vdev_geom.c revision 208142 
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/spa_impl.h>
33#include <sys/vdev_impl.h>
34#include <sys/fs/zfs.h>
35#include <sys/zio.h>
36#include <geom/geom.h>
37#include <geom/geom_int.h>
38
39/*
40 * Virtual device vector for GEOM.
41 */
42
43struct g_class zfs_vdev_class = {
44	.name = "ZFS::VDEV",
45	.version = G_VERSION,
46};
47
48DECLARE_GEOM_CLASS(zfs_vdev_class, zfs_vdev);
49
50static void
51vdev_geom_orphan(struct g_consumer *cp)
52{
53	struct g_geom *gp;
54	vdev_t *vd;
55	int error;
56
57	g_topology_assert();
58
59	vd = cp->private;
60	gp = cp->geom;
61	error = cp->provider->error;
62
63	ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
64	if (cp->acr + cp->acw + cp->ace > 0)
65		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
66	ZFS_LOG(1, "Destroyed consumer to %s.", cp->provider->name);
67	g_detach(cp);
68	g_destroy_consumer(cp);
69	/* Destroy geom if there are no consumers left. */
70	if (LIST_EMPTY(&gp->consumer)) {
71		ZFS_LOG(1, "Destroyed geom %s.", gp->name);
72		g_wither_geom(gp, error);
73	}
74	vd->vdev_tsd = NULL;
75	vd->vdev_remove_wanted = B_TRUE;
76	spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
77}
78
79static struct g_consumer *
80vdev_geom_attach(struct g_provider *pp)
81{
82	struct g_geom *gp;
83	struct g_consumer *cp;
84
85	g_topology_assert();
86
87	ZFS_LOG(1, "Attaching to %s.", pp->name);
88	/* Do we have geom already? No? Create one. */
89	LIST_FOREACH(gp, &zfs_vdev_class.geom, geom) {
90		if (gp->flags & G_GEOM_WITHER)
91			continue;
92		if (strcmp(gp->name, "zfs::vdev") != 0)
93			continue;
94		break;
95	}
96	if (gp == NULL) {
97		gp = g_new_geomf(&zfs_vdev_class, "zfs::vdev");
98		gp->orphan = vdev_geom_orphan;
99		cp = g_new_consumer(gp);
100		if (g_attach(cp, pp) != 0) {
101			g_wither_geom(gp, ENXIO);
102			return (NULL);
103		}
104		if (g_access(cp, 1, 0, 1) != 0) {
105			g_wither_geom(gp, ENXIO);
106			return (NULL);
107		}
108		ZFS_LOG(1, "Created geom and consumer for %s.", pp->name);
109	} else {
110		/* Check if we are already connected to this provider. */
111		LIST_FOREACH(cp, &gp->consumer, consumer) {
112			if (cp->provider == pp) {
113				ZFS_LOG(1, "Found consumer for %s.", pp->name);
114				break;
115			}
116		}
117		if (cp == NULL) {
118			cp = g_new_consumer(gp);
119			if (g_attach(cp, pp) != 0) {
120				g_destroy_consumer(cp);
121				return (NULL);
122			}
123			if (g_access(cp, 1, 0, 1) != 0) {
124				g_detach(cp);
125				g_destroy_consumer(cp);
126				return (NULL);
127			}
128			ZFS_LOG(1, "Created consumer for %s.", pp->name);
129		} else {
130			if (g_access(cp, 1, 0, 1) != 0)
131				return (NULL);
132			ZFS_LOG(1, "Used existing consumer for %s.", pp->name);
133		}
134	}
135	return (cp);
136}
137
138static void
139vdev_geom_detach(void *arg, int flag __unused)
140{
141	struct g_geom *gp;
142	struct g_consumer *cp;
143
144	g_topology_assert();
145	cp = arg;
146	gp = cp->geom;
147
148	ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
149	g_access(cp, -1, 0, -1);
150	/* Destroy consumer on last close. */
151	if (cp->acr == 0 && cp->ace == 0) {
152		ZFS_LOG(1, "Destroyed consumer to %s.", cp->provider->name);
153		if (cp->acw > 0)
154			g_access(cp, 0, -cp->acw, 0);
155		g_detach(cp);
156		g_destroy_consumer(cp);
157	}
158	/* Destroy geom if there are no consumers left. */
159	if (LIST_EMPTY(&gp->consumer)) {
160		ZFS_LOG(1, "Destroyed geom %s.", gp->name);
161		g_wither_geom(gp, ENXIO);
162	}
163}
164
165static uint64_t
166nvlist_get_guid(nvlist_t *list)
167{
168	nvpair_t *elem = NULL;
169	uint64_t value;
170
171	while ((elem = nvlist_next_nvpair(list, elem)) != NULL) {
172		if (nvpair_type(elem) == DATA_TYPE_UINT64 &&
173		    strcmp(nvpair_name(elem), "guid") == 0) {
174			VERIFY(nvpair_value_uint64(elem, &value) == 0);
175			return (value);
176		}
177	}
178	return (0);
179}
180
181static int
182vdev_geom_io(struct g_consumer *cp, int cmd, void *data, off_t offset, off_t size)
183{
184	struct bio *bp;
185	u_char *p;
186	off_t off;
187	int error;
188
189	ASSERT((offset % cp->provider->sectorsize) == 0);
190	ASSERT((size % cp->provider->sectorsize) == 0);
191
192	bp = g_alloc_bio();
193	off = offset;
194	offset += size;
195	p = data;
196	error = 0;
197
198	for (; off < offset; off += MAXPHYS, p += MAXPHYS, size -= MAXPHYS) {
199		bzero(bp, sizeof(*bp));
200		bp->bio_cmd = cmd;
201		bp->bio_done = NULL;
202		bp->bio_offset = off;
203		bp->bio_length = MIN(size, MAXPHYS);
204		bp->bio_data = p;
205		g_io_request(bp, cp);
206		error = biowait(bp, "vdev_geom_io");
207		if (error != 0)
208			break;
209	}
210
211	g_destroy_bio(bp);
212	return (error);
213}
214
215static uint64_t
216vdev_geom_read_guid(struct g_consumer *cp)
217{
218	struct g_provider *pp;
219	vdev_label_t *label;
220	char *p, *buf;
221	size_t buflen;
222	uint64_t psize;
223	off_t offset, size;
224	uint64_t guid;
225	int error, l, len, iszvol;
226
227	g_topology_assert_not();
228
229	pp = cp->provider;
230	ZFS_LOG(1, "Reading guid from %s...", pp->name);
231	if (g_getattr("ZFS::iszvol", cp, &iszvol) == 0 && iszvol) {
232		ZFS_LOG(1, "Skipping ZVOL-based provider %s.", pp->name);
233		return (0);
234	}
235
236	psize = pp->mediasize;
237	psize = P2ALIGN(psize, (uint64_t)sizeof(vdev_label_t));
238
239	size = sizeof(*label) + pp->sectorsize -
240	    ((sizeof(*label) - 1) % pp->sectorsize) - 1;
241
242	guid = 0;
243	label = kmem_alloc(size, KM_SLEEP);
244	buflen = sizeof(label->vl_vdev_phys.vp_nvlist);
245
246	for (l = 0; l < VDEV_LABELS; l++) {
247		nvlist_t *config = NULL;
248
249		offset = vdev_label_offset(psize, l, 0);
250		if ((offset % pp->sectorsize) != 0)
251			continue;
252
253		if (vdev_geom_io(cp, BIO_READ, label, offset, size) != 0)
254			continue;
255		buf = label->vl_vdev_phys.vp_nvlist;
256
257		if (nvlist_unpack(buf, buflen, &config, 0) != 0)
258			continue;
259
260		guid = nvlist_get_guid(config);
261		nvlist_free(config);
262		if (guid != 0)
263			break;
264	}
265
266	kmem_free(label, size);
267	if (guid != 0)
268		ZFS_LOG(1, "guid for %s is %ju", pp->name, (uintmax_t)guid);
269	return (guid);
270}
271
272struct vdev_geom_find {
273	uint64_t guid;
274	struct g_consumer *cp;
275};
276
277static void
278vdev_geom_taste_orphan(struct g_consumer *cp)
279{
280
281	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
282	    cp->provider->name));
283}
284
285static void
286vdev_geom_attach_by_guid_event(void *arg, int flags __unused)
287{
288	struct vdev_geom_find *ap;
289	struct g_class *mp;
290	struct g_geom *gp, *zgp;
291	struct g_provider *pp;
292	struct g_consumer *zcp;
293	uint64_t guid;
294
295	g_topology_assert();
296
297	ap = arg;
298
299	zgp = g_new_geomf(&zfs_vdev_class, "zfs::vdev::taste");
300	/* This orphan function should be never called. */
301	zgp->orphan = vdev_geom_taste_orphan;
302	zcp = g_new_consumer(zgp);
303
304	LIST_FOREACH(mp, &g_classes, class) {
305		if (mp == &zfs_vdev_class)
306			continue;
307		LIST_FOREACH(gp, &mp->geom, geom) {
308			if (gp->flags & G_GEOM_WITHER)
309				continue;
310			LIST_FOREACH(pp, &gp->provider, provider) {
311				if (pp->flags & G_PF_WITHER)
312					continue;
313				g_attach(zcp, pp);
314				if (g_access(zcp, 1, 0, 0) != 0) {
315					g_detach(zcp);
316					continue;
317				}
318				g_topology_unlock();
319				guid = vdev_geom_read_guid(zcp);
320				g_topology_lock();
321				g_access(zcp, -1, 0, 0);
322				g_detach(zcp);
323				if (guid != ap->guid)
324					continue;
325				ap->cp = vdev_geom_attach(pp);
326				if (ap->cp == NULL) {
327					printf("ZFS WARNING: Unable to attach to %s.\n",
328					    pp->name);
329					continue;
330				}
331				goto end;
332			}
333		}
334	}
335	ap->cp = NULL;
336end:
337	g_destroy_consumer(zcp);
338	g_destroy_geom(zgp);
339}
340
341static struct g_consumer *
342vdev_geom_attach_by_guid(uint64_t guid)
343{
344	struct vdev_geom_find *ap;
345	struct g_consumer *cp;
346
347	ap = kmem_zalloc(sizeof(*ap), KM_SLEEP);
348	ap->guid = guid;
349	g_waitfor_event(vdev_geom_attach_by_guid_event, ap, M_WAITOK, NULL);
350	cp = ap->cp;
351	kmem_free(ap, sizeof(*ap));
352	return (cp);
353}
354
355static struct g_consumer *
356vdev_geom_open_by_guid(vdev_t *vd)
357{
358	struct g_consumer *cp;
359	char *buf;
360	size_t len;
361
362	ZFS_LOG(1, "Searching by guid [%ju].", (uintmax_t)vd->vdev_guid);
363	cp = vdev_geom_attach_by_guid(vd->vdev_guid);
364	if (cp != NULL) {
365		len = strlen(cp->provider->name) + strlen("/dev/") + 1;
366		buf = kmem_alloc(len, KM_SLEEP);
367
368		snprintf(buf, len, "/dev/%s", cp->provider->name);
369		spa_strfree(vd->vdev_path);
370		vd->vdev_path = buf;
371
372		ZFS_LOG(1, "Attach by guid [%ju] succeeded, provider %s.",
373		    (uintmax_t)vd->vdev_guid, vd->vdev_path);
374	} else {
375		ZFS_LOG(1, "Search by guid [%ju] failed.",
376		    (uintmax_t)vd->vdev_guid);
377	}
378
379	return (cp);
380}
381
382static struct g_consumer *
383vdev_geom_open_by_path(vdev_t *vd, int check_guid)
384{
385	struct g_provider *pp;
386	struct g_consumer *cp;
387	uint64_t guid;
388
389	cp = NULL;
390	g_topology_lock();
391	pp = g_provider_by_name(vd->vdev_path + sizeof("/dev/") - 1);
392	if (pp != NULL) {
393		ZFS_LOG(1, "Found provider by name %s.", vd->vdev_path);
394		cp = vdev_geom_attach(pp);
395		if (cp != NULL && check_guid) {
396			g_topology_unlock();
397			guid = vdev_geom_read_guid(cp);
398			g_topology_lock();
399			if (guid != vd->vdev_guid) {
400				vdev_geom_detach(cp, 0);
401				cp = NULL;
402				ZFS_LOG(1, "guid mismatch for provider %s: "
403				    "%ju != %ju.", vd->vdev_path,
404				    (uintmax_t)vd->vdev_guid, (uintmax_t)guid);
405			} else {
406				ZFS_LOG(1, "guid match for provider %s.",
407				    vd->vdev_path);
408			}
409		}
410	}
411	g_topology_unlock();
412
413	return (cp);
414}
415
416static int
417vdev_geom_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
418{
419	struct g_provider *pp;
420	struct g_consumer *cp;
421	int error, owned;
422
423	/*
424	 * We must have a pathname, and it must be absolute.
425	 */
426	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
427		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
428		return (EINVAL);
429	}
430
431	vd->vdev_tsd = NULL;
432
433	if ((owned = mtx_owned(&Giant)))
434		mtx_unlock(&Giant);
435	error = 0;
436
437	/*
438	 * If we're creating pool, just find GEOM provider by its name
439	 * and ignore GUID mismatches.
440	 */
441	if (vd->vdev_spa->spa_load_state == SPA_LOAD_NONE)
442		cp = vdev_geom_open_by_path(vd, 0);
443	else {
444		cp = vdev_geom_open_by_path(vd, 1);
445		if (cp == NULL) {
446			/*
447			 * The device at vd->vdev_path doesn't have the
448			 * expected guid. The disks might have merely
449			 * moved around so try all other GEOM providers
450			 * to find one with the right guid.
451			 */
452			cp = vdev_geom_open_by_guid(vd);
453		}
454	}
455
456	if (cp == NULL) {
457		ZFS_LOG(1, "Provider %s not found.", vd->vdev_path);
458		error = ENOENT;
459	} else if (cp->acw == 0 && (spa_mode & FWRITE) != 0) {
460		int i;
461
462		g_topology_lock();
463		for (i = 0; i < 5; i++) {
464			error = g_access(cp, 0, 1, 0);
465			if (error == 0)
466				break;
467			g_topology_unlock();
468			tsleep(vd, 0, "vdev", hz / 2);
469			g_topology_lock();
470		}
471		if (error != 0) {
472			printf("ZFS WARNING: Unable to open %s for writing (error=%d).\n",
473			    vd->vdev_path, error);
474			vdev_geom_detach(cp, 0);
475			cp = NULL;
476		}
477		g_topology_unlock();
478	}
479	if (owned)
480		mtx_lock(&Giant);
481	if (cp == NULL) {
482		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
483		return (error);
484	}
485
486	cp->private = vd;
487	vd->vdev_tsd = cp;
488	pp = cp->provider;
489
490	/*
491	 * Determine the actual size of the device.
492	 */
493	*psize = pp->mediasize;
494
495	/*
496	 * Determine the device's minimum transfer size.
497	 */
498	*ashift = highbit(MAX(pp->sectorsize, SPA_MINBLOCKSIZE)) - 1;
499
500	/*
501	 * Clear the nowritecache bit, so that on a vdev_reopen() we will
502	 * try again.
503	 */
504	vd->vdev_nowritecache = B_FALSE;
505
506	return (0);
507}
508
509static void
510vdev_geom_close(vdev_t *vd)
511{
512	struct g_consumer *cp;
513
514	cp = vd->vdev_tsd;
515	if (cp == NULL)
516		return;
517	vd->vdev_tsd = NULL;
518	g_post_event(vdev_geom_detach, cp, M_WAITOK, NULL);
519}
520
521static void
522vdev_geom_io_intr(struct bio *bp)
523{
524	zio_t *zio;
525
526	zio = bp->bio_caller1;
527	zio->io_error = bp->bio_error;
528	if (zio->io_error == 0 && bp->bio_resid != 0)
529		zio->io_error = EIO;
530	if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == ENOTSUP) {
531		vdev_t *vd;
532
533		/*
534		 * If we get ENOTSUP, we know that no future
535		 * attempts will ever succeed.  In this case we
536		 * set a persistent bit so that we don't bother
537		 * with the ioctl in the future.
538		 */
539		vd = zio->io_vd;
540		vd->vdev_nowritecache = B_TRUE;
541	}
542	g_destroy_bio(bp);
543	zio_interrupt(zio);
544}
545
546static int
547vdev_geom_io_start(zio_t *zio)
548{
549	vdev_t *vd;
550	struct g_consumer *cp;
551	struct bio *bp;
552	int error;
553
554	vd = zio->io_vd;
555
556	if (zio->io_type == ZIO_TYPE_IOCTL) {
557		/* XXPOLICY */
558		if (!vdev_readable(vd)) {
559			zio->io_error = ENXIO;
560			return (ZIO_PIPELINE_CONTINUE);
561		}
562
563		switch (zio->io_cmd) {
564
565		case DKIOCFLUSHWRITECACHE:
566
567			if (zfs_nocacheflush)
568				break;
569
570			if (vd->vdev_nowritecache) {
571				zio->io_error = ENOTSUP;
572				break;
573			}
574
575			goto sendreq;
576		default:
577			zio->io_error = ENOTSUP;
578		}
579
580		return (ZIO_PIPELINE_CONTINUE);
581	}
582sendreq:
583	cp = vd->vdev_tsd;
584	if (cp == NULL) {
585		zio->io_error = ENXIO;
586		return (ZIO_PIPELINE_CONTINUE);
587	}
588	bp = g_alloc_bio();
589	bp->bio_caller1 = zio;
590	switch (zio->io_type) {
591	case ZIO_TYPE_READ:
592	case ZIO_TYPE_WRITE:
593		bp->bio_cmd = zio->io_type == ZIO_TYPE_READ ? BIO_READ : BIO_WRITE;
594		bp->bio_data = zio->io_data;
595		bp->bio_offset = zio->io_offset;
596		bp->bio_length = zio->io_size;
597		break;
598	case ZIO_TYPE_IOCTL:
599		bp->bio_cmd = BIO_FLUSH;
600		bp->bio_data = NULL;
601		bp->bio_offset = cp->provider->mediasize;
602		bp->bio_length = 0;
603		break;
604	}
605	bp->bio_done = vdev_geom_io_intr;
606
607	g_io_request(bp, cp);
608
609	return (ZIO_PIPELINE_STOP);
610}
611
612static void
613vdev_geom_io_done(zio_t *zio)
614{
615}
616
617vdev_ops_t vdev_geom_ops = {
618	vdev_geom_open,
619	vdev_geom_close,
620	vdev_default_asize,
621	vdev_geom_io_start,
622	vdev_geom_io_done,
623	NULL,
624	VDEV_TYPE_DISK,		/* name of this vdev type */
625	B_TRUE			/* leaf vdev */
626};
627