2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2002 Poul-Henning Kamp
5 * Copyright (c) 2002 Networks Associates Technology, Inc.
8 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
9 * and NAI Labs, the Security Research Division of Network Associates, Inc.
10 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
11 * DARPA CHATS research program.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. The names of the authors may not be used to endorse or promote
22 * products derived from this software without specific prior written
25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #include <sys/kernel.h>
46 #include <sys/ctype.h>
50 #include <sys/mutex.h>
52 #include <sys/errno.h>
55 #include <sys/fcntl.h>
56 #include <sys/limits.h>
57 #include <sys/sysctl.h>
58 #include <geom/geom.h>
59 #include <geom/geom_int.h>
60 #include <machine/stdarg.h>
65 struct cdev *sc_alias;
70 static d_open_t g_dev_open;
71 static d_close_t g_dev_close;
72 static d_strategy_t g_dev_strategy;
73 static d_ioctl_t g_dev_ioctl;
75 static struct cdevsw g_dev_cdevsw = {
76 .d_version = D_VERSION,
78 .d_close = g_dev_close,
81 .d_ioctl = g_dev_ioctl,
82 .d_strategy = g_dev_strategy,
84 .d_flags = D_DISK | D_TRACKCLOSE,
87 static g_init_t g_dev_init;
88 static g_fini_t g_dev_fini;
89 static g_taste_t g_dev_taste;
90 static g_orphan_t g_dev_orphan;
91 static g_attrchanged_t g_dev_attrchanged;
93 static struct g_class g_dev_class = {
99 .orphan = g_dev_orphan,
100 .attrchanged = g_dev_attrchanged
104 * We target 262144 (8 x 32768) sectors by default as this significantly
105 * increases the throughput on commonly used SSD's with a marginal
106 * increase in non-interruptible request latency.
108 static uint64_t g_dev_del_max_sectors = 262144;
109 SYSCTL_DECL(_kern_geom);
110 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
111 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
112 &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
113 "delete request sent to the provider. Larger requests are chunked "
114 "so they can be interrupted. (0 = disable chunking)");
116 static char *dumpdev = NULL;
118 g_dev_init(struct g_class *mp)
121 dumpdev = kern_getenv("dumpdev");
125 g_dev_fini(struct g_class *mp)
133 g_dev_setdumpdev(struct cdev *dev, struct diocskerneldump_arg *kda,
136 struct g_kerneldump kd;
137 struct g_consumer *cp;
140 if (dev == NULL || kda == NULL)
141 return (set_dumper(NULL, NULL, td, 0, 0, NULL, 0, NULL));
147 error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
151 error = set_dumper(&kd.di, devtoname(dev), td, kda->kda_compression,
152 kda->kda_encryption, kda->kda_key, kda->kda_encryptedkeysize,
153 kda->kda_encryptedkey);
155 dev->si_flags |= SI_DUMPDEV;
161 init_dumpdev(struct cdev *dev)
163 struct diocskerneldump_arg kda;
164 struct g_consumer *cp;
165 const char *devprefix = "/dev/", *devname;
169 bzero(&kda, sizeof(kda));
175 len = strlen(devprefix);
176 devname = devtoname(dev);
177 if (strcmp(devname, dumpdev) != 0 &&
178 (strncmp(dumpdev, devprefix, len) != 0 ||
179 strcmp(devname, dumpdev + len) != 0))
182 cp = (struct g_consumer *)dev->si_drv2;
183 error = g_access(cp, 1, 0, 0);
187 error = g_dev_setdumpdev(dev, &kda, curthread);
193 (void)g_access(cp, -1, 0, 0);
199 g_dev_destroy(void *arg, int flags __unused)
201 struct g_consumer *cp;
203 struct g_dev_softc *sc;
204 char buf[SPECNAMELEN + 6];
210 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
211 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
212 devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
213 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
214 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
216 g_destroy_consumer(cp);
218 mtx_destroy(&sc->sc_mtx);
228 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
229 printf("%s%s", p, gp->name);
236 g_dev_set_physpath(struct g_consumer *cp)
238 struct g_dev_softc *sc;
240 int error, physpath_len;
242 if (g_access(cp, 1, 0, 0) != 0)
246 physpath_len = MAXPATHLEN;
247 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
248 error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
249 g_access(cp, -1, 0, 0);
250 if (error == 0 && strlen(physpath) != 0) {
251 struct cdev *dev, *old_alias_dev;
252 struct cdev **alias_devp;
255 old_alias_dev = sc->sc_alias;
256 alias_devp = (struct cdev **)&sc->sc_alias;
257 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
258 old_alias_dev, physpath);
259 } else if (sc->sc_alias) {
260 destroy_dev((struct cdev *)sc->sc_alias);
267 g_dev_set_media(struct g_consumer *cp)
269 struct g_dev_softc *sc;
271 char buf[SPECNAMELEN + 6];
275 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
276 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
277 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
280 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
281 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
282 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
287 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
290 if (strcmp(attr, "GEOM::media") == 0) {
295 if (strcmp(attr, "GEOM::physpath") == 0) {
296 g_dev_set_physpath(cp);
302 g_dev_getprovider(struct cdev *dev)
304 struct g_consumer *cp;
309 if (dev->si_devsw != &g_dev_cdevsw)
312 return (cp->provider);
315 static struct g_geom *
316 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
319 struct g_geom_alias *gap;
320 struct g_consumer *cp;
321 struct g_dev_softc *sc;
323 struct cdev *dev, *adev;
324 char buf[SPECNAMELEN + 6];
326 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
328 gp = g_new_geomf(mp, "%s", pp->name);
329 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
330 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
331 cp = g_new_consumer(gp);
333 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
334 error = g_attach(cp, pp);
336 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
337 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
338 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
340 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
341 __func__, gp->name, error);
343 g_destroy_consumer(cp);
345 mtx_destroy(&sc->sc_mtx);
349 dev->si_flags |= SI_UNMAPPED;
352 dev->si_iosize_max = MAXPHYS;
354 error = init_dumpdev(dev);
356 printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
357 __func__, gp->name, error);
359 g_dev_attrchanged(cp, "GEOM::physpath");
360 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
361 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
363 * Now add all the aliases for this drive
365 LIST_FOREACH(gap, &pp->geom->aliases, ga_next) {
366 error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
367 "%s", gap->ga_alias);
369 printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
370 __func__, gap->ga_alias, error);
373 snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
374 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
381 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
383 struct g_consumer *cp;
384 struct g_dev_softc *sc;
389 return (ENXIO); /* g_dev_taste() not done yet */
390 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
391 cp->geom->name, flags, fmt, td);
393 r = flags & FREAD ? 1 : 0;
394 w = flags & FWRITE ? 1 : 0;
396 e = flags & O_EXCL ? 1 : 0;
402 * This happens on attempt to open a device node with O_EXEC.
409 * When running in very secure mode, do not allow
410 * opens for writing of any disks.
412 error = securelevel_ge(td->td_ucred, 2);
417 error = g_access(cp, r, w, e);
421 mtx_lock(&sc->sc_mtx);
422 if (sc->sc_open == 0 && sc->sc_active != 0)
423 wakeup(&sc->sc_active);
424 sc->sc_open += r + w + e;
425 mtx_unlock(&sc->sc_mtx);
431 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
433 struct g_consumer *cp;
434 struct g_dev_softc *sc;
440 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
441 cp->geom->name, flags, fmt, td);
443 r = flags & FREAD ? -1 : 0;
444 w = flags & FWRITE ? -1 : 0;
446 e = flags & O_EXCL ? -1 : 0;
452 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
453 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
454 * which would result in zero deltas, which in turn would cause
455 * panic in g_access(9).
457 * Note that we cannot zero the counters (ie. do "r = cp->acr"
458 * etc) instead, because the consumer might be opened in another
465 mtx_lock(&sc->sc_mtx);
466 sc->sc_open += r + w + e;
467 while (sc->sc_open == 0 && sc->sc_active != 0)
468 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
469 mtx_unlock(&sc->sc_mtx);
471 error = g_access(cp, r, w, e);
477 * XXX: Until we have unmessed the ioctl situation, there is a race against
478 * XXX: a concurrent orphanization. We cannot close it by holding topology
479 * XXX: since that would prevent us from doing our job, and stalling events
480 * XXX: will break (actually: stall) the BSD disklabel hacks.
483 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
485 struct g_consumer *cp;
486 struct g_provider *pp;
487 off_t offset, length, chunk, odd;
494 KASSERT(cp->acr || cp->acw,
495 ("Consumer with zero access count in g_dev_ioctl"));
497 i = IOCPARM_LEN(cmd);
499 case DIOCGSECTORSIZE:
500 *(u_int *)data = cp->provider->sectorsize;
501 if (*(u_int *)data == 0)
505 *(off_t *)data = cp->provider->mediasize;
506 if (*(off_t *)data == 0)
510 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
511 if (error == 0 && *(u_int *)data == 0)
515 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
516 if (error == 0 && *(u_int *)data == 0)
519 case DIOCGFRONTSTUFF:
520 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
522 #ifdef COMPAT_FREEBSD11
523 case DIOCSKERNELDUMP_FREEBSD11:
525 struct diocskerneldump_arg kda;
527 bzero(&kda, sizeof(kda));
528 kda.kda_encryption = KERNELDUMP_ENC_NONE;
529 kda.kda_enable = (uint8_t)*(u_int *)data;
530 if (kda.kda_enable == 0)
531 error = g_dev_setdumpdev(NULL, NULL, td);
533 error = g_dev_setdumpdev(dev, &kda, td);
537 case DIOCSKERNELDUMP:
539 struct diocskerneldump_arg *kda;
540 uint8_t *encryptedkey;
542 kda = (struct diocskerneldump_arg *)data;
543 if (kda->kda_enable == 0) {
544 error = g_dev_setdumpdev(NULL, NULL, td);
548 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
549 if (kda->kda_encryptedkeysize <= 0 ||
550 kda->kda_encryptedkeysize >
551 KERNELDUMP_ENCKEY_MAX_SIZE) {
554 encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
556 error = copyin(kda->kda_encryptedkey, encryptedkey,
557 kda->kda_encryptedkeysize);
562 kda->kda_encryptedkey = encryptedkey;
563 error = g_dev_setdumpdev(dev, kda, td);
565 if (encryptedkey != NULL) {
566 explicit_bzero(encryptedkey, kda->kda_encryptedkeysize);
567 free(encryptedkey, M_TEMP);
569 explicit_bzero(kda, sizeof(*kda));
573 error = g_io_flush(cp);
576 offset = ((off_t *)data)[0];
577 length = ((off_t *)data)[1];
578 if ((offset % cp->provider->sectorsize) != 0 ||
579 (length % cp->provider->sectorsize) != 0 || length <= 0) {
580 printf("%s: offset=%jd length=%jd\n", __func__, offset,
587 if (g_dev_del_max_sectors != 0 && chunk >
588 g_dev_del_max_sectors * cp->provider->sectorsize) {
589 chunk = g_dev_del_max_sectors *
590 cp->provider->sectorsize;
591 if (cp->provider->stripesize > 0) {
592 odd = (offset + chunk +
593 cp->provider->stripeoffset) %
594 cp->provider->stripesize;
599 error = g_delete_data(cp, offset, chunk);
605 * Since the request size can be large, the service
606 * time can be is likewise. We make this ioctl
607 * interruptible by checking for signals for each bio.
614 error = g_io_getattr("GEOM::ident", cp, &i, data);
616 case DIOCGPROVIDERNAME:
619 strlcpy(data, pp->name, i);
621 case DIOCGSTRIPESIZE:
622 *(off_t *)data = cp->provider->stripesize;
624 case DIOCGSTRIPEOFFSET:
625 *(off_t *)data = cp->provider->stripeoffset;
628 error = g_io_getattr("GEOM::physpath", cp, &i, data);
629 if (error == 0 && *(char *)data == '\0')
633 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
635 if (arg->len > sizeof(arg->value)) {
639 error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
643 struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
644 struct disk_zone_rep_entry *new_entries, *old_entries;
645 struct disk_zone_report *rep;
653 if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
655 rep = &zone_args->zone_params.report;
656 alloc_size = rep->entries_allocated *
657 sizeof(struct disk_zone_rep_entry);
659 new_entries = g_malloc(alloc_size,
661 old_entries = rep->entries;
662 rep->entries = new_entries;
664 error = g_io_zonecmd(zone_args, cp);
665 if ((zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES)
668 error = copyout(new_entries, old_entries, alloc_size);
670 if ((old_entries != NULL)
672 rep->entries = old_entries;
674 if (new_entries != NULL)
679 if (cp->provider->geom->ioctl != NULL) {
680 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
690 g_dev_done(struct bio *bp2)
692 struct g_consumer *cp;
693 struct g_dev_softc *sc;
699 bp = bp2->bio_parent;
700 bp->bio_error = bp2->bio_error;
701 bp->bio_completed = bp2->bio_completed;
702 bp->bio_resid = bp->bio_length - bp2->bio_completed;
703 if (bp2->bio_cmd == BIO_ZONE)
704 bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
706 if (bp2->bio_error != 0) {
707 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
708 bp2, bp2->bio_error);
709 bp->bio_flags |= BIO_ERROR;
711 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
712 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
716 mtx_lock(&sc->sc_mtx);
717 if ((--sc->sc_active) == 0) {
718 if (sc->sc_open == 0)
719 wakeup(&sc->sc_active);
720 if (sc->sc_dev == NULL)
723 mtx_unlock(&sc->sc_mtx);
725 g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
730 g_dev_strategy(struct bio *bp)
732 struct g_consumer *cp;
735 struct g_dev_softc *sc;
737 KASSERT(bp->bio_cmd == BIO_READ ||
738 bp->bio_cmd == BIO_WRITE ||
739 bp->bio_cmd == BIO_DELETE ||
740 bp->bio_cmd == BIO_FLUSH ||
741 bp->bio_cmd == BIO_ZONE,
742 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
746 KASSERT(cp->acr || cp->acw,
747 ("Consumer with zero access count in g_dev_strategy"));
748 biotrack(bp, __func__);
750 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
751 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
752 bp->bio_resid = bp->bio_bcount;
753 biofinish(bp, NULL, EINVAL);
757 mtx_lock(&sc->sc_mtx);
758 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
760 mtx_unlock(&sc->sc_mtx);
764 * XXX: This is not an ideal solution, but I believe it to
765 * XXX: deadlock safely, all things considered.
767 bp2 = g_clone_bio(bp);
770 pause("gdstrat", hz / 10);
772 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
773 bp2->bio_done = g_dev_done;
775 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
776 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
777 bp2->bio_data, bp2->bio_cmd);
778 g_io_request(bp2, cp);
779 KASSERT(cp->acr || cp->acw,
780 ("g_dev_strategy raced with g_dev_close and lost"));
787 * Called by devfs when asynchronous device destruction is completed.
788 * - Mark that we have no attached device any more.
789 * - If there are no outstanding requests, schedule geom destruction.
790 * Otherwise destruction will be scheduled later by g_dev_done().
794 g_dev_callback(void *arg)
796 struct g_consumer *cp;
797 struct g_dev_softc *sc;
802 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
804 mtx_lock(&sc->sc_mtx);
807 destroy = (sc->sc_active == 0);
808 mtx_unlock(&sc->sc_mtx);
810 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
816 * Called from below when the provider orphaned us.
817 * - Clear any dump settings.
818 * - Request asynchronous device destruction to prevent any more requests
819 * from coming in. The provider is already marked with an error, so
820 * anything which comes in the interim will be returned immediately.
824 g_dev_orphan(struct g_consumer *cp)
827 struct g_dev_softc *sc;
832 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
834 /* Reset any dump-area set on this device */
835 if (dev->si_flags & SI_DUMPDEV)
836 (void)set_dumper(NULL, NULL, curthread, 0, 0, NULL, 0, NULL);
838 /* Destroy the struct cdev *so we get no more requests */
839 destroy_dev_sched_cb(dev, g_dev_callback, cp);
842 DECLARE_GEOM_CLASS(g_dev_class, g_dev);