2 * Copyright (c) 2004, 2005 Lukas Ertl
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 AUTHOR 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 AUTHOR 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>
32 #include <sys/errno.h>
34 #include <sys/kernel.h>
35 #include <sys/kthread.h>
36 #include <sys/libkern.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/mutex.h>
42 #include <sys/systm.h>
45 #include <geom/geom.h>
46 #include <geom/vinum/geom_vinum_var.h>
47 #include <geom/vinum/geom_vinum.h>
48 #include <geom/vinum/geom_vinum_share.h>
50 static void gv_drive_dead(void *, int);
51 static void gv_drive_worker(void *);
54 gv_config_new_drive(struct gv_drive *d)
57 struct gv_freelist *fl;
59 KASSERT(d != NULL, ("config_new_drive: NULL d"));
61 vhdr = g_malloc(sizeof(*vhdr), M_WAITOK | M_ZERO);
62 vhdr->magic = GV_MAGIC;
63 vhdr->config_length = GV_CFG_LEN;
65 bcopy(hostname, vhdr->label.sysname, GV_HOSTNAME_LEN);
66 strncpy(vhdr->label.name, d->name, GV_MAXDRIVENAME);
67 microtime(&vhdr->label.date_of_birth);
71 LIST_INIT(&d->subdisks);
72 LIST_INIT(&d->freelist);
74 fl = g_malloc(sizeof(struct gv_freelist), M_WAITOK | M_ZERO);
75 fl->offset = GV_DATA_START;
77 LIST_INSERT_HEAD(&d->freelist, fl, freelist);
78 d->freelist_entries = 1;
80 d->bqueue = g_malloc(sizeof(struct bio_queue_head), M_WAITOK | M_ZERO);
82 mtx_init(&d->bqueue_mtx, "gv_drive", NULL, MTX_DEF);
83 kthread_create(gv_drive_worker, d, NULL, 0, 0, "gv_d %s", d->name);
84 d->flags |= GV_DRIVE_THREAD_ACTIVE;
88 gv_save_config_all(struct gv_softc *sc)
94 LIST_FOREACH(d, &sc->drives, drive) {
97 gv_save_config(NULL, d, sc);
101 /* Save the vinum configuration back to disk. */
103 gv_save_config(struct g_consumer *cp, struct gv_drive *d, struct gv_softc *sc)
106 struct g_consumer *cp2;
107 struct gv_hdr *vhdr, *hdr;
113 KASSERT(d != NULL, ("gv_save_config: null d"));
114 KASSERT(sc != NULL, ("gv_save_config: null sc"));
117 * We can't save the config on a drive that isn't up, but drives that
118 * were just created aren't officially up yet, so we check a special
121 if ((d->state != GV_DRIVE_UP) && !(d->flags && GV_DRIVE_NEWBORN))
126 KASSERT(gp != NULL, ("gv_save_config: null gp"));
127 cp2 = LIST_FIRST(&gp->consumer);
128 KASSERT(cp2 != NULL, ("gv_save_config: null cp2"));
132 vhdr = g_malloc(GV_HDR_LEN, M_WAITOK | M_ZERO);
133 vhdr->magic = GV_MAGIC;
134 vhdr->config_length = GV_CFG_LEN;
138 printf("GEOM_VINUM: drive %s has NULL hdr\n", d->name);
142 microtime(&hdr->label.last_update);
143 bcopy(&hdr->label, &vhdr->label, sizeof(struct gv_label));
145 sb = sbuf_new(NULL, NULL, GV_CFG_LEN, SBUF_FIXEDLEN);
146 gv_format_config(sc, sb, 1, NULL);
149 error = g_access(cp2, 0, 1, 0);
151 printf("GEOM_VINUM: g_access failed on drive %s, errno %d\n",
160 error = g_write_data(cp2, GV_HDR_OFFSET, vhdr, GV_HDR_LEN);
162 printf("GEOM_VINUM: writing vhdr failed on drive %s, "
163 "errno %d", d->name, error);
167 error = g_write_data(cp2, GV_CFG_OFFSET, sbuf_data(sb),
170 printf("GEOM_VINUM: writing first config copy failed "
171 "on drive %s, errno %d", d->name, error);
175 error = g_write_data(cp2, GV_CFG_OFFSET + GV_CFG_LEN,
176 sbuf_data(sb), GV_CFG_LEN);
178 printf("GEOM_VINUM: writing second config copy failed "
179 "on drive %s, errno %d", d->name, error);
183 g_access(cp2, 0, -1, 0);
191 /* This resembles g_slice_access(). */
193 gv_drive_access(struct g_provider *pp, int dr, int dw, int de)
196 struct g_consumer *cp;
197 struct g_provider *pp2;
199 struct gv_sd *s, *s2;
203 cp = LIST_FIRST(&gp->consumer);
212 KASSERT(s != NULL, ("gv_drive_access: NULL s"));
214 LIST_FOREACH(s2, &d->subdisks, from_drive) {
217 if (s->drive_offset + s->size <= s2->drive_offset)
219 if (s2->drive_offset + s2->size <= s->drive_offset)
224 KASSERT(s2 != NULL, ("gv_drive_access: NULL s2"));
225 if ((pp->acw + dw) > 0 && pp2->ace > 0)
227 if ((pp->ace + de) > 0 && pp2->acw > 0)
231 error = g_access(cp, dr, dw, de);
236 gv_drive_done(struct bio *bp)
240 /* Put the BIO on the worker queue again. */
241 d = bp->bio_from->geom->softc;
242 bp->bio_cflags |= GV_BIO_DONE;
243 mtx_lock(&d->bqueue_mtx);
244 bioq_insert_tail(d->bqueue, bp);
246 mtx_unlock(&d->bqueue_mtx);
251 gv_drive_start(struct bio *bp)
256 switch (bp->bio_cmd) {
263 g_io_deliver(bp, EOPNOTSUPP);
267 s = bp->bio_to->private;
268 if ((s->state == GV_SD_DOWN) || (s->state == GV_SD_STALE)) {
269 g_io_deliver(bp, ENXIO);
273 d = bp->bio_to->geom->softc;
276 * Put the BIO on the worker queue, where the worker thread will pick
279 mtx_lock(&d->bqueue_mtx);
280 bioq_disksort(d->bqueue, bp);
282 mtx_unlock(&d->bqueue_mtx);
287 gv_drive_worker(void *arg)
289 struct bio *bp, *cbp;
291 struct g_provider *pp;
298 mtx_lock(&d->bqueue_mtx);
300 /* We were signaled to exit. */
301 if (d->flags & GV_DRIVE_THREAD_DIE)
304 /* Take the first BIO from out queue. */
305 bp = bioq_takefirst(d->bqueue);
307 msleep(d, &d->bqueue_mtx, PRIBIO, "-", hz/10);
310 mtx_unlock(&d->bqueue_mtx);
315 /* Completed request. */
316 if (bp->bio_cflags & GV_BIO_DONE) {
317 error = bp->bio_error;
319 /* Deliver the original request. */
322 /* The request had an error, we need to clean up. */
325 gv_set_drive_state(d, GV_DRIVE_DOWN,
326 GV_SETSTATE_FORCE | GV_SETSTATE_CONFIG);
328 g_post_event(gv_drive_dead, d, M_WAITOK, d,
332 /* New request, needs to be sent downwards. */
336 if ((s->state == GV_SD_DOWN) ||
337 (s->state == GV_SD_STALE)) {
338 g_io_deliver(bp, ENXIO);
339 mtx_lock(&d->bqueue_mtx);
342 if (bp->bio_offset > s->size) {
343 g_io_deliver(bp, EINVAL);
344 mtx_lock(&d->bqueue_mtx);
348 cbp = g_clone_bio(bp);
350 g_io_deliver(bp, ENOMEM);
351 mtx_lock(&d->bqueue_mtx);
354 if (cbp->bio_offset + cbp->bio_length > s->size)
355 cbp->bio_length = s->size -
357 cbp->bio_done = gv_drive_done;
358 cbp->bio_offset += s->drive_offset;
359 g_io_request(cbp, LIST_FIRST(&gp->consumer));
362 mtx_lock(&d->bqueue_mtx);
365 while ((bp = bioq_takefirst(d->bqueue)) != NULL) {
366 mtx_unlock(&d->bqueue_mtx);
367 if (bp->bio_cflags & GV_BIO_DONE)
370 g_io_deliver(bp, ENXIO);
371 mtx_lock(&d->bqueue_mtx);
373 mtx_unlock(&d->bqueue_mtx);
374 d->flags |= GV_DRIVE_THREAD_DEAD;
381 gv_drive_orphan(struct g_consumer *cp)
388 g_trace(G_T_TOPOLOGY, "gv_drive_orphan(%s)", gp->name);
391 gv_set_drive_state(d, GV_DRIVE_DOWN,
392 GV_SETSTATE_FORCE | GV_SETSTATE_CONFIG);
393 g_post_event(gv_drive_dead, d, M_WAITOK, d, NULL);
395 g_wither_geom(gp, ENXIO);
398 static struct g_geom *
399 gv_drive_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
401 struct g_geom *gp, *gp2;
402 struct g_consumer *cp;
406 struct gv_freelist *fl;
409 char *buf, errstr[ERRBUFSIZ];
414 g_trace(G_T_TOPOLOGY, "gv_drive_taste(%s, %s)", mp->name, pp->name);
417 /* Find the VINUM class and its associated geom. */
418 gp2 = find_vinum_geom();
423 gp = g_new_geomf(mp, "%s.vinumdrive", pp->name);
424 gp->start = gv_drive_start;
425 gp->orphan = gv_drive_orphan;
426 gp->access = gv_drive_access;
427 gp->start = gv_drive_start;
429 cp = g_new_consumer(gp);
431 error = g_access(cp, 1, 0, 0);
434 g_destroy_consumer(cp);
441 /* Now check if the provided slice is a valid vinum drive. */
443 vhdr = g_read_data(cp, GV_HDR_OFFSET, pp->sectorsize, NULL);
446 if (vhdr->magic != GV_MAGIC) {
451 /* A valid vinum drive, let's parse the on-disk information. */
452 buf = g_read_data(cp, GV_CFG_OFFSET, GV_CFG_LEN, NULL);
458 gv_parse_config(sc, buf, 1);
462 * Let's see if this drive is already known in the
465 d = gv_find_drive(sc, vhdr->label.name);
467 /* We already know about this drive. */
469 /* Check if this drive already has a geom. */
470 if (d->geom != NULL) {
475 bcopy(vhdr, d->hdr, sizeof(*vhdr));
478 /* This is a new drive. */
480 d = g_malloc(sizeof(*d), M_WAITOK | M_ZERO);
482 /* Initialize all needed variables. */
483 d->size = pp->mediasize - GV_DATA_START;
486 strncpy(d->name, vhdr->label.name, GV_MAXDRIVENAME);
487 LIST_INIT(&d->subdisks);
488 LIST_INIT(&d->freelist);
490 /* We also need a freelist entry. */
491 fl = g_malloc(sizeof(*fl), M_WAITOK | M_ZERO);
492 fl->offset = GV_DATA_START;
494 LIST_INSERT_HEAD(&d->freelist, fl, freelist);
495 d->freelist_entries = 1;
497 /* Save it into the main configuration. */
498 LIST_INSERT_HEAD(&sc->drives, d, drive);
502 * Create bio queue, queue mutex and a worker thread, if
505 if (d->bqueue == NULL) {
506 d->bqueue = g_malloc(sizeof(struct bio_queue_head),
508 bioq_init(d->bqueue);
510 if (mtx_initialized(&d->bqueue_mtx) == 0)
511 mtx_init(&d->bqueue_mtx, "gv_drive", NULL, MTX_DEF);
513 if (!(d->flags & GV_DRIVE_THREAD_ACTIVE)) {
514 kthread_create(gv_drive_worker, d, NULL, 0, 0,
516 d->flags |= GV_DRIVE_THREAD_ACTIVE;
519 g_access(cp, -1, 0, 0);
524 strncpy(d->device, pp->name, GV_MAXDRIVENAME);
527 * Find out which subdisks belong to this drive and crosslink
530 LIST_FOREACH(s, &sc->subdisks, sd) {
531 if (!strncmp(s->drive, d->name, GV_MAXDRIVENAME))
532 /* XXX: errors ignored */
533 gv_sd_to_drive(sc, d, s, errstr,
537 /* This drive is now up for sure. */
538 gv_set_drive_state(d, GV_DRIVE_UP, 0);
541 * If there are subdisks on this drive, we need to create
542 * providers for them.
552 g_access(cp, -1, 0, 0);
555 g_destroy_consumer(cp);
561 * Modify the providers for the given drive 'd'. It is assumed that the
562 * subdisk list of 'd' is already correctly set up.
565 gv_drive_modify(struct gv_drive *d)
568 struct g_consumer *cp;
569 struct g_provider *pp, *pp2;
572 KASSERT(d != NULL, ("gv_drive_modify: null d"));
574 KASSERT(gp != NULL, ("gv_drive_modify: null gp"));
575 cp = LIST_FIRST(&gp->consumer);
576 KASSERT(cp != NULL, ("gv_drive_modify: null cp"));
578 KASSERT(pp != NULL, ("gv_drive_modify: null pp"));
582 LIST_FOREACH(s, &d->subdisks, from_drive) {
583 /* This subdisk already has a provider. */
584 if (s->provider != NULL)
586 pp2 = g_new_providerf(gp, "gvinum/sd/%s", s->name);
587 pp2->mediasize = s->size;
588 pp2->sectorsize = pp->sectorsize;
589 g_error_provider(pp2, 0);
596 gv_drive_dead(void *arg, int flag)
599 struct g_consumer *cp;
604 KASSERT(arg != NULL, ("gv_drive_dead: NULL arg"));
606 if (flag == EV_CANCEL)
610 if (d->state != GV_DRIVE_DOWN)
613 g_trace(G_T_TOPOLOGY, "gv_drive_dead(%s)", d->name);
619 LIST_FOREACH(cp, &gp->consumer, consumer) {
620 if (cp->nstart != cp->nend) {
621 printf("GEOM_VINUM: dead drive '%s' has still "
622 "active requests, can't detach consumer\n",
624 g_post_event(gv_drive_dead, d, M_WAITOK, d,
628 if (cp->acr != 0 || cp->acw != 0 || cp->ace != 0)
629 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
632 printf("GEOM_VINUM: lost drive '%s'\n", d->name);
634 LIST_FOREACH(s, &d->subdisks, from_drive) {
638 gv_kill_drive_thread(d);
640 g_wither_geom(gp, ENXIO);
644 gv_drive_destroy_geom(struct gctl_req *req, struct g_class *mp,
649 g_trace(G_T_TOPOLOGY, "gv_drive_destroy_geom: %s", gp->name);
653 gv_kill_drive_thread(d);
655 g_wither_geom(gp, ENXIO);
659 #define VINUMDRIVE_CLASS_NAME "VINUMDRIVE"
661 static struct g_class g_vinum_drive_class = {
662 .name = VINUMDRIVE_CLASS_NAME,
663 .version = G_VERSION,
664 .taste = gv_drive_taste,
665 .destroy_geom = gv_drive_destroy_geom
668 DECLARE_GEOM_CLASS(g_vinum_drive_class, g_vinum_drive);