2 * Copyright (c) 2004-2005 Pawel Jakub Dawidek <pjd@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
35 #include <sys/mutex.h>
38 #include <sys/sysctl.h>
39 #include <sys/malloc.h>
41 #include <geom/geom.h>
42 #include <geom/stripe/g_stripe.h>
44 FEATURE(geom_stripe, "GEOM striping support");
46 static MALLOC_DEFINE(M_STRIPE, "stripe_data", "GEOM_STRIPE Data");
48 static uma_zone_t g_stripe_zone;
50 static int g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force);
51 static int g_stripe_destroy_geom(struct gctl_req *req, struct g_class *mp,
54 static g_taste_t g_stripe_taste;
55 static g_ctl_req_t g_stripe_config;
56 static g_dumpconf_t g_stripe_dumpconf;
57 static g_init_t g_stripe_init;
58 static g_fini_t g_stripe_fini;
60 struct g_class g_stripe_class = {
61 .name = G_STRIPE_CLASS_NAME,
63 .ctlreq = g_stripe_config,
64 .taste = g_stripe_taste,
65 .destroy_geom = g_stripe_destroy_geom,
66 .init = g_stripe_init,
70 SYSCTL_DECL(_kern_geom);
71 static SYSCTL_NODE(_kern_geom, OID_AUTO, stripe, CTLFLAG_RW, 0,
73 static u_int g_stripe_debug = 0;
74 TUNABLE_INT("kern.geom.stripe.debug", &g_stripe_debug);
75 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, debug, CTLFLAG_RW, &g_stripe_debug, 0,
77 static int g_stripe_fast = 0;
78 TUNABLE_INT("kern.geom.stripe.fast", &g_stripe_fast);
80 g_sysctl_stripe_fast(SYSCTL_HANDLER_ARGS)
85 error = sysctl_handle_int(oidp, &fast, 0, req);
86 if (error == 0 && req->newptr != NULL)
90 SYSCTL_PROC(_kern_geom_stripe, OID_AUTO, fast, CTLTYPE_INT | CTLFLAG_RW,
91 NULL, 0, g_sysctl_stripe_fast, "I", "Fast, but memory-consuming, mode");
92 static u_int g_stripe_maxmem = MAXPHYS * 100;
93 TUNABLE_INT("kern.geom.stripe.maxmem", &g_stripe_maxmem);
94 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, maxmem, CTLFLAG_RD, &g_stripe_maxmem,
95 0, "Maximum memory that can be allocated in \"fast\" mode (in bytes)");
96 static u_int g_stripe_fast_failed = 0;
97 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, fast_failed, CTLFLAG_RD,
98 &g_stripe_fast_failed, 0, "How many times \"fast\" mode failed");
101 * Greatest Common Divisor.
104 gcd(u_int a, u_int b)
117 * Least Common Multiple.
120 lcm(u_int a, u_int b)
123 return ((a * b) / gcd(a, b));
127 g_stripe_init(struct g_class *mp __unused)
130 g_stripe_zone = uma_zcreate("g_stripe_zone", MAXPHYS, NULL, NULL,
132 g_stripe_maxmem -= g_stripe_maxmem % MAXPHYS;
133 uma_zone_set_max(g_stripe_zone, g_stripe_maxmem / MAXPHYS);
137 g_stripe_fini(struct g_class *mp __unused)
140 uma_zdestroy(g_stripe_zone);
144 * Return the number of valid disks.
147 g_stripe_nvalid(struct g_stripe_softc *sc)
152 for (i = 0; i < sc->sc_ndisks; i++) {
153 if (sc->sc_disks[i] != NULL)
161 g_stripe_remove_disk(struct g_consumer *cp)
163 struct g_stripe_softc *sc;
166 KASSERT(cp != NULL, ("Non-valid disk in %s.", __func__));
167 sc = (struct g_stripe_softc *)cp->geom->softc;
168 KASSERT(sc != NULL, ("NULL sc in %s.", __func__));
170 if (cp->private == NULL) {
171 G_STRIPE_DEBUG(0, "Disk %s removed from %s.",
172 cp->provider->name, sc->sc_name);
173 cp->private = (void *)(uintptr_t)-1;
176 if (sc->sc_provider != NULL) {
177 sc->sc_provider->flags |= G_PF_WITHER;
178 G_STRIPE_DEBUG(0, "Device %s deactivated.",
179 sc->sc_provider->name);
180 g_orphan_provider(sc->sc_provider, ENXIO);
181 sc->sc_provider = NULL;
184 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
186 sc->sc_disks[cp->index] = NULL;
189 g_destroy_consumer(cp);
190 /* If there are no valid disks anymore, remove device. */
191 if (LIST_EMPTY(&sc->sc_geom->consumer))
192 g_stripe_destroy(sc, 1);
196 g_stripe_orphan(struct g_consumer *cp)
198 struct g_stripe_softc *sc;
207 g_stripe_remove_disk(cp);
211 g_stripe_access(struct g_provider *pp, int dr, int dw, int de)
213 struct g_consumer *cp1, *cp2, *tmp;
214 struct g_stripe_softc *sc;
221 KASSERT(sc != NULL, ("NULL sc in %s.", __func__));
223 /* On first open, grab an extra "exclusive" bit */
224 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
226 /* ... and let go of it on last close */
227 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0)
230 LIST_FOREACH_SAFE(cp1, &gp->consumer, consumer, tmp) {
231 error = g_access(cp1, dr, dw, de);
234 if (cp1->acr == 0 && cp1->acw == 0 && cp1->ace == 0 &&
235 cp1->private != NULL) {
236 g_stripe_remove_disk(cp1); /* May destroy geom. */
242 LIST_FOREACH(cp2, &gp->consumer, consumer) {
245 g_access(cp2, -dr, -dw, -de);
251 g_stripe_copy(struct g_stripe_softc *sc, char *src, char *dst, off_t offset,
252 off_t length, int mode)
257 stripesize = sc->sc_stripesize;
258 len = (size_t)(stripesize - (offset & (stripesize - 1)));
260 bcopy(src, dst, len);
262 dst += len + stripesize * (sc->sc_ndisks - 1);
266 src += len + stripesize * (sc->sc_ndisks - 1);
270 ("Length < 0 (stripesize=%zu, offset=%jd, length=%jd).",
271 (size_t)stripesize, (intmax_t)offset, (intmax_t)length));
272 if (length > stripesize)
276 } while (length > 0);
280 g_stripe_done(struct bio *bp)
282 struct g_stripe_softc *sc;
285 pbp = bp->bio_parent;
286 sc = pbp->bio_to->geom->softc;
287 if (bp->bio_cmd == BIO_READ && bp->bio_caller1 != NULL) {
288 g_stripe_copy(sc, bp->bio_data, bp->bio_caller1, bp->bio_offset,
290 bp->bio_data = bp->bio_caller1;
291 bp->bio_caller1 = NULL;
293 mtx_lock(&sc->sc_lock);
294 if (pbp->bio_error == 0)
295 pbp->bio_error = bp->bio_error;
296 pbp->bio_completed += bp->bio_completed;
298 if (pbp->bio_children == pbp->bio_inbed) {
299 mtx_unlock(&sc->sc_lock);
300 if (pbp->bio_driver1 != NULL)
301 uma_zfree(g_stripe_zone, pbp->bio_driver1);
302 g_io_deliver(pbp, pbp->bio_error);
304 mtx_unlock(&sc->sc_lock);
309 g_stripe_start_fast(struct bio *bp, u_int no, off_t offset, off_t length)
311 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
312 u_int nparts = 0, stripesize;
313 struct g_stripe_softc *sc;
314 char *addr, *data = NULL;
318 sc = bp->bio_to->geom->softc;
321 stripesize = sc->sc_stripesize;
323 cbp = g_clone_bio(bp);
328 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
331 * Fill in the component buf structure.
333 cbp->bio_done = g_stripe_done;
334 cbp->bio_offset = offset;
335 cbp->bio_data = addr;
336 cbp->bio_caller1 = NULL;
337 cbp->bio_length = length;
338 cbp->bio_caller2 = sc->sc_disks[no];
340 /* offset -= offset % stripesize; */
341 offset -= offset & (stripesize - 1);
343 length = bp->bio_length - length;
344 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
345 if (no > sc->sc_ndisks - 1) {
347 offset += stripesize;
349 if (nparts >= sc->sc_ndisks) {
350 cbp = TAILQ_NEXT(cbp, bio_queue);
352 cbp = TAILQ_FIRST(&queue);
355 * Update bio structure.
358 * MIN() is in case when
359 * (bp->bio_length % sc->sc_stripesize) != 0.
361 cbp->bio_length += MIN(stripesize, length);
362 if (cbp->bio_caller1 == NULL) {
363 cbp->bio_caller1 = cbp->bio_data;
364 cbp->bio_data = NULL;
366 data = uma_zalloc(g_stripe_zone,
375 cbp = g_clone_bio(bp);
380 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
383 * Fill in the component buf structure.
385 cbp->bio_done = g_stripe_done;
386 cbp->bio_offset = offset;
387 cbp->bio_data = addr;
388 cbp->bio_caller1 = NULL;
390 * MIN() is in case when
391 * (bp->bio_length % sc->sc_stripesize) != 0.
393 cbp->bio_length = MIN(stripesize, length);
394 cbp->bio_caller2 = sc->sc_disks[no];
398 bp->bio_driver1 = data;
400 * Fire off all allocated requests!
402 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
403 struct g_consumer *cp;
405 TAILQ_REMOVE(&queue, cbp, bio_queue);
406 cp = cbp->bio_caller2;
407 cbp->bio_caller2 = NULL;
408 cbp->bio_to = cp->provider;
409 if (cbp->bio_caller1 != NULL) {
410 cbp->bio_data = data;
411 if (bp->bio_cmd == BIO_WRITE) {
412 g_stripe_copy(sc, cbp->bio_caller1, data,
413 cbp->bio_offset, cbp->bio_length, 0);
415 data += cbp->bio_length;
417 G_STRIPE_LOGREQ(cbp, "Sending request.");
418 g_io_request(cbp, cp);
423 uma_zfree(g_stripe_zone, data);
424 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
425 TAILQ_REMOVE(&queue, cbp, bio_queue);
426 if (cbp->bio_caller1 != NULL) {
427 cbp->bio_data = cbp->bio_caller1;
428 cbp->bio_caller1 = NULL;
437 g_stripe_start_economic(struct bio *bp, u_int no, off_t offset, off_t length)
439 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
440 struct g_stripe_softc *sc;
446 sc = bp->bio_to->geom->softc;
448 stripesize = sc->sc_stripesize;
450 cbp = g_clone_bio(bp);
455 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
457 * Fill in the component buf structure.
459 if (bp->bio_length == length)
460 cbp->bio_done = g_std_done; /* Optimized lockless case. */
462 cbp->bio_done = g_stripe_done;
463 cbp->bio_offset = offset;
464 cbp->bio_length = length;
465 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
466 bp->bio_ma_n = round_page(bp->bio_ma_offset +
467 bp->bio_length) / PAGE_SIZE;
471 cbp->bio_caller2 = sc->sc_disks[no];
473 /* offset -= offset % stripesize; */
474 offset -= offset & (stripesize - 1);
476 length = bp->bio_length - length;
477 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
478 if (no > sc->sc_ndisks - 1) {
480 offset += stripesize;
482 cbp = g_clone_bio(bp);
487 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
490 * Fill in the component buf structure.
492 cbp->bio_done = g_stripe_done;
493 cbp->bio_offset = offset;
495 * MIN() is in case when
496 * (bp->bio_length % sc->sc_stripesize) != 0.
498 cbp->bio_length = MIN(stripesize, length);
499 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
500 cbp->bio_ma_offset += (uintptr_t)addr;
501 cbp->bio_ma += cbp->bio_ma_offset / PAGE_SIZE;
502 cbp->bio_ma_offset %= PAGE_SIZE;
503 cbp->bio_ma_n = round_page(cbp->bio_ma_offset +
504 cbp->bio_length) / PAGE_SIZE;
506 cbp->bio_data = addr;
508 cbp->bio_caller2 = sc->sc_disks[no];
511 * Fire off all allocated requests!
513 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
514 struct g_consumer *cp;
516 TAILQ_REMOVE(&queue, cbp, bio_queue);
517 cp = cbp->bio_caller2;
518 cbp->bio_caller2 = NULL;
519 cbp->bio_to = cp->provider;
520 G_STRIPE_LOGREQ(cbp, "Sending request.");
521 g_io_request(cbp, cp);
525 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
526 TAILQ_REMOVE(&queue, cbp, bio_queue);
534 g_stripe_flush(struct g_stripe_softc *sc, struct bio *bp)
536 struct bio_queue_head queue;
537 struct g_consumer *cp;
542 for (no = 0; no < sc->sc_ndisks; no++) {
543 cbp = g_clone_bio(bp);
545 for (cbp = bioq_first(&queue); cbp != NULL;
546 cbp = bioq_first(&queue)) {
547 bioq_remove(&queue, cbp);
550 if (bp->bio_error == 0)
551 bp->bio_error = ENOMEM;
552 g_io_deliver(bp, bp->bio_error);
555 bioq_insert_tail(&queue, cbp);
556 cbp->bio_done = g_stripe_done;
557 cbp->bio_caller2 = sc->sc_disks[no];
558 cbp->bio_to = sc->sc_disks[no]->provider;
560 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) {
561 bioq_remove(&queue, cbp);
562 G_STRIPE_LOGREQ(cbp, "Sending request.");
563 cp = cbp->bio_caller2;
564 cbp->bio_caller2 = NULL;
565 g_io_request(cbp, cp);
570 g_stripe_start(struct bio *bp)
572 off_t offset, start, length, nstripe;
573 struct g_stripe_softc *sc;
574 u_int no, stripesize;
577 sc = bp->bio_to->geom->softc;
579 * If sc == NULL, provider's error should be set and g_stripe_start()
580 * should not be called at all.
583 ("Provider's error should be set (error=%d)(device=%s).",
584 bp->bio_to->error, bp->bio_to->name));
586 G_STRIPE_LOGREQ(bp, "Request received.");
588 switch (bp->bio_cmd) {
594 g_stripe_flush(sc, bp);
597 /* To which provider it should be delivered? */
599 g_io_deliver(bp, EOPNOTSUPP);
603 stripesize = sc->sc_stripesize;
606 * Calculations are quite messy, but fast I hope.
610 /* nstripe = bp->bio_offset / stripesize; */
611 nstripe = bp->bio_offset >> (off_t)sc->sc_stripebits;
613 no = nstripe % sc->sc_ndisks;
614 /* Start position in stripe. */
615 /* start = bp->bio_offset % stripesize; */
616 start = bp->bio_offset & (stripesize - 1);
617 /* Start position in disk. */
618 /* offset = (nstripe / sc->sc_ndisks) * stripesize + start; */
619 offset = ((nstripe / sc->sc_ndisks) << sc->sc_stripebits) + start;
620 /* Length of data to operate. */
621 length = MIN(bp->bio_length, stripesize - start);
624 * Do use "fast" mode when:
625 * 1. "Fast" mode is ON.
627 * 2. Request size is less than or equal to MAXPHYS,
628 * which should always be true.
630 * 3. Request size is bigger than stripesize * ndisks. If it isn't,
631 * there will be no need to send more than one I/O request to
632 * a provider, so there is nothing to optmize.
634 * 4. Request is not unmapped.
636 if (g_stripe_fast && bp->bio_length <= MAXPHYS &&
637 bp->bio_length >= stripesize * sc->sc_ndisks &&
638 (bp->bio_flags & BIO_UNMAPPED) == 0) {
643 error = g_stripe_start_fast(bp, no, offset, length);
645 g_stripe_fast_failed++;
648 * Do use "economic" when:
649 * 1. "Economic" mode is ON.
651 * 2. "Fast" mode failed. It can only fail if there is no memory.
653 if (!fast || error != 0)
654 error = g_stripe_start_economic(bp, no, offset, length);
656 if (bp->bio_error == 0)
657 bp->bio_error = error;
658 g_io_deliver(bp, bp->bio_error);
663 g_stripe_check_and_run(struct g_stripe_softc *sc)
665 struct g_provider *dp;
667 u_int no, sectorsize = 0;
670 if (g_stripe_nvalid(sc) != sc->sc_ndisks)
673 sc->sc_provider = g_new_providerf(sc->sc_geom, "stripe/%s",
675 sc->sc_provider->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
676 if (g_stripe_fast == 0)
677 sc->sc_provider->flags |= G_PF_ACCEPT_UNMAPPED;
679 * Find the smallest disk.
681 mediasize = sc->sc_disks[0]->provider->mediasize;
682 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
683 mediasize -= sc->sc_disks[0]->provider->sectorsize;
684 mediasize -= mediasize % sc->sc_stripesize;
685 sectorsize = sc->sc_disks[0]->provider->sectorsize;
686 for (no = 1; no < sc->sc_ndisks; no++) {
687 dp = sc->sc_disks[no]->provider;
689 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
690 ms -= dp->sectorsize;
691 ms -= ms % sc->sc_stripesize;
694 sectorsize = lcm(sectorsize, dp->sectorsize);
696 /* A provider underneath us doesn't support unmapped */
697 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) {
698 G_STRIPE_DEBUG(1, "Cancelling unmapped "
699 "because of %s.", dp->name);
700 sc->sc_provider->flags &= ~G_PF_ACCEPT_UNMAPPED;
703 sc->sc_provider->sectorsize = sectorsize;
704 sc->sc_provider->mediasize = mediasize * sc->sc_ndisks;
705 sc->sc_provider->stripesize = sc->sc_stripesize;
706 sc->sc_provider->stripeoffset = 0;
707 g_error_provider(sc->sc_provider, 0);
709 G_STRIPE_DEBUG(0, "Device %s activated.", sc->sc_provider->name);
713 g_stripe_read_metadata(struct g_consumer *cp, struct g_stripe_metadata *md)
715 struct g_provider *pp;
721 error = g_access(cp, 1, 0, 0);
726 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
729 g_access(cp, -1, 0, 0);
733 /* Decode metadata. */
734 stripe_metadata_decode(buf, md);
741 * Add disk to given device.
744 g_stripe_add_disk(struct g_stripe_softc *sc, struct g_provider *pp, u_int no)
746 struct g_consumer *cp, *fcp;
751 /* Metadata corrupted? */
752 if (no >= sc->sc_ndisks)
755 /* Check if disk is not already attached. */
756 if (sc->sc_disks[no] != NULL)
760 fcp = LIST_FIRST(&gp->consumer);
762 cp = g_new_consumer(gp);
763 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
766 error = g_attach(cp, pp);
768 g_destroy_consumer(cp);
772 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) {
773 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
776 g_destroy_consumer(cp);
780 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) {
781 struct g_stripe_metadata md;
783 /* Reread metadata. */
784 error = g_stripe_read_metadata(cp, &md);
788 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0 ||
789 strcmp(md.md_name, sc->sc_name) != 0 ||
790 md.md_id != sc->sc_id) {
791 G_STRIPE_DEBUG(0, "Metadata on %s changed.", pp->name);
796 sc->sc_disks[no] = cp;
797 G_STRIPE_DEBUG(0, "Disk %s attached to %s.", pp->name, sc->sc_name);
798 g_stripe_check_and_run(sc);
802 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0))
803 g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace);
805 g_destroy_consumer(cp);
809 static struct g_geom *
810 g_stripe_create(struct g_class *mp, const struct g_stripe_metadata *md,
813 struct g_stripe_softc *sc;
818 G_STRIPE_DEBUG(1, "Creating device %s (id=%u).", md->md_name,
821 /* Two disks is minimum. */
822 if (md->md_all < 2) {
823 G_STRIPE_DEBUG(0, "Too few disks defined for %s.", md->md_name);
827 /* Stripe size have to be grater than or equal to sector size. */
828 if (md->md_stripesize < sectorsize) {
829 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name);
833 /* Stripe size have to be power of 2. */
834 if (!powerof2(md->md_stripesize)) {
835 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name);
839 /* Check for duplicate unit */
840 LIST_FOREACH(gp, &mp->geom, geom) {
842 if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) {
843 G_STRIPE_DEBUG(0, "Device %s already configured.",
848 gp = g_new_geomf(mp, "%s", md->md_name);
849 sc = malloc(sizeof(*sc), M_STRIPE, M_WAITOK | M_ZERO);
850 gp->start = g_stripe_start;
851 gp->spoiled = g_stripe_orphan;
852 gp->orphan = g_stripe_orphan;
853 gp->access = g_stripe_access;
854 gp->dumpconf = g_stripe_dumpconf;
856 sc->sc_id = md->md_id;
857 sc->sc_stripesize = md->md_stripesize;
858 sc->sc_stripebits = bitcount32(sc->sc_stripesize - 1);
859 sc->sc_ndisks = md->md_all;
860 sc->sc_disks = malloc(sizeof(struct g_consumer *) * sc->sc_ndisks,
861 M_STRIPE, M_WAITOK | M_ZERO);
862 for (no = 0; no < sc->sc_ndisks; no++)
863 sc->sc_disks[no] = NULL;
865 mtx_init(&sc->sc_lock, "gstripe lock", NULL, MTX_DEF);
869 sc->sc_provider = NULL;
871 G_STRIPE_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id);
877 g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force)
879 struct g_provider *pp;
880 struct g_consumer *cp, *cp1;
888 pp = sc->sc_provider;
889 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
891 G_STRIPE_DEBUG(0, "Device %s is still open, so it "
892 "can't be definitely removed.", pp->name);
895 "Device %s is still open (r%dw%de%d).", pp->name,
896 pp->acr, pp->acw, pp->ace);
902 LIST_FOREACH_SAFE(cp, &gp->consumer, consumer, cp1) {
903 g_stripe_remove_disk(cp);
905 return (0); /* Recursion happened. */
907 if (!LIST_EMPTY(&gp->consumer))
908 return (EINPROGRESS);
911 KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)",
913 free(sc->sc_disks, M_STRIPE);
914 mtx_destroy(&sc->sc_lock);
916 G_STRIPE_DEBUG(0, "Device %s destroyed.", gp->name);
917 g_wither_geom(gp, ENXIO);
922 g_stripe_destroy_geom(struct gctl_req *req __unused,
923 struct g_class *mp __unused, struct g_geom *gp)
925 struct g_stripe_softc *sc;
928 return (g_stripe_destroy(sc, 0));
931 static struct g_geom *
932 g_stripe_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
934 struct g_stripe_metadata md;
935 struct g_stripe_softc *sc;
936 struct g_consumer *cp;
940 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
943 /* Skip providers that are already open for writing. */
947 G_STRIPE_DEBUG(3, "Tasting %s.", pp->name);
949 gp = g_new_geomf(mp, "stripe:taste");
950 gp->start = g_stripe_start;
951 gp->access = g_stripe_access;
952 gp->orphan = g_stripe_orphan;
953 cp = g_new_consumer(gp);
955 error = g_stripe_read_metadata(cp, &md);
957 g_destroy_consumer(cp);
963 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0)
965 if (md.md_version > G_STRIPE_VERSION) {
966 printf("geom_stripe.ko module is too old to handle %s.\n",
971 * Backward compatibility:
973 /* There was no md_provider field in earlier versions of metadata. */
974 if (md.md_version < 2)
975 bzero(md.md_provider, sizeof(md.md_provider));
976 /* There was no md_provsize field in earlier versions of metadata. */
977 if (md.md_version < 3)
978 md.md_provsize = pp->mediasize;
980 if (md.md_provider[0] != '\0' &&
981 !g_compare_names(md.md_provider, pp->name))
983 if (md.md_provsize != pp->mediasize)
987 * Let's check if device already exists.
990 LIST_FOREACH(gp, &mp->geom, geom) {
994 if (sc->sc_type != G_STRIPE_TYPE_AUTOMATIC)
996 if (strcmp(md.md_name, sc->sc_name) != 0)
998 if (md.md_id != sc->sc_id)
1003 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
1004 error = g_stripe_add_disk(sc, pp, md.md_no);
1007 "Cannot add disk %s to %s (error=%d).", pp->name,
1012 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_AUTOMATIC);
1014 G_STRIPE_DEBUG(0, "Cannot create device %s.",
1019 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
1020 error = g_stripe_add_disk(sc, pp, md.md_no);
1023 "Cannot add disk %s to %s (error=%d).", pp->name,
1025 g_stripe_destroy(sc, 1);
1034 g_stripe_ctl_create(struct gctl_req *req, struct g_class *mp)
1037 struct g_stripe_metadata md;
1038 struct g_provider *pp;
1039 struct g_stripe_softc *sc;
1042 intmax_t *stripesize;
1047 g_topology_assert();
1048 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1049 if (nargs == NULL) {
1050 gctl_error(req, "No '%s' argument.", "nargs");
1054 gctl_error(req, "Too few arguments.");
1058 strlcpy(md.md_magic, G_STRIPE_MAGIC, sizeof(md.md_magic));
1059 md.md_version = G_STRIPE_VERSION;
1060 name = gctl_get_asciiparam(req, "arg0");
1062 gctl_error(req, "No 'arg%u' argument.", 0);
1065 strlcpy(md.md_name, name, sizeof(md.md_name));
1066 md.md_id = arc4random();
1068 md.md_all = *nargs - 1;
1069 stripesize = gctl_get_paraml(req, "stripesize", sizeof(*stripesize));
1070 if (stripesize == NULL) {
1071 gctl_error(req, "No '%s' argument.", "stripesize");
1074 md.md_stripesize = *stripesize;
1075 bzero(md.md_provider, sizeof(md.md_provider));
1076 /* This field is not important here. */
1079 /* Check all providers are valid */
1080 for (no = 1; no < *nargs; no++) {
1081 snprintf(param, sizeof(param), "arg%u", no);
1082 name = gctl_get_asciiparam(req, param);
1084 gctl_error(req, "No 'arg%u' argument.", no);
1087 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1088 name += strlen("/dev/");
1089 pp = g_provider_by_name(name);
1091 G_STRIPE_DEBUG(1, "Disk %s is invalid.", name);
1092 gctl_error(req, "Disk %s is invalid.", name);
1097 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_MANUAL);
1099 gctl_error(req, "Can't configure %s.", md.md_name);
1104 sb = sbuf_new_auto();
1105 sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name);
1106 for (attached = 0, no = 1; no < *nargs; no++) {
1107 snprintf(param, sizeof(param), "arg%u", no);
1108 name = gctl_get_asciiparam(req, param);
1110 gctl_error(req, "No 'arg%u' argument.", no);
1113 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1114 name += strlen("/dev/");
1115 pp = g_provider_by_name(name);
1116 KASSERT(pp != NULL, ("Provider %s disappear?!", name));
1117 if (g_stripe_add_disk(sc, pp, no - 1) != 0) {
1118 G_STRIPE_DEBUG(1, "Disk %u (%s) not attached to %s.",
1119 no, pp->name, gp->name);
1120 sbuf_printf(sb, " %s", pp->name);
1126 if (md.md_all != attached) {
1127 g_stripe_destroy(gp->softc, 1);
1128 gctl_error(req, "%s", sbuf_data(sb));
1133 static struct g_stripe_softc *
1134 g_stripe_find_device(struct g_class *mp, const char *name)
1136 struct g_stripe_softc *sc;
1139 LIST_FOREACH(gp, &mp->geom, geom) {
1143 if (strcmp(sc->sc_name, name) == 0)
1150 g_stripe_ctl_destroy(struct gctl_req *req, struct g_class *mp)
1152 struct g_stripe_softc *sc;
1153 int *force, *nargs, error;
1158 g_topology_assert();
1160 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1161 if (nargs == NULL) {
1162 gctl_error(req, "No '%s' argument.", "nargs");
1166 gctl_error(req, "Missing device(s).");
1169 force = gctl_get_paraml(req, "force", sizeof(*force));
1170 if (force == NULL) {
1171 gctl_error(req, "No '%s' argument.", "force");
1175 for (i = 0; i < (u_int)*nargs; i++) {
1176 snprintf(param, sizeof(param), "arg%u", i);
1177 name = gctl_get_asciiparam(req, param);
1179 gctl_error(req, "No 'arg%u' argument.", i);
1182 sc = g_stripe_find_device(mp, name);
1184 gctl_error(req, "No such device: %s.", name);
1187 error = g_stripe_destroy(sc, *force);
1189 gctl_error(req, "Cannot destroy device %s (error=%d).",
1190 sc->sc_name, error);
1197 g_stripe_config(struct gctl_req *req, struct g_class *mp, const char *verb)
1201 g_topology_assert();
1203 version = gctl_get_paraml(req, "version", sizeof(*version));
1204 if (version == NULL) {
1205 gctl_error(req, "No '%s' argument.", "version");
1208 if (*version != G_STRIPE_VERSION) {
1209 gctl_error(req, "Userland and kernel parts are out of sync.");
1213 if (strcmp(verb, "create") == 0) {
1214 g_stripe_ctl_create(req, mp);
1216 } else if (strcmp(verb, "destroy") == 0 ||
1217 strcmp(verb, "stop") == 0) {
1218 g_stripe_ctl_destroy(req, mp);
1222 gctl_error(req, "Unknown verb.");
1226 g_stripe_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1227 struct g_consumer *cp, struct g_provider *pp)
1229 struct g_stripe_softc *sc;
1236 } else if (cp != NULL) {
1237 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent,
1240 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
1241 sbuf_printf(sb, "%s<Stripesize>%u</Stripesize>\n", indent,
1242 (u_int)sc->sc_stripesize);
1243 sbuf_printf(sb, "%s<Type>", indent);
1244 switch (sc->sc_type) {
1245 case G_STRIPE_TYPE_AUTOMATIC:
1246 sbuf_printf(sb, "AUTOMATIC");
1248 case G_STRIPE_TYPE_MANUAL:
1249 sbuf_printf(sb, "MANUAL");
1252 sbuf_printf(sb, "UNKNOWN");
1255 sbuf_printf(sb, "</Type>\n");
1256 sbuf_printf(sb, "%s<Status>Total=%u, Online=%u</Status>\n",
1257 indent, sc->sc_ndisks, g_stripe_nvalid(sc));
1258 sbuf_printf(sb, "%s<State>", indent);
1259 if (sc->sc_provider != NULL && sc->sc_provider->error == 0)
1260 sbuf_printf(sb, "UP");
1262 sbuf_printf(sb, "DOWN");
1263 sbuf_printf(sb, "</State>\n");
1267 DECLARE_GEOM_CLASS(g_stripe_class, g_stripe);