2 * Copyright (c) 2002 Poul-Henning Kamp
3 * Copyright (c) 2002 Networks Associates Technology, Inc.
6 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
7 * and NAI Labs, the Security Research Division of Network Associates, Inc.
8 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 * DARPA CHATS research program.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The names of the authors may not be used to endorse or promote
20 * products derived from this software without specific prior written
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
46 #include <sys/ctype.h>
47 #include <sys/fcntl.h>
48 #include <sys/malloc.h>
50 #include <sys/sysctl.h>
51 #include <sys/devicestat.h>
52 #include <machine/md_var.h>
55 #include <sys/mutex.h>
56 #include <geom/geom.h>
57 #include <geom/geom_disk.h>
58 #include <geom/geom_int.h>
60 #include <dev/led/led.h>
65 struct sysctl_ctx_list sysctl_ctx;
66 struct sysctl_oid *sysctl_tree;
71 static g_access_t g_disk_access;
72 static g_start_t g_disk_start;
73 static g_ioctl_t g_disk_ioctl;
74 static g_dumpconf_t g_disk_dumpconf;
75 static g_provgone_t g_disk_providergone;
77 static struct g_class g_disk_class = {
80 .start = g_disk_start,
81 .access = g_disk_access,
82 .ioctl = g_disk_ioctl,
83 .providergone = g_disk_providergone,
84 .dumpconf = g_disk_dumpconf,
87 SYSCTL_DECL(_kern_geom);
88 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0,
91 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
94 g_disk_lock_giant(struct disk *dp)
97 if (dp->d_flags & DISKFLAG_NEEDSGIANT)
102 g_disk_unlock_giant(struct disk *dp)
105 if (dp->d_flags & DISKFLAG_NEEDSGIANT)
110 g_disk_access(struct g_provider *pp, int r, int w, int e)
113 struct g_disk_softc *sc;
116 g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
120 if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
122 * Allow decreasing access count even if disk is not
125 if (r <= 0 && w <= 0 && e <= 0)
133 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
134 if (dp->d_open != NULL) {
135 g_disk_lock_giant(dp);
136 error = dp->d_open(dp);
137 if (bootverbose && error != 0)
138 printf("Opened disk %s -> %d\n",
140 g_disk_unlock_giant(dp);
144 pp->mediasize = dp->d_mediasize;
145 pp->sectorsize = dp->d_sectorsize;
146 if (dp->d_flags & DISKFLAG_CANDELETE)
147 pp->flags |= G_PF_CANDELETE;
149 pp->flags &= ~G_PF_CANDELETE;
150 pp->stripeoffset = dp->d_stripeoffset;
151 pp->stripesize = dp->d_stripesize;
152 dp->d_flags |= DISKFLAG_OPEN;
153 if (dp->d_maxsize == 0) {
154 printf("WARNING: Disk drive %s%d has no d_maxsize\n",
155 dp->d_name, dp->d_unit);
156 dp->d_maxsize = DFLTPHYS;
158 if (dp->d_delmaxsize == 0) {
159 if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
160 printf("WARNING: Disk drive %s%d has no "
161 "d_delmaxsize\n", dp->d_name, dp->d_unit);
163 dp->d_delmaxsize = dp->d_maxsize;
165 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
166 if (dp->d_close != NULL) {
167 g_disk_lock_giant(dp);
168 error = dp->d_close(dp);
170 printf("Closed disk %s -> %d\n",
172 g_disk_unlock_giant(dp);
174 sc->state = G_STATE_ACTIVE;
176 led_set(sc->led, "0");
177 dp->d_flags &= ~DISKFLAG_OPEN;
183 g_disk_kerneldump(struct bio *bp, struct disk *dp)
185 struct g_kerneldump *gkd;
188 gkd = (struct g_kerneldump*)bp->bio_data;
189 gp = bp->bio_to->geom;
190 g_trace(G_T_TOPOLOGY, "g_disk_kernedump(%s, %jd, %jd)",
191 gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
192 if (dp->d_dump == NULL) {
193 g_io_deliver(bp, ENODEV);
196 gkd->di.dumper = dp->d_dump;
198 gkd->di.blocksize = dp->d_sectorsize;
199 gkd->di.maxiosize = dp->d_maxsize;
200 gkd->di.mediaoffset = gkd->offset;
201 if ((gkd->offset + gkd->length) > dp->d_mediasize)
202 gkd->length = dp->d_mediasize - gkd->offset;
203 gkd->di.mediasize = gkd->length;
208 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
212 memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
213 if (sc->led[0] != 0) {
218 case G_STATE_REBUILD:
228 led_set(sc->led, cmd);
234 g_disk_done(struct bio *bp)
237 struct g_disk_softc *sc;
239 /* See "notes" for why we need a mutex here */
240 /* XXX: will witness accept a mix of Giant/unGiant drivers here ? */
241 bp2 = bp->bio_parent;
242 sc = bp2->bio_to->private;
243 bp->bio_completed = bp->bio_length - bp->bio_resid;
244 mtx_lock(&sc->done_mtx);
245 if (bp2->bio_error == 0)
246 bp2->bio_error = bp->bio_error;
247 bp2->bio_completed += bp->bio_completed;
248 if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) != 0)
249 devstat_end_transaction_bio(sc->dp->d_devstat, bp);
252 if (bp2->bio_children == bp2->bio_inbed) {
253 bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
254 g_io_deliver(bp2, bp2->bio_error);
256 mtx_unlock(&sc->done_mtx);
260 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
263 struct g_disk_softc *sc;
269 if (dp->d_ioctl == NULL)
271 g_disk_lock_giant(dp);
272 error = dp->d_ioctl(dp, cmd, data, fflag, td);
273 g_disk_unlock_giant(dp);
278 g_disk_start(struct bio *bp)
280 struct bio *bp2, *bp3;
282 struct g_disk_softc *sc;
286 sc = bp->bio_to->private;
287 if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
288 g_io_deliver(bp, ENXIO);
292 switch(bp->bio_cmd) {
294 if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
303 bp2 = g_clone_bio(bp);
311 d_maxsize = (bp->bio_cmd == BIO_DELETE &&
312 (dp->d_flags & DISKFLAG_LACKS_DELMAX) == 0) ?
313 dp->d_delmaxsize : dp->d_maxsize;
315 bp2->bio_offset += off;
316 bp2->bio_length -= off;
317 if ((bp->bio_flags & BIO_UNMAPPED) == 0) {
318 bp2->bio_data += off;
320 KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO)
322 ("unmapped bio not supported by disk %s",
324 bp2->bio_ma += off / PAGE_SIZE;
325 bp2->bio_ma_offset += off;
326 bp2->bio_ma_offset %= PAGE_SIZE;
327 bp2->bio_ma_n -= off / PAGE_SIZE;
329 if (bp2->bio_length > d_maxsize) {
331 * XXX: If we have a stripesize we should really
332 * use it here. Care should be taken in the delete
333 * case if this is done as deletes can be very
334 * sensitive to size given how they are processed.
336 bp2->bio_length = d_maxsize;
337 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
338 bp2->bio_ma_n = howmany(
340 bp2->bio_length, PAGE_SIZE);
344 * To avoid a race, we need to grab the next bio
345 * before we schedule this one. See "notes".
347 bp3 = g_clone_bio(bp);
349 bp->bio_error = ENOMEM;
351 bp2->bio_done = g_disk_done;
352 bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
353 bp2->bio_bcount = bp2->bio_length;
355 devstat_start_transaction_bio(dp->d_devstat, bp2);
356 g_disk_lock_giant(dp);
358 g_disk_unlock_giant(dp);
361 } while (bp2 != NULL);
364 /* Give the driver a chance to override */
365 if (dp->d_getattr != NULL) {
366 if (bp->bio_disk == NULL)
368 error = dp->d_getattr(bp);
373 if (g_handleattr_int(bp, "GEOM::candelete",
374 (dp->d_flags & DISKFLAG_CANDELETE) != 0))
376 else if (g_handleattr_int(bp, "GEOM::fwsectors",
379 else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
381 else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0))
383 else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
385 else if (g_handleattr(bp, "GEOM::hba_vendor",
386 &dp->d_hba_vendor, 2))
388 else if (g_handleattr(bp, "GEOM::hba_device",
389 &dp->d_hba_device, 2))
391 else if (g_handleattr(bp, "GEOM::hba_subvendor",
392 &dp->d_hba_subvendor, 2))
394 else if (g_handleattr(bp, "GEOM::hba_subdevice",
395 &dp->d_hba_subdevice, 2))
397 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
398 g_disk_kerneldump(bp, dp);
399 else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
400 g_disk_setstate(bp, sc);
405 g_trace(G_T_BIO, "g_disk_flushcache(%s)",
407 if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
411 bp2 = g_clone_bio(bp);
413 g_io_deliver(bp, ENOMEM);
416 bp2->bio_done = g_disk_done;
418 g_disk_lock_giant(dp);
420 g_disk_unlock_giant(dp);
426 if (error != EJUSTRETURN)
427 g_io_deliver(bp, error);
432 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
436 struct g_disk_softc *sc;
441 if (sc == NULL || (dp = sc->dp) == NULL)
443 if (indent == NULL) {
444 sbuf_printf(sb, " hd %u", dp->d_fwheads);
445 sbuf_printf(sb, " sc %u", dp->d_fwsectors);
449 sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
450 indent, dp->d_fwheads);
451 sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
452 indent, dp->d_fwsectors);
453 if (dp->d_getattr != NULL) {
454 buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
457 bp->bio_attribute = "GEOM::ident";
458 bp->bio_length = DISK_IDENT_SIZE;
460 res = dp->d_getattr(bp);
461 sbuf_printf(sb, "%s<ident>%s</ident>\n", indent,
462 res == 0 ? buf: dp->d_ident);
463 bp->bio_attribute = "GEOM::lunid";
464 bp->bio_length = DISK_IDENT_SIZE;
466 if (dp->d_getattr(bp) == 0)
467 sbuf_printf(sb, "%s<lunid>%s</lunid>\n",
472 sbuf_printf(sb, "%s<ident>%s</ident>\n", indent,
474 sbuf_printf(sb, "%s<descr>%s</descr>\n", indent, dp->d_descr);
479 g_disk_create(void *arg, int flag)
482 struct g_provider *pp;
484 struct g_disk_softc *sc;
487 if (flag == EV_CANCEL)
491 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
492 mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
494 gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
496 pp = g_new_providerf(gp, "%s", gp->name);
497 pp->mediasize = dp->d_mediasize;
498 pp->sectorsize = dp->d_sectorsize;
499 if (dp->d_flags & DISKFLAG_CANDELETE)
500 pp->flags |= G_PF_CANDELETE;
501 pp->stripeoffset = dp->d_stripeoffset;
502 pp->stripesize = dp->d_stripesize;
503 if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
504 pp->flags |= G_PF_ACCEPT_UNMAPPED;
506 printf("GEOM: new disk %s\n", gp->name);
507 sysctl_ctx_init(&sc->sysctl_ctx);
508 snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
509 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
510 SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
511 CTLFLAG_RD, 0, tmpstr);
512 if (sc->sysctl_tree != NULL) {
513 snprintf(tmpstr, sizeof(tmpstr),
514 "kern.geom.disk.%s.led", gp->name);
515 TUNABLE_STR_FETCH(tmpstr, sc->led, sizeof(sc->led));
516 SYSCTL_ADD_STRING(&sc->sysctl_ctx,
517 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
518 CTLFLAG_RW | CTLFLAG_TUN, sc->led, sizeof(sc->led),
523 g_error_provider(pp, 0);
527 * We get this callback after all of the consumers have gone away, and just
528 * before the provider is freed. If the disk driver provided a d_gone
529 * callback, let them know that it is okay to free resources -- they won't
530 * be getting any more accesses from GEOM.
533 g_disk_providergone(struct g_provider *pp)
536 struct g_disk_softc *sc;
538 sc = (struct g_disk_softc *)pp->private;
542 * FreeBSD 9 started with VERSION_01 of the struct disk structure.
543 * However, g_gone was added in the middle of the branch. To
544 * cope with version being missing from struct disk, we set a flag
545 * in g_disk_create for VERSION_01 and avoid touching the d_gone
546 * field for old consumers.
548 if (dp != NULL && (dp->d_flags & DISKFLAG_LACKS_GONE) == 0 &&
551 if (sc->sysctl_tree != NULL) {
552 sysctl_ctx_free(&sc->sysctl_ctx);
553 sc->sysctl_tree = NULL;
555 if (sc->led[0] != 0) {
556 led_set(sc->led, "0");
560 pp->geom->softc = NULL;
561 mtx_destroy(&sc->done_mtx);
566 g_disk_destroy(void *ptr, int flag)
570 struct g_disk_softc *sc;
580 g_wither_geom(gp, ENXIO);
586 * We only allow printable characters in disk ident,
587 * the rest is converted to 'x<HH>'.
590 g_disk_ident_adjust(char *ident, size_t size)
592 char *p, tmp[4], newid[DISK_IDENT_SIZE];
595 for (p = ident; *p != '\0'; p++) {
600 snprintf(tmp, sizeof(tmp), "x%02hhx",
601 *(unsigned char *)p);
603 if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
607 strlcpy(ident, newid, size);
614 return (g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO));
618 disk_create(struct disk *dp, int version)
621 if (version != DISK_VERSION_03 && version != DISK_VERSION_02 &&
622 version != DISK_VERSION_01) {
623 printf("WARNING: Attempt to add disk %s%d %s",
624 dp->d_name, dp->d_unit,
625 " using incompatible ABI version of disk(9)\n");
626 printf("WARNING: Ignoring disk %s%d\n",
627 dp->d_name, dp->d_unit);
630 if (version == DISK_VERSION_01)
631 dp->d_flags |= DISKFLAG_LACKS_GONE;
632 if (version < DISK_VERSION_03)
633 dp->d_flags |= DISKFLAG_LACKS_DELMAX;
634 KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
635 KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
636 KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
637 KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
638 if (dp->d_devstat == NULL)
639 dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
640 dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
641 DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
643 g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
644 g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
648 disk_destroy(struct disk *dp)
653 if (dp->d_devstat != NULL)
654 devstat_remove_entry(dp->d_devstat);
655 g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
659 disk_gone(struct disk *dp)
662 struct g_provider *pp;
666 pp = LIST_FIRST(&gp->provider);
668 KASSERT(LIST_NEXT(pp, provider) == NULL,
669 ("geom %p has more than one provider", gp));
670 g_wither_provider(pp, ENXIO);
676 disk_attr_changed(struct disk *dp, const char *attr, int flag)
679 struct g_provider *pp;
683 LIST_FOREACH(pp, &gp->provider, provider)
684 (void)g_attr_changed(pp, attr, flag);
688 disk_media_changed(struct disk *dp, int flag)
691 struct g_provider *pp;
695 pp = LIST_FIRST(&gp->provider);
697 KASSERT(LIST_NEXT(pp, provider) == NULL,
698 ("geom %p has more than one provider", gp));
699 g_media_changed(pp, flag);
705 disk_media_gone(struct disk *dp, int flag)
708 struct g_provider *pp;
712 pp = LIST_FIRST(&gp->provider);
714 KASSERT(LIST_NEXT(pp, provider) == NULL,
715 ("geom %p has more than one provider", gp));
716 g_media_gone(pp, flag);
722 g_kern_disks(void *p, int flag __unused)
731 LIST_FOREACH(gp, &g_disk_class.geom, geom) {
732 sbuf_printf(sb, "%s%s", sp, gp->name);
739 sysctl_disks(SYSCTL_HANDLER_ARGS)
744 sb = sbuf_new_auto();
745 g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
746 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
751 SYSCTL_PROC(_kern, OID_AUTO, disks,
752 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
753 sysctl_disks, "A", "names of available disks");