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>
37 #include <sys/sysctl.h>
38 #include <sys/malloc.h>
39 #include <geom/geom.h>
40 #include <geom/concat/g_concat.h>
43 static MALLOC_DEFINE(M_CONCAT, "concat_data", "GEOM_CONCAT Data");
45 SYSCTL_DECL(_kern_geom);
46 SYSCTL_NODE(_kern_geom, OID_AUTO, concat, CTLFLAG_RW, 0, "GEOM_CONCAT stuff");
47 static u_int g_concat_debug = 0;
48 TUNABLE_INT("kern.geom.concat.debug", &g_concat_debug);
49 SYSCTL_UINT(_kern_geom_concat, OID_AUTO, debug, CTLFLAG_RW, &g_concat_debug, 0,
52 static int g_concat_destroy(struct g_concat_softc *sc, boolean_t force);
53 static int g_concat_destroy_geom(struct gctl_req *req, struct g_class *mp,
56 static g_taste_t g_concat_taste;
57 static g_ctl_req_t g_concat_config;
58 static g_dumpconf_t g_concat_dumpconf;
60 struct g_class g_concat_class = {
61 .name = G_CONCAT_CLASS_NAME,
63 .ctlreq = g_concat_config,
64 .taste = g_concat_taste,
65 .destroy_geom = g_concat_destroy_geom
70 * Greatest Common Divisor.
86 * Least Common Multiple.
92 return ((a * b) / gcd(a, b));
96 * Return the number of valid disks.
99 g_concat_nvalid(struct g_concat_softc *sc)
104 for (i = 0; i < sc->sc_ndisks; i++) {
105 if (sc->sc_disks[i].d_consumer != NULL)
113 g_concat_remove_disk(struct g_concat_disk *disk)
115 struct g_consumer *cp;
116 struct g_concat_softc *sc;
118 KASSERT(disk->d_consumer != NULL, ("Non-valid disk in %s.", __func__));
120 cp = disk->d_consumer;
122 G_CONCAT_DEBUG(0, "Disk %s removed from %s.", cp->provider->name,
125 disk->d_consumer = NULL;
126 if (sc->sc_provider != NULL) {
127 sc->sc_provider->flags |= G_PF_WITHER;
128 g_orphan_provider(sc->sc_provider, ENXIO);
129 sc->sc_provider = NULL;
130 G_CONCAT_DEBUG(0, "Device %s removed.", sc->sc_name);
133 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
134 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
136 g_destroy_consumer(cp);
140 g_concat_orphan(struct g_consumer *cp)
142 struct g_concat_softc *sc;
143 struct g_concat_disk *disk;
153 if (disk == NULL) /* Possible? */
155 g_concat_remove_disk(disk);
157 /* If there are no valid disks anymore, remove device. */
158 if (g_concat_nvalid(sc) == 0)
159 g_concat_destroy(sc, 1);
163 g_concat_access(struct g_provider *pp, int dr, int dw, int de)
165 struct g_consumer *cp1, *cp2;
166 struct g_concat_softc *sc;
175 * It looks like geom is being withered.
176 * In that case we allow only negative requests.
178 KASSERT(dr <= 0 && dw <= 0 && de <= 0,
179 ("Positive access request (device=%s).", pp->name));
180 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 &&
181 (pp->ace + de) == 0) {
182 G_CONCAT_DEBUG(0, "Device %s definitely destroyed.",
188 /* On first open, grab an extra "exclusive" bit */
189 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
191 /* ... and let go of it on last close */
192 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0)
196 LIST_FOREACH(cp1, &gp->consumer, consumer) {
197 error = g_access(cp1, dr, dw, de);
201 * If we fail here, backout all previous changes.
203 LIST_FOREACH(cp2, &gp->consumer, consumer) {
206 g_access(cp2, -dr, -dw, -de);
215 g_concat_flush(struct g_concat_softc *sc, struct bio *bp)
217 struct bio_queue_head queue;
218 struct g_consumer *cp;
223 for (no = 0; no < sc->sc_ndisks; no++) {
224 cbp = g_clone_bio(bp);
226 for (cbp = bioq_first(&queue); cbp != NULL;
227 cbp = bioq_first(&queue)) {
228 bioq_remove(&queue, cbp);
231 if (bp->bio_error == 0)
232 bp->bio_error = ENOMEM;
233 g_io_deliver(bp, bp->bio_error);
236 bioq_insert_tail(&queue, cbp);
237 cbp->bio_done = g_std_done;
238 cbp->bio_caller1 = sc->sc_disks[no].d_consumer;
239 cbp->bio_to = sc->sc_disks[no].d_consumer->provider;
241 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) {
242 bioq_remove(&queue, cbp);
243 G_CONCAT_LOGREQ(cbp, "Sending request.");
244 cp = cbp->bio_caller1;
245 cbp->bio_caller1 = NULL;
246 g_io_request(cbp, cp);
251 g_concat_start(struct bio *bp)
253 struct bio_queue_head queue;
254 struct g_concat_softc *sc;
255 struct g_concat_disk *disk;
256 struct g_provider *pp;
257 off_t offset, end, length, off, len;
263 sc = pp->geom->softc;
265 * If sc == NULL, provider's error should be set and g_concat_start()
266 * should not be called at all.
269 ("Provider's error should be set (error=%d)(device=%s).",
270 bp->bio_to->error, bp->bio_to->name));
272 G_CONCAT_LOGREQ(bp, "Request received.");
274 switch (bp->bio_cmd) {
280 g_concat_flush(sc, bp);
283 /* To which provider it should be delivered? */
285 g_io_deliver(bp, EOPNOTSUPP);
289 offset = bp->bio_offset;
290 length = bp->bio_length;
292 end = offset + length;
295 for (no = 0; no < sc->sc_ndisks; no++) {
296 disk = &sc->sc_disks[no];
297 if (disk->d_end <= offset)
299 if (disk->d_start >= end)
302 off = offset - disk->d_start;
303 len = MIN(length, disk->d_end - offset);
307 cbp = g_clone_bio(bp);
309 for (cbp = bioq_first(&queue); cbp != NULL;
310 cbp = bioq_first(&queue)) {
311 bioq_remove(&queue, cbp);
314 if (bp->bio_error == 0)
315 bp->bio_error = ENOMEM;
316 g_io_deliver(bp, bp->bio_error);
319 bioq_insert_tail(&queue, cbp);
321 * Fill in the component buf structure.
323 cbp->bio_done = g_std_done;
324 cbp->bio_offset = off;
325 cbp->bio_data = addr;
327 cbp->bio_length = len;
328 cbp->bio_to = disk->d_consumer->provider;
329 cbp->bio_caller1 = disk;
335 ("Length is still greater than 0 (class=%s, name=%s).",
336 bp->bio_to->geom->class->name, bp->bio_to->geom->name));
337 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) {
338 bioq_remove(&queue, cbp);
339 G_CONCAT_LOGREQ(cbp, "Sending request.");
340 disk = cbp->bio_caller1;
341 cbp->bio_caller1 = NULL;
342 g_io_request(cbp, disk->d_consumer);
347 g_concat_check_and_run(struct g_concat_softc *sc)
349 struct g_concat_disk *disk;
350 struct g_provider *pp;
351 u_int no, sectorsize = 0;
354 if (g_concat_nvalid(sc) != sc->sc_ndisks)
357 pp = g_new_providerf(sc->sc_geom, "concat/%s", sc->sc_name);
359 for (no = 0; no < sc->sc_ndisks; no++) {
360 disk = &sc->sc_disks[no];
361 disk->d_start = start;
362 disk->d_end = disk->d_start +
363 disk->d_consumer->provider->mediasize;
364 if (sc->sc_type == G_CONCAT_TYPE_AUTOMATIC)
365 disk->d_end -= disk->d_consumer->provider->sectorsize;
368 sectorsize = disk->d_consumer->provider->sectorsize;
370 sectorsize = lcm(sectorsize,
371 disk->d_consumer->provider->sectorsize);
374 pp->sectorsize = sectorsize;
375 /* We have sc->sc_disks[sc->sc_ndisks - 1].d_end in 'start'. */
376 pp->mediasize = start;
377 pp->stripesize = sc->sc_disks[0].d_consumer->provider->stripesize;
378 pp->stripeoffset = sc->sc_disks[0].d_consumer->provider->stripeoffset;
379 sc->sc_provider = pp;
380 g_error_provider(pp, 0);
382 G_CONCAT_DEBUG(0, "Device %s activated.", sc->sc_name);
386 g_concat_read_metadata(struct g_consumer *cp, struct g_concat_metadata *md)
388 struct g_provider *pp;
394 error = g_access(cp, 1, 0, 0);
399 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
402 g_access(cp, -1, 0, 0);
406 /* Decode metadata. */
407 concat_metadata_decode(buf, md);
414 * Add disk to given device.
417 g_concat_add_disk(struct g_concat_softc *sc, struct g_provider *pp, u_int no)
419 struct g_concat_disk *disk;
420 struct g_consumer *cp, *fcp;
424 /* Metadata corrupted? */
425 if (no >= sc->sc_ndisks)
428 disk = &sc->sc_disks[no];
429 /* Check if disk is not already attached. */
430 if (disk->d_consumer != NULL)
434 fcp = LIST_FIRST(&gp->consumer);
436 cp = g_new_consumer(gp);
437 error = g_attach(cp, pp);
439 g_destroy_consumer(cp);
443 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) {
444 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
447 g_destroy_consumer(cp);
451 if (sc->sc_type == G_CONCAT_TYPE_AUTOMATIC) {
452 struct g_concat_metadata md;
454 /* Re-read metadata. */
455 error = g_concat_read_metadata(cp, &md);
459 if (strcmp(md.md_magic, G_CONCAT_MAGIC) != 0 ||
460 strcmp(md.md_name, sc->sc_name) != 0 ||
461 md.md_id != sc->sc_id) {
462 G_CONCAT_DEBUG(0, "Metadata on %s changed.", pp->name);
468 disk->d_consumer = cp;
470 disk->d_start = 0; /* not yet */
471 disk->d_end = 0; /* not yet */
473 G_CONCAT_DEBUG(0, "Disk %s attached to %s.", pp->name, sc->sc_name);
475 g_concat_check_and_run(sc);
479 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0))
480 g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace);
482 g_destroy_consumer(cp);
486 static struct g_geom *
487 g_concat_create(struct g_class *mp, const struct g_concat_metadata *md,
490 struct g_concat_softc *sc;
494 G_CONCAT_DEBUG(1, "Creating device %s (id=%u).", md->md_name,
497 /* One disks is minimum. */
501 /* Check for duplicate unit */
502 LIST_FOREACH(gp, &mp->geom, geom) {
504 if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) {
505 G_CONCAT_DEBUG(0, "Device %s already configured.",
510 gp = g_new_geomf(mp, "%s", md->md_name);
511 gp->softc = NULL; /* for a moment */
513 sc = malloc(sizeof(*sc), M_CONCAT, M_WAITOK | M_ZERO);
514 gp->start = g_concat_start;
515 gp->spoiled = g_concat_orphan;
516 gp->orphan = g_concat_orphan;
517 gp->access = g_concat_access;
518 gp->dumpconf = g_concat_dumpconf;
520 sc->sc_id = md->md_id;
521 sc->sc_ndisks = md->md_all;
522 sc->sc_disks = malloc(sizeof(struct g_concat_disk) * sc->sc_ndisks,
523 M_CONCAT, M_WAITOK | M_ZERO);
524 for (no = 0; no < sc->sc_ndisks; no++)
525 sc->sc_disks[no].d_consumer = NULL;
530 sc->sc_provider = NULL;
532 G_CONCAT_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id);
538 g_concat_destroy(struct g_concat_softc *sc, boolean_t force)
540 struct g_provider *pp;
549 pp = sc->sc_provider;
550 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
552 G_CONCAT_DEBUG(0, "Device %s is still open, so it "
553 "can't be definitely removed.", pp->name);
556 "Device %s is still open (r%dw%de%d).", pp->name,
557 pp->acr, pp->acw, pp->ace);
562 for (no = 0; no < sc->sc_ndisks; no++) {
563 if (sc->sc_disks[no].d_consumer != NULL)
564 g_concat_remove_disk(&sc->sc_disks[no]);
569 KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)",
571 free(sc->sc_disks, M_CONCAT);
574 pp = LIST_FIRST(&gp->provider);
575 if (pp == NULL || (pp->acr == 0 && pp->acw == 0 && pp->ace == 0))
576 G_CONCAT_DEBUG(0, "Device %s destroyed.", gp->name);
578 g_wither_geom(gp, ENXIO);
584 g_concat_destroy_geom(struct gctl_req *req __unused,
585 struct g_class *mp __unused, struct g_geom *gp)
587 struct g_concat_softc *sc;
590 return (g_concat_destroy(sc, 0));
593 static struct g_geom *
594 g_concat_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
596 struct g_concat_metadata md;
597 struct g_concat_softc *sc;
598 struct g_consumer *cp;
602 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
605 /* Skip providers that are already open for writing. */
609 G_CONCAT_DEBUG(3, "Tasting %s.", pp->name);
611 gp = g_new_geomf(mp, "concat:taste");
612 gp->start = g_concat_start;
613 gp->access = g_concat_access;
614 gp->orphan = g_concat_orphan;
615 cp = g_new_consumer(gp);
617 error = g_concat_read_metadata(cp, &md);
619 g_destroy_consumer(cp);
625 if (strcmp(md.md_magic, G_CONCAT_MAGIC) != 0)
627 if (md.md_version > G_CONCAT_VERSION) {
628 printf("geom_concat.ko module is too old to handle %s.\n",
633 * Backward compatibility:
635 /* There was no md_provider field in earlier versions of metadata. */
636 if (md.md_version < 3)
637 bzero(md.md_provider, sizeof(md.md_provider));
638 /* There was no md_provsize field in earlier versions of metadata. */
639 if (md.md_version < 4)
640 md.md_provsize = pp->mediasize;
642 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0)
644 if (md.md_provsize != pp->mediasize)
648 * Let's check if device already exists.
651 LIST_FOREACH(gp, &mp->geom, geom) {
655 if (sc->sc_type != G_CONCAT_TYPE_AUTOMATIC)
657 if (strcmp(md.md_name, sc->sc_name) != 0)
659 if (md.md_id != sc->sc_id)
664 G_CONCAT_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
665 error = g_concat_add_disk(sc, pp, md.md_no);
668 "Cannot add disk %s to %s (error=%d).", pp->name,
673 gp = g_concat_create(mp, &md, G_CONCAT_TYPE_AUTOMATIC);
675 G_CONCAT_DEBUG(0, "Cannot create device %s.",
680 G_CONCAT_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
681 error = g_concat_add_disk(sc, pp, md.md_no);
684 "Cannot add disk %s to %s (error=%d).", pp->name,
686 g_concat_destroy(sc, 1);
695 g_concat_ctl_create(struct gctl_req *req, struct g_class *mp)
698 struct g_concat_metadata md;
699 struct g_provider *pp;
700 struct g_concat_softc *sc;
708 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
710 gctl_error(req, "No '%s' argument.", "nargs");
714 gctl_error(req, "Too few arguments.");
718 strlcpy(md.md_magic, G_CONCAT_MAGIC, sizeof(md.md_magic));
719 md.md_version = G_CONCAT_VERSION;
720 name = gctl_get_asciiparam(req, "arg0");
722 gctl_error(req, "No 'arg%u' argument.", 0);
725 strlcpy(md.md_name, name, sizeof(md.md_name));
726 md.md_id = arc4random();
728 md.md_all = *nargs - 1;
729 bzero(md.md_provider, sizeof(md.md_provider));
730 /* This field is not important here. */
733 /* Check all providers are valid */
734 for (no = 1; no < *nargs; no++) {
735 snprintf(param, sizeof(param), "arg%u", no);
736 name = gctl_get_asciiparam(req, param);
738 gctl_error(req, "No 'arg%u' argument.", no);
741 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
742 name += strlen("/dev/");
743 pp = g_provider_by_name(name);
745 G_CONCAT_DEBUG(1, "Disk %s is invalid.", name);
746 gctl_error(req, "Disk %s is invalid.", name);
751 gp = g_concat_create(mp, &md, G_CONCAT_TYPE_MANUAL);
753 gctl_error(req, "Can't configure %s.", md.md_name);
758 sb = sbuf_new_auto();
759 sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name);
760 for (attached = 0, no = 1; no < *nargs; no++) {
761 snprintf(param, sizeof(param), "arg%u", no);
762 name = gctl_get_asciiparam(req, param);
764 gctl_error(req, "No 'arg%d' argument.", no);
767 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
768 name += strlen("/dev/");
769 pp = g_provider_by_name(name);
770 KASSERT(pp != NULL, ("Provider %s disappear?!", name));
771 if (g_concat_add_disk(sc, pp, no - 1) != 0) {
772 G_CONCAT_DEBUG(1, "Disk %u (%s) not attached to %s.",
773 no, pp->name, gp->name);
774 sbuf_printf(sb, " %s", pp->name);
780 if (md.md_all != attached) {
781 g_concat_destroy(gp->softc, 1);
782 gctl_error(req, "%s", sbuf_data(sb));
787 static struct g_concat_softc *
788 g_concat_find_device(struct g_class *mp, const char *name)
790 struct g_concat_softc *sc;
793 LIST_FOREACH(gp, &mp->geom, geom) {
797 if (strcmp(sc->sc_name, name) == 0)
804 g_concat_ctl_destroy(struct gctl_req *req, struct g_class *mp)
806 struct g_concat_softc *sc;
807 int *force, *nargs, error;
814 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
816 gctl_error(req, "No '%s' argument.", "nargs");
820 gctl_error(req, "Missing device(s).");
823 force = gctl_get_paraml(req, "force", sizeof(*force));
825 gctl_error(req, "No '%s' argument.", "force");
829 for (i = 0; i < (u_int)*nargs; i++) {
830 snprintf(param, sizeof(param), "arg%u", i);
831 name = gctl_get_asciiparam(req, param);
833 gctl_error(req, "No 'arg%u' argument.", i);
836 sc = g_concat_find_device(mp, name);
838 gctl_error(req, "No such device: %s.", name);
841 error = g_concat_destroy(sc, *force);
843 gctl_error(req, "Cannot destroy device %s (error=%d).",
851 g_concat_config(struct gctl_req *req, struct g_class *mp, const char *verb)
857 version = gctl_get_paraml(req, "version", sizeof(*version));
858 if (version == NULL) {
859 gctl_error(req, "No '%s' argument.", "version");
862 if (*version != G_CONCAT_VERSION) {
863 gctl_error(req, "Userland and kernel parts are out of sync.");
867 if (strcmp(verb, "create") == 0) {
868 g_concat_ctl_create(req, mp);
870 } else if (strcmp(verb, "destroy") == 0 ||
871 strcmp(verb, "stop") == 0) {
872 g_concat_ctl_destroy(req, mp);
875 gctl_error(req, "Unknown verb.");
879 g_concat_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
880 struct g_consumer *cp, struct g_provider *pp)
882 struct g_concat_softc *sc;
890 } else if (cp != NULL) {
891 struct g_concat_disk *disk;
896 sbuf_printf(sb, "%s<End>%jd</End>\n", indent,
897 (intmax_t)disk->d_end);
898 sbuf_printf(sb, "%s<Start>%jd</Start>\n", indent,
899 (intmax_t)disk->d_start);
901 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
902 sbuf_printf(sb, "%s<Type>", indent);
903 switch (sc->sc_type) {
904 case G_CONCAT_TYPE_AUTOMATIC:
905 sbuf_printf(sb, "AUTOMATIC");
907 case G_CONCAT_TYPE_MANUAL:
908 sbuf_printf(sb, "MANUAL");
911 sbuf_printf(sb, "UNKNOWN");
914 sbuf_printf(sb, "</Type>\n");
915 sbuf_printf(sb, "%s<Status>Total=%u, Online=%u</Status>\n",
916 indent, sc->sc_ndisks, g_concat_nvalid(sc));
917 sbuf_printf(sb, "%s<State>", indent);
918 if (sc->sc_provider != NULL && sc->sc_provider->error == 0)
919 sbuf_printf(sb, "UP");
921 sbuf_printf(sb, "DOWN");
922 sbuf_printf(sb, "</State>\n");
926 DECLARE_GEOM_CLASS(g_concat_class, g_concat);