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 = 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 const char *devprefix = "/dev/", *devname;
161 len = strlen(devprefix);
162 devname = devtoname(dev);
163 if (strcmp(devname, dumpdev) != 0 &&
164 (strncmp(dumpdev, devprefix, len) != 0 ||
165 strcmp(devname, dumpdev + len) != 0))
167 if (g_dev_setdumpdev(dev, curthread) == 0) {
174 g_dev_destroy(void *arg, int flags __unused)
176 struct g_consumer *cp;
178 struct g_dev_softc *sc;
179 char buf[SPECNAMELEN + 6];
185 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
186 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
187 devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
188 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
189 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
191 g_destroy_consumer(cp);
193 mtx_destroy(&sc->sc_mtx);
203 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
204 printf("%s%s", p, gp->name);
211 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
213 struct g_dev_softc *sc;
215 char buf[SPECNAMELEN + 6];
218 if (strcmp(attr, "GEOM::media") == 0) {
220 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
221 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
222 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
225 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
226 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf,
228 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf,
234 if (strcmp(attr, "GEOM::physpath") != 0)
237 if (g_access(cp, 1, 0, 0) == 0) {
239 int error, physpath_len;
241 physpath_len = MAXPATHLEN;
242 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
244 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
245 g_access(cp, -1, 0, 0);
246 if (error == 0 && strlen(physpath) != 0) {
247 struct cdev *old_alias_dev;
248 struct cdev **alias_devp;
251 old_alias_dev = sc->sc_alias;
252 alias_devp = (struct cdev **)&sc->sc_alias;
253 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp,
254 dev, old_alias_dev, physpath);
255 } else if (sc->sc_alias) {
256 destroy_dev((struct cdev *)sc->sc_alias);
264 g_dev_getprovider(struct cdev *dev)
266 struct g_consumer *cp;
271 if (dev->si_devsw != &g_dev_cdevsw)
274 return (cp->provider);
277 static struct g_geom *
278 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
281 struct g_consumer *cp;
282 struct g_dev_softc *sc;
284 struct cdev *dev, *adev;
285 char buf[SPECNAMELEN + 6], *val;
287 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
289 gp = g_new_geomf(mp, "%s", pp->name);
290 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
291 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
292 cp = g_new_consumer(gp);
294 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
295 error = g_attach(cp, pp);
297 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
298 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
299 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
301 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
302 __func__, gp->name, error);
304 g_destroy_consumer(cp);
306 mtx_destroy(&sc->sc_mtx);
310 dev->si_flags |= SI_UNMAPPED;
313 /* Search for device alias name and create it if found. */
315 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
316 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
320 snprintf(buf, sizeof(buf), "%s%s",
321 val, gp->name + len);
323 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
324 &adev, dev, "%s", buf);
325 adev->si_flags |= SI_UNMAPPED;
330 dev->si_iosize_max = MAXPHYS;
334 adev->si_iosize_max = MAXPHYS;
339 g_dev_attrchanged(cp, "GEOM::physpath");
340 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
341 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
347 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
349 struct g_consumer *cp;
350 struct g_dev_softc *sc;
355 return(ENXIO); /* g_dev_taste() not done yet */
356 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
357 cp->geom->name, flags, fmt, td);
359 r = flags & FREAD ? 1 : 0;
360 w = flags & FWRITE ? 1 : 0;
362 e = flags & O_EXCL ? 1 : 0;
368 * This happens on attempt to open a device node with O_EXEC.
375 * When running in very secure mode, do not allow
376 * opens for writing of any disks.
378 error = securelevel_ge(td->td_ucred, 2);
383 error = g_access(cp, r, w, e);
387 mtx_lock(&sc->sc_mtx);
388 if (sc->sc_open == 0 && sc->sc_active != 0)
389 wakeup(&sc->sc_active);
390 sc->sc_open += r + w + e;
391 mtx_unlock(&sc->sc_mtx);
397 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
399 struct g_consumer *cp;
400 struct g_dev_softc *sc;
406 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
407 cp->geom->name, flags, fmt, td);
409 r = flags & FREAD ? -1 : 0;
410 w = flags & FWRITE ? -1 : 0;
412 e = flags & O_EXCL ? -1 : 0;
418 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
419 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
420 * which would result in zero deltas, which in turn would cause
421 * panic in g_access(9).
423 * Note that we cannot zero the counters (ie. do "r = cp->acr"
424 * etc) instead, because the consumer might be opened in another
431 mtx_lock(&sc->sc_mtx);
432 sc->sc_open += r + w + e;
433 while (sc->sc_open == 0 && sc->sc_active != 0)
434 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
435 mtx_unlock(&sc->sc_mtx);
437 error = g_access(cp, r, w, e);
443 * XXX: Until we have unmessed the ioctl situation, there is a race against
444 * XXX: a concurrent orphanization. We cannot close it by holding topology
445 * XXX: since that would prevent us from doing our job, and stalling events
446 * XXX: will break (actually: stall) the BSD disklabel hacks.
449 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
451 struct g_consumer *cp;
452 struct g_provider *pp;
453 off_t offset, length, chunk;
460 KASSERT(cp->acr || cp->acw,
461 ("Consumer with zero access count in g_dev_ioctl"));
463 i = IOCPARM_LEN(cmd);
465 case DIOCGSECTORSIZE:
466 *(u_int *)data = cp->provider->sectorsize;
467 if (*(u_int *)data == 0)
471 *(off_t *)data = cp->provider->mediasize;
472 if (*(off_t *)data == 0)
476 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
477 if (error == 0 && *(u_int *)data == 0)
481 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
482 if (error == 0 && *(u_int *)data == 0)
485 case DIOCGFRONTSTUFF:
486 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
488 case DIOCSKERNELDUMP:
489 if (*(u_int *)data == 0)
490 error = g_dev_setdumpdev(NULL, td);
492 error = g_dev_setdumpdev(dev, td);
495 error = g_io_flush(cp);
498 offset = ((off_t *)data)[0];
499 length = ((off_t *)data)[1];
500 if ((offset % cp->provider->sectorsize) != 0 ||
501 (length % cp->provider->sectorsize) != 0 || length <= 0) {
502 printf("%s: offset=%jd length=%jd\n", __func__, offset,
509 if (g_dev_del_max_sectors != 0 && chunk >
510 g_dev_del_max_sectors * cp->provider->sectorsize) {
511 chunk = g_dev_del_max_sectors *
512 cp->provider->sectorsize;
514 error = g_delete_data(cp, offset, chunk);
520 * Since the request size can be large, the service
521 * time can be is likewise. We make this ioctl
522 * interruptible by checking for signals for each bio.
529 error = g_io_getattr("GEOM::ident", cp, &i, data);
531 case DIOCGPROVIDERNAME:
534 strlcpy(data, pp->name, i);
536 case DIOCGSTRIPESIZE:
537 *(off_t *)data = cp->provider->stripesize;
539 case DIOCGSTRIPEOFFSET:
540 *(off_t *)data = cp->provider->stripeoffset;
543 error = g_io_getattr("GEOM::physpath", cp, &i, data);
544 if (error == 0 && *(char *)data == '\0')
548 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
550 if (arg->len > sizeof(arg->value)) {
554 error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
558 if (cp->provider->geom->ioctl != NULL) {
559 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
569 g_dev_done(struct bio *bp2)
571 struct g_consumer *cp;
572 struct g_dev_softc *sc;
578 bp = bp2->bio_parent;
579 bp->bio_error = bp2->bio_error;
580 bp->bio_completed = bp2->bio_completed;
581 bp->bio_resid = bp->bio_length - bp2->bio_completed;
582 if (bp2->bio_error != 0) {
583 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
584 bp2, bp2->bio_error);
585 bp->bio_flags |= BIO_ERROR;
587 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
588 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
592 mtx_lock(&sc->sc_mtx);
593 if ((--sc->sc_active) == 0) {
594 if (sc->sc_open == 0)
595 wakeup(&sc->sc_active);
596 if (sc->sc_dev == NULL)
599 mtx_unlock(&sc->sc_mtx);
601 g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
606 g_dev_strategy(struct bio *bp)
608 struct g_consumer *cp;
611 struct g_dev_softc *sc;
613 KASSERT(bp->bio_cmd == BIO_READ ||
614 bp->bio_cmd == BIO_WRITE ||
615 bp->bio_cmd == BIO_DELETE ||
616 bp->bio_cmd == BIO_FLUSH,
617 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
621 KASSERT(cp->acr || cp->acw,
622 ("Consumer with zero access count in g_dev_strategy"));
624 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
625 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
626 bp->bio_resid = bp->bio_bcount;
627 biofinish(bp, NULL, EINVAL);
631 mtx_lock(&sc->sc_mtx);
632 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
634 mtx_unlock(&sc->sc_mtx);
638 * XXX: This is not an ideal solution, but I belive it to
639 * XXX: deadlock safe, all things considered.
641 bp2 = g_clone_bio(bp);
644 pause("gdstrat", hz / 10);
646 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
647 bp2->bio_done = g_dev_done;
649 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
650 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
651 bp2->bio_data, bp2->bio_cmd);
652 g_io_request(bp2, cp);
653 KASSERT(cp->acr || cp->acw,
654 ("g_dev_strategy raced with g_dev_close and lost"));
661 * Called by devfs when asynchronous device destruction is completed.
662 * - Mark that we have no attached device any more.
663 * - If there are no outstanding requests, schedule geom destruction.
664 * Otherwise destruction will be scheduled later by g_dev_done().
668 g_dev_callback(void *arg)
670 struct g_consumer *cp;
671 struct g_dev_softc *sc;
676 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
678 mtx_lock(&sc->sc_mtx);
681 destroy = (sc->sc_active == 0);
682 mtx_unlock(&sc->sc_mtx);
684 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
690 * Called from below when the provider orphaned us.
691 * - Clear any dump settings.
692 * - Request asynchronous device destruction to prevent any more requests
693 * from coming in. The provider is already marked with an error, so
694 * anything which comes in in the interrim will be returned immediately.
698 g_dev_orphan(struct g_consumer *cp)
701 struct g_dev_softc *sc;
706 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
708 /* Reset any dump-area set on this device */
709 if (dev->si_flags & SI_DUMPDEV)
710 (void)set_dumper(NULL, NULL, curthread);
712 /* Destroy the struct cdev *so we get no more requests */
713 destroy_dev_sched_cb(dev, g_dev_callback, cp);
716 DECLARE_GEOM_CLASS(g_dev_class, g_dev);