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 <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/kernel.h>
44 #include <sys/ctype.h>
48 #include <sys/mutex.h>
50 #include <sys/errno.h>
53 #include <sys/fcntl.h>
54 #include <sys/limits.h>
55 #include <sys/sysctl.h>
56 #include <geom/geom.h>
57 #include <geom/geom_int.h>
58 #include <machine/stdarg.h>
63 struct cdev *sc_alias;
68 static d_open_t g_dev_open;
69 static d_close_t g_dev_close;
70 static d_strategy_t g_dev_strategy;
71 static d_ioctl_t g_dev_ioctl;
73 static struct cdevsw g_dev_cdevsw = {
74 .d_version = D_VERSION,
76 .d_close = g_dev_close,
79 .d_ioctl = g_dev_ioctl,
80 .d_strategy = g_dev_strategy,
82 .d_flags = D_DISK | D_TRACKCLOSE,
85 static g_init_t g_dev_init;
86 static g_fini_t g_dev_fini;
87 static g_taste_t g_dev_taste;
88 static g_orphan_t g_dev_orphan;
89 static g_attrchanged_t g_dev_attrchanged;
91 static struct g_class g_dev_class = {
97 .orphan = g_dev_orphan,
98 .attrchanged = g_dev_attrchanged
102 * We target 262144 (8 x 32768) sectors by default as this significantly
103 * increases the throughput on commonly used SSD's with a marginal
104 * increase in non-interruptible request latency.
106 static uint64_t g_dev_del_max_sectors = 262144;
107 SYSCTL_DECL(_kern_geom);
108 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
109 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
110 &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
111 "delete request sent to the provider. Larger requests are chunked "
112 "so they can be interrupted. (0 = disable chunking)");
114 static char *dumpdev = NULL;
116 g_dev_init(struct g_class *mp)
119 dumpdev = kern_getenv("dumpdev");
123 g_dev_fini(struct g_class *mp)
131 g_dev_setdumpdev(struct cdev *dev, struct thread *td)
133 struct g_kerneldump kd;
134 struct g_consumer *cp;
138 return (set_dumper(NULL, NULL, td));
144 error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
146 error = set_dumper(&kd.di, devtoname(dev), td);
148 dev->si_flags |= SI_DUMPDEV;
154 init_dumpdev(struct cdev *dev)
156 struct g_consumer *cp;
157 const char *devprefix = "/dev/", *devname;
164 len = strlen(devprefix);
165 devname = devtoname(dev);
166 if (strcmp(devname, dumpdev) != 0 &&
167 (strncmp(dumpdev, devprefix, len) != 0 ||
168 strcmp(devname, dumpdev + len) != 0))
171 cp = (struct g_consumer *)dev->si_drv2;
172 error = g_access(cp, 1, 0, 0);
176 error = g_dev_setdumpdev(dev, curthread);
182 (void)g_access(cp, -1, 0, 0);
188 g_dev_destroy(void *arg, int flags __unused)
190 struct g_consumer *cp;
192 struct g_dev_softc *sc;
193 char buf[SPECNAMELEN + 6];
199 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
200 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
201 devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
202 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
203 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
205 g_destroy_consumer(cp);
207 mtx_destroy(&sc->sc_mtx);
217 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
218 printf("%s%s", p, gp->name);
225 g_dev_set_physpath(struct g_consumer *cp)
227 struct g_dev_softc *sc;
229 int error, physpath_len;
231 if (g_access(cp, 1, 0, 0) != 0)
235 physpath_len = MAXPATHLEN;
236 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
237 error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
238 g_access(cp, -1, 0, 0);
239 if (error == 0 && strlen(physpath) != 0) {
240 struct cdev *dev, *old_alias_dev;
241 struct cdev **alias_devp;
244 old_alias_dev = sc->sc_alias;
245 alias_devp = (struct cdev **)&sc->sc_alias;
246 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
247 old_alias_dev, physpath);
248 } else if (sc->sc_alias) {
249 destroy_dev((struct cdev *)sc->sc_alias);
256 g_dev_set_media(struct g_consumer *cp)
258 struct g_dev_softc *sc;
260 char buf[SPECNAMELEN + 6];
264 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
265 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
266 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
269 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
270 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
271 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
276 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
279 if (strcmp(attr, "GEOM::media") == 0) {
284 if (strcmp(attr, "GEOM::physpath") == 0) {
285 g_dev_set_physpath(cp);
291 g_dev_getprovider(struct cdev *dev)
293 struct g_consumer *cp;
298 if (dev->si_devsw != &g_dev_cdevsw)
301 return (cp->provider);
304 static struct g_geom *
305 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
308 struct g_consumer *cp;
309 struct g_dev_softc *sc;
312 char buf[SPECNAMELEN + 6];
314 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
316 gp = g_new_geomf(mp, "%s", pp->name);
317 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
318 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
319 cp = g_new_consumer(gp);
321 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
322 error = g_attach(cp, pp);
324 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
325 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
326 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
328 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
329 __func__, gp->name, error);
331 g_destroy_consumer(cp);
333 mtx_destroy(&sc->sc_mtx);
337 dev->si_flags |= SI_UNMAPPED;
340 dev->si_iosize_max = MAXPHYS;
342 error = init_dumpdev(dev);
344 printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
345 __func__, gp->name, error);
347 g_dev_attrchanged(cp, "GEOM::physpath");
348 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
349 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
355 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
357 struct g_consumer *cp;
358 struct g_dev_softc *sc;
363 return (ENXIO); /* g_dev_taste() not done yet */
364 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
365 cp->geom->name, flags, fmt, td);
367 r = flags & FREAD ? 1 : 0;
368 w = flags & FWRITE ? 1 : 0;
370 e = flags & O_EXCL ? 1 : 0;
376 * This happens on attempt to open a device node with O_EXEC.
383 * When running in very secure mode, do not allow
384 * opens for writing of any disks.
386 error = securelevel_ge(td->td_ucred, 2);
391 error = g_access(cp, r, w, e);
395 mtx_lock(&sc->sc_mtx);
396 if (sc->sc_open == 0 && sc->sc_active != 0)
397 wakeup(&sc->sc_active);
398 sc->sc_open += r + w + e;
399 mtx_unlock(&sc->sc_mtx);
405 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
407 struct g_consumer *cp;
408 struct g_dev_softc *sc;
414 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
415 cp->geom->name, flags, fmt, td);
417 r = flags & FREAD ? -1 : 0;
418 w = flags & FWRITE ? -1 : 0;
420 e = flags & O_EXCL ? -1 : 0;
426 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
427 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
428 * which would result in zero deltas, which in turn would cause
429 * panic in g_access(9).
431 * Note that we cannot zero the counters (ie. do "r = cp->acr"
432 * etc) instead, because the consumer might be opened in another
439 mtx_lock(&sc->sc_mtx);
440 sc->sc_open += r + w + e;
441 while (sc->sc_open == 0 && sc->sc_active != 0)
442 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
443 mtx_unlock(&sc->sc_mtx);
445 error = g_access(cp, r, w, e);
451 * XXX: Until we have unmessed the ioctl situation, there is a race against
452 * XXX: a concurrent orphanization. We cannot close it by holding topology
453 * XXX: since that would prevent us from doing our job, and stalling events
454 * XXX: will break (actually: stall) the BSD disklabel hacks.
457 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
459 struct g_consumer *cp;
460 struct g_provider *pp;
461 off_t offset, length, chunk;
468 KASSERT(cp->acr || cp->acw,
469 ("Consumer with zero access count in g_dev_ioctl"));
471 i = IOCPARM_LEN(cmd);
473 case DIOCGSECTORSIZE:
474 *(u_int *)data = cp->provider->sectorsize;
475 if (*(u_int *)data == 0)
479 *(off_t *)data = cp->provider->mediasize;
480 if (*(off_t *)data == 0)
484 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
485 if (error == 0 && *(u_int *)data == 0)
489 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
490 if (error == 0 && *(u_int *)data == 0)
493 case DIOCGFRONTSTUFF:
494 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
496 case DIOCSKERNELDUMP:
497 if (*(u_int *)data == 0)
498 error = g_dev_setdumpdev(NULL, td);
500 error = g_dev_setdumpdev(dev, td);
503 error = g_io_flush(cp);
506 offset = ((off_t *)data)[0];
507 length = ((off_t *)data)[1];
508 if ((offset % cp->provider->sectorsize) != 0 ||
509 (length % cp->provider->sectorsize) != 0 || length <= 0) {
510 printf("%s: offset=%jd length=%jd\n", __func__, offset,
517 if (g_dev_del_max_sectors != 0 && chunk >
518 g_dev_del_max_sectors * cp->provider->sectorsize) {
519 chunk = g_dev_del_max_sectors *
520 cp->provider->sectorsize;
522 error = g_delete_data(cp, offset, chunk);
528 * Since the request size can be large, the service
529 * time can be is likewise. We make this ioctl
530 * interruptible by checking for signals for each bio.
537 error = g_io_getattr("GEOM::ident", cp, &i, data);
539 case DIOCGPROVIDERNAME:
542 strlcpy(data, pp->name, i);
544 case DIOCGSTRIPESIZE:
545 *(off_t *)data = cp->provider->stripesize;
547 case DIOCGSTRIPEOFFSET:
548 *(off_t *)data = cp->provider->stripeoffset;
551 error = g_io_getattr("GEOM::physpath", cp, &i, data);
552 if (error == 0 && *(char *)data == '\0')
556 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
558 if (arg->len > sizeof(arg->value)) {
562 error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
566 struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
567 struct disk_zone_rep_entry *new_entries, *old_entries;
568 struct disk_zone_report *rep;
576 if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
578 rep = &zone_args->zone_params.report;
579 alloc_size = rep->entries_allocated *
580 sizeof(struct disk_zone_rep_entry);
582 new_entries = g_malloc(alloc_size,
584 old_entries = rep->entries;
585 rep->entries = new_entries;
587 error = g_io_zonecmd(zone_args, cp);
588 if ((zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES)
591 error = copyout(new_entries, old_entries, alloc_size);
593 if ((old_entries != NULL)
595 rep->entries = old_entries;
597 if (new_entries != NULL)
602 if (cp->provider->geom->ioctl != NULL) {
603 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
613 g_dev_done(struct bio *bp2)
615 struct g_consumer *cp;
616 struct g_dev_softc *sc;
622 bp = bp2->bio_parent;
623 bp->bio_error = bp2->bio_error;
624 bp->bio_completed = bp2->bio_completed;
625 bp->bio_resid = bp->bio_length - bp2->bio_completed;
626 if (bp2->bio_cmd == BIO_ZONE)
627 bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
629 if (bp2->bio_error != 0) {
630 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
631 bp2, bp2->bio_error);
632 bp->bio_flags |= BIO_ERROR;
634 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
635 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
639 mtx_lock(&sc->sc_mtx);
640 if ((--sc->sc_active) == 0) {
641 if (sc->sc_open == 0)
642 wakeup(&sc->sc_active);
643 if (sc->sc_dev == NULL)
646 mtx_unlock(&sc->sc_mtx);
648 g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
653 g_dev_strategy(struct bio *bp)
655 struct g_consumer *cp;
658 struct g_dev_softc *sc;
660 KASSERT(bp->bio_cmd == BIO_READ ||
661 bp->bio_cmd == BIO_WRITE ||
662 bp->bio_cmd == BIO_DELETE ||
663 bp->bio_cmd == BIO_FLUSH ||
664 bp->bio_cmd == BIO_ZONE,
665 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
669 KASSERT(cp->acr || cp->acw,
670 ("Consumer with zero access count in g_dev_strategy"));
671 biotrack(bp, __func__);
673 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
674 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
675 bp->bio_resid = bp->bio_bcount;
676 biofinish(bp, NULL, EINVAL);
680 mtx_lock(&sc->sc_mtx);
681 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
683 mtx_unlock(&sc->sc_mtx);
687 * XXX: This is not an ideal solution, but I believe it to
688 * XXX: deadlock safely, all things considered.
690 bp2 = g_clone_bio(bp);
693 pause("gdstrat", hz / 10);
695 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
696 bp2->bio_done = g_dev_done;
698 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
699 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
700 bp2->bio_data, bp2->bio_cmd);
701 g_io_request(bp2, cp);
702 KASSERT(cp->acr || cp->acw,
703 ("g_dev_strategy raced with g_dev_close and lost"));
710 * Called by devfs when asynchronous device destruction is completed.
711 * - Mark that we have no attached device any more.
712 * - If there are no outstanding requests, schedule geom destruction.
713 * Otherwise destruction will be scheduled later by g_dev_done().
717 g_dev_callback(void *arg)
719 struct g_consumer *cp;
720 struct g_dev_softc *sc;
725 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
727 mtx_lock(&sc->sc_mtx);
730 destroy = (sc->sc_active == 0);
731 mtx_unlock(&sc->sc_mtx);
733 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
739 * Called from below when the provider orphaned us.
740 * - Clear any dump settings.
741 * - Request asynchronous device destruction to prevent any more requests
742 * from coming in. The provider is already marked with an error, so
743 * anything which comes in the interim will be returned immediately.
747 g_dev_orphan(struct g_consumer *cp)
750 struct g_dev_softc *sc;
755 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
757 /* Reset any dump-area set on this device */
758 if (dev->si_flags & SI_DUMPDEV)
759 (void)set_dumper(NULL, NULL, curthread);
761 /* Destroy the struct cdev *so we get no more requests */
762 destroy_dev_sched_cb(dev, g_dev_callback, cp);
765 DECLARE_GEOM_CLASS(g_dev_class, g_dev);