vdev_geom.c revision 185029 
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
23 * All rights reserved.
24 */
25
26#include <sys/zfs_context.h>
27#include <sys/param.h>
28#include <sys/kernel.h>
29#include <sys/bio.h>
30#include <sys/disk.h>
31#include <sys/spa.h>
32#include <sys/vdev_impl.h>
33#include <sys/fs/zfs.h>
34#include <sys/zio.h>
35#include <geom/geom.h>
36#include <geom/geom_int.h>
37
38/*
39 * Virtual device vector for GEOM.
40 */
41
42struct g_class zfs_vdev_class = {
43	.name = "ZFS::VDEV",
44	.version = G_VERSION,
45};
46
47DECLARE_GEOM_CLASS(zfs_vdev_class, zfs_vdev);
48
49typedef struct vdev_geom_ctx {
50	struct g_consumer *gc_consumer;
51	int gc_state;
52	struct bio_queue_head gc_queue;
53	struct mtx gc_queue_mtx;
54} vdev_geom_ctx_t;
55
56static void
57vdev_geom_release(vdev_t *vd)
58{
59	vdev_geom_ctx_t *ctx;
60
61	ctx = vd->vdev_tsd;
62	vd->vdev_tsd = NULL;
63
64	mtx_lock(&ctx->gc_queue_mtx);
65	ctx->gc_state = 1;
66	wakeup_one(&ctx->gc_queue);
67	while (ctx->gc_state != 2)
68		msleep(&ctx->gc_state, &ctx->gc_queue_mtx, 0, "vgeom:w", 0);
69	mtx_unlock(&ctx->gc_queue_mtx);
70	mtx_destroy(&ctx->gc_queue_mtx);
71	kmem_free(ctx, sizeof(*ctx));
72}
73
74static void
75vdev_geom_orphan(struct g_consumer *cp)
76{
77	struct g_geom *gp;
78	vdev_t *vd;
79	int error;
80
81	g_topology_assert();
82
83	vd = cp->private;
84	gp = cp->geom;
85	error = cp->provider->error;
86
87	ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
88	if (cp->acr + cp->acw + cp->ace > 0)
89		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
90	ZFS_LOG(1, "Destroyed consumer to %s.", cp->provider->name);
91	g_detach(cp);
92	g_destroy_consumer(cp);
93	/* Destroy geom if there are no consumers left. */
94	if (LIST_EMPTY(&gp->consumer)) {
95		ZFS_LOG(1, "Destroyed geom %s.", gp->name);
96		g_wither_geom(gp, error);
97	}
98	vdev_geom_release(vd);
99
100	vd->vdev_remove_wanted = B_TRUE;
101	spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
102}
103
104static struct g_consumer *
105vdev_geom_attach(struct g_provider *pp, int write)
106{
107	struct g_geom *gp;
108	struct g_consumer *cp;
109
110	g_topology_assert();
111
112	ZFS_LOG(1, "Attaching to %s.", pp->name);
113	/* Do we have geom already? No? Create one. */
114	LIST_FOREACH(gp, &zfs_vdev_class.geom, geom) {
115		if (gp->flags & G_GEOM_WITHER)
116			continue;
117		if (strcmp(gp->name, "zfs::vdev") != 0)
118			continue;
119		break;
120	}
121	if (gp == NULL) {
122		gp = g_new_geomf(&zfs_vdev_class, "zfs::vdev");
123		gp->orphan = vdev_geom_orphan;
124		cp = g_new_consumer(gp);
125		if (g_attach(cp, pp) != 0) {
126			g_wither_geom(gp, ENXIO);
127			return (NULL);
128		}
129		if (g_access(cp, 1, write, 1) != 0) {
130			g_wither_geom(gp, ENXIO);
131			return (NULL);
132		}
133		ZFS_LOG(1, "Created geom and consumer for %s.", pp->name);
134	} else {
135		/* Check if we are already connected to this provider. */
136		LIST_FOREACH(cp, &gp->consumer, consumer) {
137			if (cp->provider == pp) {
138				ZFS_LOG(1, "Found consumer for %s.", pp->name);
139				break;
140			}
141		}
142		if (cp == NULL) {
143			cp = g_new_consumer(gp);
144			if (g_attach(cp, pp) != 0) {
145				g_destroy_consumer(cp);
146				return (NULL);
147			}
148			if (g_access(cp, 1, write, 1) != 0) {
149				g_detach(cp);
150				g_destroy_consumer(cp);
151				return (NULL);
152			}
153			ZFS_LOG(1, "Created consumer for %s.", pp->name);
154		} else {
155			if (g_access(cp, 1, cp->acw > 0 ? 0 : write, 1) != 0)
156				return (NULL);
157			ZFS_LOG(1, "Used existing consumer for %s.", pp->name);
158		}
159	}
160	return (cp);
161}
162
163static void
164vdev_geom_detach(void *arg, int flag __unused)
165{
166	struct g_geom *gp;
167	struct g_consumer *cp;
168
169	g_topology_assert();
170	cp = arg;
171	gp = cp->geom;
172
173	ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
174	g_access(cp, -1, 0, -1);
175	/* Destroy consumer on last close. */
176	if (cp->acr == 0 && cp->ace == 0) {
177		ZFS_LOG(1, "Destroyed consumer to %s.", cp->provider->name);
178		if (cp->acw > 0)
179			g_access(cp, 0, -cp->acw, 0);
180		g_detach(cp);
181		g_destroy_consumer(cp);
182	}
183	/* Destroy geom if there are no consumers left. */
184	if (LIST_EMPTY(&gp->consumer)) {
185		ZFS_LOG(1, "Destroyed geom %s.", gp->name);
186		g_wither_geom(gp, ENXIO);
187	}
188}
189
190static void
191vdev_geom_worker(void *arg)
192{
193	vdev_geom_ctx_t *ctx;
194	zio_t *zio;
195	struct bio *bp;
196
197	ctx = arg;
198	for (;;) {
199		mtx_lock(&ctx->gc_queue_mtx);
200		bp = bioq_takefirst(&ctx->gc_queue);
201		if (bp == NULL) {
202			if (ctx->gc_state == 1) {
203				ctx->gc_state = 2;
204				wakeup_one(&ctx->gc_state);
205				mtx_unlock(&ctx->gc_queue_mtx);
206				kproc_exit(0);
207			}
208			msleep(&ctx->gc_queue, &ctx->gc_queue_mtx,
209			    PRIBIO | PDROP, "vgeom:io", 0);
210			continue;
211		}
212		mtx_unlock(&ctx->gc_queue_mtx);
213		zio = bp->bio_caller1;
214		zio->io_error = bp->bio_error;
215		if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == ENOTSUP) {
216			vdev_t *vd;
217
218			/*
219			 * If we get ENOTSUP, we know that no future
220			 * attempts will ever succeed.  In this case we
221			 * set a persistent bit so that we don't bother
222			 * with the ioctl in the future.
223			 */
224			vd = zio->io_vd;
225			vd->vdev_nowritecache = B_TRUE;
226		}
227		g_destroy_bio(bp);
228		zio_interrupt(zio);
229	}
230}
231
232static char *
233vdev_geom_get_id(struct g_consumer *cp)
234{
235	char *id;
236	int len;
237
238	g_topology_assert_not();
239	len = DISK_IDENT_SIZE;
240	id = kmem_zalloc(len, KM_SLEEP);
241	if (g_io_getattr("GEOM::ident", cp, &len, id) != 0) {
242		kmem_free(id, DISK_IDENT_SIZE);
243		return (NULL);
244	}
245	return (id);
246}
247
248static uint64_t
249nvlist_get_guid(nvlist_t *list)
250{
251	nvpair_t *elem = NULL;
252	uint64_t value;
253
254	while ((elem = nvlist_next_nvpair(list, elem)) != NULL) {
255		if (nvpair_type(elem) == DATA_TYPE_UINT64 &&
256		    strcmp(nvpair_name(elem), "guid") == 0) {
257			VERIFY(nvpair_value_uint64(elem, &value) == 0);
258			return (value);
259		}
260	}
261	return (0);
262}
263
264static char *
265nvlist_get_devid(nvlist_t *list, uint64_t guid)
266{
267	nvpair_t *elem = NULL;
268	int progress;
269	char *id;
270
271	progress = 0;
272	id = NULL;
273
274	while ((elem = nvlist_next_nvpair(list, elem)) != NULL) {
275		switch (nvpair_type(elem)) {
276		case DATA_TYPE_STRING:
277		    {
278			char *value;
279
280			VERIFY(nvpair_value_string(elem, &value) == 0);
281			if (strcmp(nvpair_name(elem), "type") == 0 &&
282			    strcmp(value, "disk") == 0) {
283				progress |= 0x01;
284			} else if (strcmp(nvpair_name(elem), "devid") == 0) {
285				progress |= 0x02;
286				id = value;
287			}
288			break;
289		    }
290		case DATA_TYPE_UINT64:
291		    {
292			uint64_t value;
293
294			VERIFY(nvpair_value_uint64(elem, &value) == 0);
295			if (strcmp(nvpair_name(elem), "guid") == 0 &&
296			    value == guid) {
297				progress |= 0x04;
298			}
299			break;
300		    }
301		case DATA_TYPE_NVLIST:
302		    {
303			nvlist_t *value;
304			char *lid;
305
306			VERIFY(nvpair_value_nvlist(elem, &value) == 0);
307			lid = nvlist_get_devid(value, guid);
308			if (lid != NULL)
309				return (lid);
310			break;
311		    }
312		case DATA_TYPE_NVLIST_ARRAY:
313		    {
314			nvlist_t **value;
315			u_int c, count;
316			char *lid;
317
318			VERIFY(nvpair_value_nvlist_array(elem, &value,
319			    &count) == 0);
320
321			for (c = 0; c < count; c++) {
322				lid = nvlist_get_devid(value[c], guid);
323				if (lid != NULL)
324					return (lid);
325			}
326			break;
327		    }
328		}
329		if (progress == 0x07)
330			break;
331	}
332	if (progress != 0x07)
333		id = NULL;
334	return (id);
335}
336
337static int
338vdev_geom_io(struct g_consumer *cp, int cmd, void *data, off_t offset, off_t size)
339{
340	struct bio *bp;
341	u_char *p;
342	off_t off;
343	int error;
344
345	ASSERT((offset % cp->provider->sectorsize) == 0);
346	ASSERT((size % cp->provider->sectorsize) == 0);
347
348	bp = g_alloc_bio();
349	off = offset;
350	offset += size;
351	p = data;
352	error = 0;
353
354	for (; off < offset; off += MAXPHYS, p += MAXPHYS, size -= MAXPHYS) {
355		bzero(bp, sizeof(*bp));
356		bp->bio_cmd = cmd;
357		bp->bio_done = NULL;
358		bp->bio_offset = off;
359		bp->bio_length = MIN(size, MAXPHYS);
360		bp->bio_data = p;
361		g_io_request(bp, cp);
362		error = biowait(bp, "vdev_geom_io");
363		if (error != 0)
364			break;
365	}
366
367	g_destroy_bio(bp);
368	return (error);
369}
370
371static char *
372vdev_geom_read_id(struct g_consumer *cp)
373{
374	struct g_provider *pp;
375	vdev_label_t *label;
376	char *p, *buf;
377	size_t buflen;
378	uint64_t psize;
379	off_t offset, size;
380	char *id;
381	int error, l, len;
382
383	g_topology_assert_not();
384
385	pp = cp->provider;
386
387	psize = pp->mediasize;
388	psize = P2ALIGN(psize, (uint64_t)sizeof(vdev_label_t));
389
390	size = sizeof(*label) + pp->sectorsize -
391	    ((sizeof(*label) - 1) % pp->sectorsize) - 1;
392
393	id = NULL;
394	label = kmem_alloc(size, KM_SLEEP);
395	buflen = sizeof(label->vl_vdev_phys.vp_nvlist);
396
397	for (l = 0; l < VDEV_LABELS && id == NULL; l++) {
398		nvlist_t *config = NULL;
399		uint64_t guid;
400
401		offset = vdev_label_offset(psize, l, 0);
402		if ((offset % pp->sectorsize) != 0)
403			continue;
404
405		error = vdev_geom_io(cp, BIO_READ, label, offset, size);
406		if (error != 0)
407			continue;
408		buf = label->vl_vdev_phys.vp_nvlist;
409
410		if (nvlist_unpack(buf, buflen, &config, 0) != 0)
411			continue;
412
413		guid = nvlist_get_guid(config);
414		if (guid == 0) {
415			nvlist_free(config);
416			continue;
417		}
418		id = nvlist_get_devid(config, guid);
419		if (id != NULL) {
420			char *tmp;
421
422			tmp = kmem_zalloc(DISK_IDENT_SIZE, KM_SLEEP);
423			strlcpy(tmp, id, DISK_IDENT_SIZE);
424			id = tmp;
425		}
426
427		nvlist_free(config);
428	}
429
430	kmem_free(label, size);
431	if (id != NULL)
432		ZFS_LOG(1, "ID of %s: %s", pp->name, id);
433	return (id);
434}
435
436static void
437vdev_geom_free_id(char *id)
438{
439
440	if (id != NULL)
441		kmem_free(id, DISK_IDENT_SIZE);
442}
443
444struct vdev_geom_find {
445	const char *id;
446	int write;
447	struct g_consumer *cp;
448};
449
450static void
451vdev_geom_taste_orphan(struct g_consumer *cp)
452{
453
454	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
455	    cp->provider->name));
456}
457
458static void
459vdev_geom_attach_by_id_event(void *arg, int flags __unused)
460{
461	struct vdev_geom_find *ap;
462	struct g_class *mp;
463	struct g_geom *gp, *zgp;
464	struct g_provider *pp;
465	struct g_consumer *zcp;
466	char *id;
467
468	g_topology_assert();
469
470	ap = arg;
471
472	zgp = g_new_geomf(&zfs_vdev_class, "zfs::vdev::taste");
473	/* This orphan function should be never called. */
474	zgp->orphan = vdev_geom_taste_orphan;
475	zcp = g_new_consumer(zgp);
476
477	/* First round tries to get provider's ID without reading metadata. */
478	LIST_FOREACH(mp, &g_classes, class) {
479		if (mp == &zfs_vdev_class)
480			continue;
481		LIST_FOREACH(gp, &mp->geom, geom) {
482			if (gp->flags & G_GEOM_WITHER)
483				continue;
484			LIST_FOREACH(pp, &gp->provider, provider) {
485				if (pp->flags & G_PF_WITHER)
486					continue;
487				g_attach(zcp, pp);
488				if (g_access(zcp, 1, 0, 0) != 0) {
489					g_detach(zcp);
490					continue;
491				}
492				g_topology_unlock();
493				id = vdev_geom_get_id(zcp);
494				g_topology_lock();
495				g_access(zcp, -1, 0, 0);
496				g_detach(zcp);
497				if (id == NULL || strcmp(id, ap->id) != 0) {
498					vdev_geom_free_id(id);
499					continue;
500				}
501				vdev_geom_free_id(id);
502				ap->cp = vdev_geom_attach(pp, ap->write);
503				if (ap->cp == NULL) {
504					printf("ZFS WARNING: Cannot open %s "
505					    "for writting.\n", pp->name);
506					continue;
507				}
508				goto end;
509			}
510		}
511	}
512	/* Second round looks for ID by reading ZFS metadata. */
513	LIST_FOREACH(mp, &g_classes, class) {
514		if (mp == &zfs_vdev_class)
515			continue;
516		LIST_FOREACH(gp, &mp->geom, geom) {
517			if (gp->flags & G_GEOM_WITHER)
518				continue;
519			LIST_FOREACH(pp, &gp->provider, provider) {
520				if (pp->flags & G_PF_WITHER)
521					continue;
522				g_attach(zcp, pp);
523				if (g_access(zcp, 1, 0, 0) != 0) {
524					g_detach(zcp);
525					continue;
526				}
527				g_topology_unlock();
528				id = vdev_geom_read_id(zcp);
529				g_topology_lock();
530				g_access(zcp, -1, 0, 0);
531				g_detach(zcp);
532				if (id == NULL || strcmp(id, ap->id) != 0) {
533					vdev_geom_free_id(id);
534					continue;
535				}
536				vdev_geom_free_id(id);
537				ap->cp = vdev_geom_attach(pp, ap->write);
538				if (ap->cp == NULL) {
539					printf("ZFS WARNING: Cannot open %s "
540					    "for writting.\n", pp->name);
541					continue;
542				}
543				goto end;
544			}
545		}
546	}
547	ap->cp = NULL;
548end:
549	g_destroy_consumer(zcp);
550	g_destroy_geom(zgp);
551}
552
553static struct g_consumer *
554vdev_geom_attach_by_id(const char *id, int write)
555{
556	struct vdev_geom_find *ap;
557	struct g_consumer *cp;
558
559	ap = kmem_zalloc(sizeof(*ap), KM_SLEEP);
560	ap->id = id;
561	ap->write = write;
562	g_waitfor_event(vdev_geom_attach_by_id_event, ap, M_WAITOK, NULL);
563	cp = ap->cp;
564	kmem_free(ap, sizeof(*ap));
565	return (cp);
566}
567
568static struct g_consumer *
569vdev_geom_open_by_path_and_devid(vdev_t *vd)
570{
571	struct g_provider *pp;
572	struct g_consumer *cp;
573	char *id;
574
575	cp = NULL;
576	g_topology_lock();
577	pp = g_provider_by_name(vd->vdev_path + sizeof("/dev/") - 1);
578	if (pp != NULL) {
579		ZFS_LOG(1, "Found provider by name %s.", vd->vdev_path);
580		cp = vdev_geom_attach(pp, !!(spa_mode & FWRITE));
581		if (cp != NULL && vd->vdev_devid != NULL) {
582			g_topology_unlock();
583			id = vdev_geom_get_id(cp);
584			g_topology_lock();
585			if (id == NULL || strcmp(id, vd->vdev_devid) != 0) {
586				vdev_geom_detach(cp, 0);
587				cp = NULL;
588				ZFS_LOG(1, "ID mismatch for provider %s: "
589				    "[%s]!=[%s].", vd->vdev_path,
590				    vd->vdev_devid, id);
591			} else
592				ZFS_LOG(1, "ID match for provider %s.",
593				    vd->vdev_path);
594			vdev_geom_free_id(id);
595		}
596	}
597	g_topology_unlock();
598
599	return (cp);
600}
601
602static struct g_consumer *
603vdev_geom_open_by_devid(vdev_t *vd)
604{
605	struct g_consumer *cp;
606	char *buf;
607	size_t len;
608
609	/*
610	 * We can't search by devid if it's missing.
611	 */
612	if (vd->vdev_devid == NULL)
613		return (NULL);
614
615	ZFS_LOG(1, "Searching by ID [%s].", vd->vdev_devid);
616	cp = vdev_geom_attach_by_id(vd->vdev_devid, !!(spa_mode & FWRITE));
617	if (cp != NULL) {
618		len = strlen(cp->provider->name) + strlen("/dev/") + 1;
619		buf = kmem_alloc(len, KM_SLEEP);
620
621		snprintf(buf, len, "/dev/%s", cp->provider->name);
622		spa_strfree(vd->vdev_path);
623		vd->vdev_path = buf;
624
625		ZFS_LOG(1, "Attach by ID [%s] succeeded, provider %s.",
626		    vd->vdev_devid, vd->vdev_path);
627	} else
628		ZFS_LOG(1, "Search by ID [%s] failed.", vd->vdev_devid);
629
630	return (cp);
631}
632
633static int
634vdev_geom_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
635{
636	vdev_geom_ctx_t *ctx;
637	struct g_provider *pp;
638	struct g_consumer *cp;
639	int owned;
640
641	/*
642	 * We must have a pathname, and it must be absolute.
643	 */
644	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
645		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
646		return (EINVAL);
647	}
648
649	vd->vdev_tsd = NULL;
650
651	if ((owned = mtx_owned(&Giant)))
652		mtx_unlock(&Giant);
653	cp = vdev_geom_open_by_path_and_devid(vd);
654	if (cp == NULL) {
655		/*
656		 * The device at vd->vdev_path doesn't have the right devid.
657		 * The disks might have merely moved around so try all other
658		 * geom providers to find one with the right devid.
659		 */
660		cp = vdev_geom_open_by_devid(vd);
661		if (cp == NULL) {
662			ZFS_LOG(1, "Provider %s not found.", vd->vdev_path);
663			vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
664			if (owned)
665				mtx_lock(&Giant);
666			return (EACCES);
667		}
668	}
669	if (owned)
670		mtx_lock(&Giant);
671
672	cp->private = vd;
673
674	ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP);
675	bioq_init(&ctx->gc_queue);
676	mtx_init(&ctx->gc_queue_mtx, "zfs:vdev:geom:queue", NULL, MTX_DEF);
677	ctx->gc_consumer = cp;
678	ctx->gc_state = 0;
679
680	vd->vdev_tsd = ctx;
681	pp = cp->provider;
682
683	kproc_create(vdev_geom_worker, ctx, NULL, 0, 0, "vdev:worker %s",
684	    pp->name);
685
686	/*
687	 * Determine the actual size of the device.
688	 */
689	*psize = pp->mediasize;
690
691	/*
692	 * Determine the device's minimum transfer size.
693	 */
694	*ashift = highbit(MAX(pp->sectorsize, SPA_MINBLOCKSIZE)) - 1;
695
696	/*
697	 * Clear the nowritecache bit, so that on a vdev_reopen() we will
698	 * try again.
699	 */
700	vd->vdev_nowritecache = B_FALSE;
701
702	return (0);
703}
704
705static void
706vdev_geom_close(vdev_t *vd)
707{
708	vdev_geom_ctx_t *ctx;
709	struct g_consumer *cp;
710
711	if ((ctx = vd->vdev_tsd) == NULL)
712		return;
713	if ((cp = ctx->gc_consumer) == NULL)
714		return;
715	vdev_geom_release(vd);
716	g_post_event(vdev_geom_detach, cp, M_WAITOK, NULL);
717}
718
719static void
720vdev_geom_io_intr(struct bio *bp)
721{
722	vdev_geom_ctx_t *ctx;
723	zio_t *zio;
724
725	zio = bp->bio_caller1;
726	ctx = zio->io_vd->vdev_tsd;
727
728	if ((zio->io_error = bp->bio_error) == 0 && bp->bio_resid != 0)
729		zio->io_error = EIO;
730
731	mtx_lock(&ctx->gc_queue_mtx);
732	bioq_insert_tail(&ctx->gc_queue, bp);
733	wakeup_one(&ctx->gc_queue);
734	mtx_unlock(&ctx->gc_queue_mtx);
735}
736
737static int
738vdev_geom_io_start(zio_t *zio)
739{
740	vdev_t *vd;
741	vdev_geom_ctx_t *ctx;
742	struct g_consumer *cp;
743	struct bio *bp;
744	int error;
745
746	cp = NULL;
747
748	vd = zio->io_vd;
749	ctx = vd->vdev_tsd;
750	if (ctx != NULL)
751		cp = ctx->gc_consumer;
752
753	if (zio->io_type == ZIO_TYPE_IOCTL) {
754		/* XXPOLICY */
755		if (!vdev_readable(vd)) {
756			zio->io_error = ENXIO;
757			return (ZIO_PIPELINE_CONTINUE);
758		}
759
760		switch (zio->io_cmd) {
761
762		case DKIOCFLUSHWRITECACHE:
763
764			if (zfs_nocacheflush)
765				break;
766
767			if (vd->vdev_nowritecache) {
768				zio->io_error = ENOTSUP;
769				break;
770			}
771
772			goto sendreq;
773		default:
774			zio->io_error = ENOTSUP;
775		}
776
777		return (ZIO_PIPELINE_CONTINUE);
778	}
779sendreq:
780	if (cp == NULL) {
781		zio->io_error = ENXIO;
782		return (ZIO_PIPELINE_CONTINUE);
783	}
784	bp = g_alloc_bio();
785	bp->bio_caller1 = zio;
786	switch (zio->io_type) {
787	case ZIO_TYPE_READ:
788	case ZIO_TYPE_WRITE:
789		bp->bio_cmd = zio->io_type == ZIO_TYPE_READ ? BIO_READ : BIO_WRITE;
790		bp->bio_data = zio->io_data;
791		bp->bio_offset = zio->io_offset;
792		bp->bio_length = zio->io_size;
793		break;
794	case ZIO_TYPE_IOCTL:
795		bp->bio_cmd = BIO_FLUSH;
796		bp->bio_data = NULL;
797		bp->bio_offset = cp->provider->mediasize;
798		bp->bio_length = 0;
799		break;
800	}
801	bp->bio_done = vdev_geom_io_intr;
802
803	g_io_request(bp, cp);
804
805	return (ZIO_PIPELINE_STOP);
806}
807
808static void
809vdev_geom_io_done(zio_t *zio)
810{
811
812	/*
813	 * If the device returned ENXIO, then attempt we should verify if GEOM
814	 * provider has been removed. If this is the case, then we trigger an
815	 * asynchronous removal of the device.
816	 */
817	if (zio->io_error == ENXIO) {
818		vdev_t *vd = zio->io_vd;
819		vdev_geom_ctx_t *ctx;
820		struct g_provider *pp = NULL;
821
822		ctx = vd->vdev_tsd;
823		if (ctx != NULL && ctx->gc_consumer != NULL)
824			pp = ctx->gc_consumer->provider;
825
826		if (pp == NULL || (pp->flags & G_PF_ORPHAN)) {
827			vd->vdev_remove_wanted = B_TRUE;
828			spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
829		}
830	}
831}
832
833vdev_ops_t vdev_geom_ops = {
834	vdev_geom_open,
835	vdev_geom_close,
836	vdev_default_asize,
837	vdev_geom_io_start,
838	vdev_geom_io_done,
839	NULL,
840	VDEV_TYPE_DISK,		/* name of this vdev type */
841	B_TRUE			/* leaf vdev */
842};
843