sectorsize = pp->sectorsize;
1121 		}
1122 		if (error != 0)
1123 			return (error);
1124 
1125 		if (sectorsize <= 0) {
1126 			gctl_error(req, "Can't get sector size.");
1127 			return (-8);
1128 		}
1129 
1130 		/* Reserve space for metadata. */
1131 		size -= 2 * sectorsize;
1132 
1133 		/* Handle size argument. */
1134 		len = sizeof(*sizearg);
1135 		sizearg = gctl_get_param(req, "size", &len);
1136 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1137 		    *sizearg > 0) {
1138 			if (*sizearg > size) {
1139 				gctl_error(req, "Size too big %lld > %lld.",
1140 				    (long long)*sizearg, (long long)size);
1141 				return (-9);
1142 			}
1143 			size = *sizearg;
1144 		}
1145 
1146 		/* Handle strip argument. */
1147 		strip = 131072;
1148 		len = sizeof(*striparg);
1149 		striparg = gctl_get_param(req, "strip", &len);
1150 		if (striparg != NULL && len == sizeof(*striparg) &&
1151 		    *striparg > 0) {
1152 			if (*striparg < sectorsize) {
1153 				gctl_error(req, "Strip size too small.");
1154 				return (-10);
1155 			}
1156 			if (*striparg % sectorsize != 0) {
1157 				gctl_error(req, "Incorrect strip size.");
1158 				return (-11);
1159 			}
1160 			if (strip > 65535 * sectorsize) {
1161 				gctl_error(req, "Strip size too big.");
1162 				return (-12);
1163 			}
1164 			strip = *striparg;
1165 		}
1166 
1167 		/* Round size down to strip or sector. */
1168 		if (level == G_RAID_VOLUME_RL_RAID1)
1169 			size -= (size % sectorsize);
1170 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1171 		    (numdisks & 1) != 0)
1172 			size -= (size % (2 * strip));
1173 		else
1174 			size -= (size % strip);
1175 		if (size <= 0) {
1176 			gctl_error(req, "Size too small.");
1177 			return (-13);
1178 		}
1179 
1180 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1181 		    level == G_RAID_VOLUME_RL_CONCAT ||
1182 		    level == G_RAID_VOLUME_RL_SINGLE)
1183 			volsize = size * numdisks;
1184 		else if (level == G_RAID_VOLUME_RL_RAID1)
1185 			volsize = size;
1186 		else if (level == G_RAID_VOLUME_RL_RAID5)
1187 			volsize = size * (numdisks - 1);
1188 		else { /* RAID1E */
1189 			volsize = ((size * numdisks) / strip / 2) *
1190 			    strip;
1191 		}
1192 		if (volsize > 0xffffffffllu * sectorsize) {
1193 			gctl_error(req, "Size too big.");
1194 			return (-14);
1195 		}
1196 
1197 		/* We have all we need, create things: volume, ... */
1198 		mdi->mdio_total_disks = numdisks;
1199 		mdi->mdio_started = 1;
1200 		vol = g_raid_create_volume(sc, volname, -1);
1201 		vol->v_md_data = (void *)(intptr_t)0;
1202 		vol->v_raid_level = level;
1203 		vol->v_raid_level_qualifier = qual;
1204 		vol->v_strip_size = strip;
1205 		vol->v_disks_count = numdisks;
1206 		vol->v_mediasize = volsize;
1207 		vol->v_sectorsize = sectorsize;
1208 		g_raid_start_volume(vol);
1209 
1210 		/* , and subdisks. */
1211 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1212 			pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1213 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1214 			sd->sd_disk = disk;
1215 			sd->sd_offset = 0;
1216 			sd->sd_size = size;
1217 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1218 			if (sd->sd_disk->d_consumer != NULL) {
1219 				g_raid_change_disk_state(disk,
1220 				    G_RAID_DISK_S_ACTIVE);
1221 				g_raid_change_subdisk_state(sd,
1222 				    G_RAID_SUBDISK_S_ACTIVE);
1223 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1224 				    G_RAID_EVENT_SUBDISK);
1225 			} else {
1226 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1227 			}
1228 		}
1229 
1230 		/* Write metadata based on created entities. */
1231 		G_RAID_DEBUG1(0, sc, "Array started.");
1232 		g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1233 
1234 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1235 		g_raid_md_nvidia_refill(sc);
1236 
1237 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1238 		    G_RAID_EVENT_VOLUME);
1239 		return (0);
1240 	}
1241 	if (strcmp(verb, "delete") == 0) {
1242 
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, "/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