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("$FreeBSD$");
39 #include "opt_compat.h"
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, NULL, 0, NULL));
147 error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
151 error = set_dumper(&kd.di, devtoname(dev), td, kda->kda_encryption,
152 kda->kda_key, kda->kda_encryptedkeysize, kda->kda_encryptedkey);
154 dev->si_flags |= SI_DUMPDEV;
160 init_dumpdev(struct cdev *dev)
162 struct diocskerneldump_arg kda;
163 struct g_consumer *cp;
164 const char *devprefix = "/dev/", *devname;
168 bzero(&kda, sizeof(kda));
174 len = strlen(devprefix);
175 devname = devtoname(dev);
176 if (strcmp(devname, dumpdev) != 0 &&
177 (strncmp(dumpdev, devprefix, len) != 0 ||
178 strcmp(devname, dumpdev + len) != 0))
181 cp = (struct g_consumer *)dev->si_drv2;
182 error = g_access(cp, 1, 0, 0);
186 error = g_dev_setdumpdev(dev, &kda, curthread);
192 (void)g_access(cp, -1, 0, 0);
198 g_dev_destroy(void *arg, int flags __unused)
200 struct g_consumer *cp;
202 struct g_dev_softc *sc;
203 char buf[SPECNAMELEN + 6];
209 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
210 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
211 devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
212 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
213 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
215 g_destroy_consumer(cp);
217 mtx_destroy(&sc->sc_mtx);
227 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
228 printf("%s%s", p, gp->name);
235 g_dev_set_physpath(struct g_consumer *cp)
237 struct g_dev_softc *sc;
239 int error, physpath_len;
241 if (g_access(cp, 1, 0, 0) != 0)
245 physpath_len = MAXPATHLEN;
246 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
247 error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
248 g_access(cp, -1, 0, 0);
249 if (error == 0 && strlen(physpath) != 0) {
250 struct cdev *dev, *old_alias_dev;
251 struct cdev **alias_devp;
254 old_alias_dev = sc->sc_alias;
255 alias_devp = (struct cdev **)&sc->sc_alias;
256 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
257 old_alias_dev, physpath);
258 } else if (sc->sc_alias) {
259 destroy_dev((struct cdev *)sc->sc_alias);
266 g_dev_set_media(struct g_consumer *cp)
268 struct g_dev_softc *sc;
270 char buf[SPECNAMELEN + 6];
274 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
275 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
276 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
279 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
280 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
281 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
286 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
289 if (strcmp(attr, "GEOM::media") == 0) {
294 if (strcmp(attr, "GEOM::physpath") == 0) {
295 g_dev_set_physpath(cp);
301 g_dev_getprovider(struct cdev *dev)
303 struct g_consumer *cp;
308 if (dev->si_devsw != &g_dev_cdevsw)
311 return (cp->provider);
314 static struct g_geom *
315 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
318 struct g_geom_alias *gap;
319 struct g_consumer *cp;
320 struct g_dev_softc *sc;
322 struct cdev *dev, *adev;
323 char buf[SPECNAMELEN + 6];
325 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
327 gp = g_new_geomf(mp, "%s", pp->name);
328 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
329 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
330 cp = g_new_consumer(gp);
332 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
333 error = g_attach(cp, pp);
335 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
336 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
337 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
339 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
340 __func__, gp->name, error);
342 g_destroy_consumer(cp);
344 mtx_destroy(&sc->sc_mtx);
348 dev->si_flags |= SI_UNMAPPED;
351 dev->si_iosize_max = MAXPHYS;
353 error = init_dumpdev(dev);
355 printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
356 __func__, gp->name, error);
358 g_dev_attrchanged(cp, "GEOM::physpath");
359 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
360 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
362 * Now add all the aliases for this drive
364 LIST_FOREACH(gap, &pp->geom->aliases, ga_next) {
365 error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
366 "%s", gap->ga_alias);
368 printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
369 __func__, gap->ga_alias, error);
372 snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
373 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
380 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
382 struct g_consumer *cp;
383 struct g_dev_softc *sc;
388 return (ENXIO); /* g_dev_taste() not done yet */
389 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
390 cp->geom->name, flags, fmt, td);
392 r = flags & FREAD ? 1 : 0;
393 w = flags & FWRITE ? 1 : 0;
395 e = flags & O_EXCL ? 1 : 0;
401 * This happens on attempt to open a device node with O_EXEC.
408 * When running in very secure mode, do not allow
409 * opens for writing of any disks.
411 error = securelevel_ge(td->td_ucred, 2);
416 error = g_access(cp, r, w, e);
420 mtx_lock(&sc->sc_mtx);
421 if (sc->sc_open == 0 && sc->sc_active != 0)
422 wakeup(&sc->sc_active);
423 sc->sc_open += r + w + e;
424 mtx_unlock(&sc->sc_mtx);
430 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
432 struct g_consumer *cp;
433 struct g_dev_softc *sc;
439 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
440 cp->geom->name, flags, fmt, td);
442 r = flags & FREAD ? -1 : 0;
443 w = flags & FWRITE ? -1 : 0;
445 e = flags & O_EXCL ? -1 : 0;
451 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
452 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
453 * which would result in zero deltas, which in turn would cause
454 * panic in g_access(9).
456 * Note that we cannot zero the counters (ie. do "r = cp->acr"
457 * etc) instead, because the consumer might be opened in another
464 mtx_lock(&sc->sc_mtx);
465 sc->sc_open += r + w + e;
466 while (sc->sc_open == 0 && sc->sc_active != 0)
467 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
468 mtx_unlock(&sc->sc_mtx);
470 error = g_access(cp, r, w, e);
476 * XXX: Until we have unmessed the ioctl situation, there is a race against
477 * XXX: a concurrent orphanization. We cannot close it by holding topology
478 * XXX: since that would prevent us from doing our job, and stalling events
479 * XXX: will break (actually: stall) the BSD disklabel hacks.
482 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
484 struct g_consumer *cp;
485 struct g_provider *pp;
486 off_t offset, length, chunk, odd;
493 KASSERT(cp->acr || cp->acw,
494 ("Consumer with zero access count in g_dev_ioctl"));
496 i = IOCPARM_LEN(cmd);
498 case DIOCGSECTORSIZE:
499 *(u_int *)data = cp->provider->sectorsize;
500 if (*(u_int *)data == 0)
504 *(off_t *)data = cp->provider->mediasize;
505 if (*(off_t *)data == 0)
509 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
510 if (error == 0 && *(u_int *)data == 0)
514 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
515 if (error == 0 && *(u_int *)data == 0)
518 case DIOCGFRONTSTUFF:
519 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
521 #ifdef COMPAT_FREEBSD11
522 case DIOCSKERNELDUMP_FREEBSD11:
524 struct diocskerneldump_arg kda;
526 bzero(&kda, sizeof(kda));
527 kda.kda_encryption = KERNELDUMP_ENC_NONE;
528 kda.kda_enable = (uint8_t)*(u_int *)data;
529 if (kda.kda_enable == 0)
530 error = g_dev_setdumpdev(NULL, NULL, td);
532 error = g_dev_setdumpdev(dev, &kda, td);
536 case DIOCSKERNELDUMP:
538 struct diocskerneldump_arg *kda;
539 uint8_t *encryptedkey;
541 kda = (struct diocskerneldump_arg *)data;
542 if (kda->kda_enable == 0) {
543 error = g_dev_setdumpdev(NULL, NULL, td);
547 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
548 if (kda->kda_encryptedkeysize <= 0 ||
549 kda->kda_encryptedkeysize >
550 KERNELDUMP_ENCKEY_MAX_SIZE) {
553 encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
555 error = copyin(kda->kda_encryptedkey, encryptedkey,
556 kda->kda_encryptedkeysize);
561 kda->kda_encryptedkey = encryptedkey;
562 error = g_dev_setdumpdev(dev, kda, td);
564 if (encryptedkey != NULL) {
565 explicit_bzero(encryptedkey, kda->kda_encryptedkeysize);
566 free(encryptedkey, M_TEMP);
568 explicit_bzero(kda, sizeof(*kda));
572 error = g_io_flush(cp);
575 offset = ((off_t *)data)[0];
576 length = ((off_t *)data)[1];
577 if ((offset % cp->provider->sectorsize) != 0 ||
578 (length % cp->provider->sectorsize) != 0 || length <= 0) {
579 printf("%s: offset=%jd length=%jd\n", __func__, offset,
586 if (g_dev_del_max_sectors != 0 && chunk >
587 g_dev_del_max_sectors * cp->provider->sectorsize) {
588 chunk = g_dev_del_max_sectors *
589 cp->provider->sectorsize;
590 if (cp->provider->stripesize > 0) {
591 odd = (offset + chunk +
592 cp->provider->stripeoffset) %
593 cp->provider->stripesize;
598 error = g_delete_data(cp, offset, chunk);
604 * Since the request size can be large, the service
605 * time can be is likewise. We make this ioctl
606 * interruptible by checking for signals for each bio.
613 error = g_io_getattr("GEOM::ident", cp, &i, data);
615 case DIOCGPROVIDERNAME:
618 strlcpy(data, pp->name, i);
620 case DIOCGSTRIPESIZE:
621 *(off_t *)data = cp->provider->stripesize;
623 case DIOCGSTRIPEOFFSET:
624 *(off_t *)data = cp->provider->stripeoffset;
627 error = g_io_getattr("GEOM::physpath", cp, &i, data);
628 if (error == 0 && *(char *)data == '\0')
632 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
634 if (arg->len > sizeof(arg->value)) {
638 error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
642 struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
643 struct disk_zone_rep_entry *new_entries, *old_entries;
644 struct disk_zone_report *rep;
652 if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
654 rep = &zone_args->zone_params.report;
655 alloc_size = rep->entries_allocated *
656 sizeof(struct disk_zone_rep_entry);
658 new_entries = g_malloc(alloc_size,
660 old_entries = rep->entries;
661 rep->entries = new_entries;
663 error = g_io_zonecmd(zone_args, cp);
664 if ((zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES)
667 error = copyout(new_entries, old_entries, alloc_size);
669 if ((old_entries != NULL)
671 rep->entries = old_entries;
673 if (new_entries != NULL)
678 if (cp->provider->geom->ioctl != NULL) {
679 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
689 g_dev_done(struct bio *bp2)
691 struct g_consumer *cp;
692 struct g_dev_softc *sc;
698 bp = bp2->bio_parent;
699 bp->bio_error = bp2->bio_error;
700 bp->bio_completed = bp2->bio_completed;
701 bp->bio_resid = bp->bio_length - bp2->bio_completed;
702 if (bp2->bio_cmd == BIO_ZONE)
703 bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
705 if (bp2->bio_error != 0) {
706 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
707 bp2, bp2->bio_error);
708 bp->bio_flags |= BIO_ERROR;
710 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
711 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
715 mtx_lock(&sc->sc_mtx);
716 if ((--sc->sc_active) == 0) {
717 if (sc->sc_open == 0)
718 wakeup(&sc->sc_active);
719 if (sc->sc_dev == NULL)
722 mtx_unlock(&sc->sc_mtx);
724 g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
729 g_dev_strategy(struct bio *bp)
731 struct g_consumer *cp;
734 struct g_dev_softc *sc;
736 KASSERT(bp->bio_cmd == BIO_READ ||
737 bp->bio_cmd == BIO_WRITE ||
738 bp->bio_cmd == BIO_DELETE ||
739 bp->bio_cmd == BIO_FLUSH ||
740 bp->bio_cmd == BIO_ZONE,
741 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
745 KASSERT(cp->acr || cp->acw,
746 ("Consumer with zero access count in g_dev_strategy"));
747 biotrack(bp, __func__);
749 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
750 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
751 bp->bio_resid = bp->bio_bcount;
752 biofinish(bp, NULL, EINVAL);
756 mtx_lock(&sc->sc_mtx);
757 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
759 mtx_unlock(&sc->sc_mtx);
763 * XXX: This is not an ideal solution, but I believe it to
764 * XXX: deadlock safely, all things considered.
766 bp2 = g_clone_bio(bp);
769 pause("gdstrat", hz / 10);
771 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
772 bp2->bio_done = g_dev_done;
774 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
775 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
776 bp2->bio_data, bp2->bio_cmd);
777 g_io_request(bp2, cp);
778 KASSERT(cp->acr || cp->acw,
779 ("g_dev_strategy raced with g_dev_close and lost"));
786 * Called by devfs when asynchronous device destruction is completed.
787 * - Mark that we have no attached device any more.
788 * - If there are no outstanding requests, schedule geom destruction.
789 * Otherwise destruction will be scheduled later by g_dev_done().
793 g_dev_callback(void *arg)
795 struct g_consumer *cp;
796 struct g_dev_softc *sc;
801 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
803 mtx_lock(&sc->sc_mtx);
806 destroy = (sc->sc_active == 0);
807 mtx_unlock(&sc->sc_mtx);
809 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
815 * Called from below when the provider orphaned us.
816 * - Clear any dump settings.
817 * - Request asynchronous device destruction to prevent any more requests
818 * from coming in. The provider is already marked with an error, so
819 * anything which comes in the interim will be returned immediately.
823 g_dev_orphan(struct g_consumer *cp)
826 struct g_dev_softc *sc;
831 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
833 /* Reset any dump-area set on this device */
834 if (dev->si_flags & SI_DUMPDEV)
835 (void)set_dumper(NULL, NULL, curthread, 0, NULL, 0, NULL);
837 /* Destroy the struct cdev *so we get no more requests */
838 destroy_dev_sched_cb(dev, g_dev_callback, cp);
841 DECLARE_GEOM_CLASS(g_dev_class, g_dev);