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)
130 g_dev_setdumpdev(struct cdev *dev)
132 struct g_kerneldump kd;
133 struct g_consumer *cp;
137 return (set_dumper(NULL, NULL));
143 error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
145 error = set_dumper(&kd.di, devtoname(dev));
147 dev->si_flags |= SI_DUMPDEV;
153 init_dumpdev(struct cdev *dev)
158 if (strcmp(devtoname(dev), dumpdev) != 0)
160 if (g_dev_setdumpdev(dev) == 0) {
167 g_dev_destroy(void *arg, int flags __unused)
169 struct g_consumer *cp;
171 struct g_dev_softc *sc;
177 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
178 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
179 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
181 g_destroy_consumer(cp);
183 mtx_destroy(&sc->sc_mtx);
193 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
194 printf("%s%s", p, gp->name);
201 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
203 struct g_dev_softc *sc;
205 char buf[SPECNAMELEN + 6];
208 if (strcmp(attr, "GEOM::media") == 0) {
210 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
211 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
214 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
215 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf,
221 if (strcmp(attr, "GEOM::physpath") != 0)
224 if (g_access(cp, 1, 0, 0) == 0) {
226 int error, physpath_len;
228 physpath_len = MAXPATHLEN;
229 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
231 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
232 g_access(cp, -1, 0, 0);
233 if (error == 0 && strlen(physpath) != 0) {
234 struct cdev *old_alias_dev;
235 struct cdev **alias_devp;
238 old_alias_dev = sc->sc_alias;
239 alias_devp = (struct cdev **)&sc->sc_alias;
240 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp,
241 dev, old_alias_dev, physpath);
242 } else if (sc->sc_alias) {
243 destroy_dev((struct cdev *)sc->sc_alias);
251 g_dev_getprovider(struct cdev *dev)
253 struct g_consumer *cp;
258 if (dev->si_devsw != &g_dev_cdevsw)
261 return (cp->provider);
264 static struct g_geom *
265 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
268 struct g_consumer *cp;
269 struct g_dev_softc *sc;
271 struct cdev *dev, *adev;
274 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
276 gp = g_new_geomf(mp, "%s", pp->name);
277 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
278 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
279 cp = g_new_consumer(gp);
281 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
282 error = g_attach(cp, pp);
284 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
285 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
286 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
288 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
289 __func__, gp->name, error);
291 g_destroy_consumer(cp);
293 mtx_destroy(&sc->sc_mtx);
297 dev->si_flags |= SI_UNMAPPED;
300 /* Search for device alias name and create it if found. */
302 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
303 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
307 snprintf(buf, sizeof(buf), "%s%s",
308 val, gp->name + len);
310 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
311 &adev, dev, "%s", buf);
312 adev->si_flags |= SI_UNMAPPED;
317 dev->si_iosize_max = MAXPHYS;
321 adev->si_iosize_max = MAXPHYS;
326 g_dev_attrchanged(cp, "GEOM::physpath");
332 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
334 struct g_consumer *cp;
335 struct g_dev_softc *sc;
340 return(ENXIO); /* g_dev_taste() not done yet */
341 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
342 cp->geom->name, flags, fmt, td);
344 r = flags & FREAD ? 1 : 0;
345 w = flags & FWRITE ? 1 : 0;
347 e = flags & O_EXCL ? 1 : 0;
353 * When running in very secure mode, do not allow
354 * opens for writing of any disks.
356 error = securelevel_ge(td->td_ucred, 2);
361 error = g_access(cp, r, w, e);
365 mtx_lock(&sc->sc_mtx);
366 if (sc->sc_open == 0 && sc->sc_active != 0)
367 wakeup(&sc->sc_active);
368 sc->sc_open += r + w + e;
369 mtx_unlock(&sc->sc_mtx);
375 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
377 struct g_consumer *cp;
378 struct g_dev_softc *sc;
384 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
385 cp->geom->name, flags, fmt, td);
387 r = flags & FREAD ? -1 : 0;
388 w = flags & FWRITE ? -1 : 0;
390 e = flags & O_EXCL ? -1 : 0;
395 mtx_lock(&sc->sc_mtx);
396 sc->sc_open += r + w + e;
397 while (sc->sc_open == 0 && sc->sc_active != 0)
398 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
399 mtx_unlock(&sc->sc_mtx);
401 error = g_access(cp, r, w, e);
407 * XXX: Until we have unmessed the ioctl situation, there is a race against
408 * XXX: a concurrent orphanization. We cannot close it by holding topology
409 * XXX: since that would prevent us from doing our job, and stalling events
410 * XXX: will break (actually: stall) the BSD disklabel hacks.
413 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
415 struct g_consumer *cp;
416 struct g_provider *pp;
417 off_t offset, length, chunk;
424 KASSERT(cp->acr || cp->acw,
425 ("Consumer with zero access count in g_dev_ioctl"));
427 i = IOCPARM_LEN(cmd);
429 case DIOCGSECTORSIZE:
430 *(u_int *)data = cp->provider->sectorsize;
431 if (*(u_int *)data == 0)
435 *(off_t *)data = cp->provider->mediasize;
436 if (*(off_t *)data == 0)
440 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
441 if (error == 0 && *(u_int *)data == 0)
445 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
446 if (error == 0 && *(u_int *)data == 0)
449 case DIOCGFRONTSTUFF:
450 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
452 case DIOCSKERNELDUMP:
453 if (*(u_int *)data == 0)
454 error = g_dev_setdumpdev(NULL);
456 error = g_dev_setdumpdev(dev);
459 error = g_io_flush(cp);
462 offset = ((off_t *)data)[0];
463 length = ((off_t *)data)[1];
464 if ((offset % cp->provider->sectorsize) != 0 ||
465 (length % cp->provider->sectorsize) != 0 || length <= 0) {
466 printf("%s: offset=%jd length=%jd\n", __func__, offset,
473 if (g_dev_del_max_sectors != 0 && chunk >
474 g_dev_del_max_sectors * cp->provider->sectorsize) {
475 chunk = g_dev_del_max_sectors *
476 cp->provider->sectorsize;
478 error = g_delete_data(cp, offset, chunk);
484 * Since the request size can be large, the service
485 * time can be is likewise. We make this ioctl
486 * interruptible by checking for signals for each bio.
493 error = g_io_getattr("GEOM::ident", cp, &i, data);
495 case DIOCGPROVIDERNAME:
498 strlcpy(data, pp->name, i);
500 case DIOCGSTRIPESIZE:
501 *(off_t *)data = cp->provider->stripesize;
503 case DIOCGSTRIPEOFFSET:
504 *(off_t *)data = cp->provider->stripeoffset;
507 error = g_io_getattr("GEOM::physpath", cp, &i, data);
508 if (error == 0 && *(char *)data == '\0')
512 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
514 if (arg->len > sizeof(arg->value)) {
518 error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
522 if (cp->provider->geom->ioctl != NULL) {
523 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
533 g_dev_done(struct bio *bp2)
535 struct g_consumer *cp;
536 struct g_dev_softc *sc;
542 bp = bp2->bio_parent;
543 bp->bio_error = bp2->bio_error;
544 bp->bio_completed = bp2->bio_completed;
545 bp->bio_resid = bp->bio_length - bp2->bio_completed;
546 if (bp2->bio_error != 0) {
547 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
548 bp2, bp2->bio_error);
549 bp->bio_flags |= BIO_ERROR;
551 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
552 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
556 mtx_lock(&sc->sc_mtx);
557 if ((--sc->sc_active) == 0) {
558 if (sc->sc_open == 0)
559 wakeup(&sc->sc_active);
560 if (sc->sc_dev == NULL)
563 mtx_unlock(&sc->sc_mtx);
565 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
570 g_dev_strategy(struct bio *bp)
572 struct g_consumer *cp;
575 struct g_dev_softc *sc;
577 KASSERT(bp->bio_cmd == BIO_READ ||
578 bp->bio_cmd == BIO_WRITE ||
579 bp->bio_cmd == BIO_DELETE ||
580 bp->bio_cmd == BIO_FLUSH,
581 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
585 KASSERT(cp->acr || cp->acw,
586 ("Consumer with zero access count in g_dev_strategy"));
588 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
589 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
590 bp->bio_resid = bp->bio_bcount;
591 biofinish(bp, NULL, EINVAL);
595 mtx_lock(&sc->sc_mtx);
596 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
598 mtx_unlock(&sc->sc_mtx);
602 * XXX: This is not an ideal solution, but I belive it to
603 * XXX: deadlock safe, all things considered.
605 bp2 = g_clone_bio(bp);
608 pause("gdstrat", hz / 10);
610 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
611 bp2->bio_done = g_dev_done;
613 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
614 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
615 bp2->bio_data, bp2->bio_cmd);
616 g_io_request(bp2, cp);
617 KASSERT(cp->acr || cp->acw,
618 ("g_dev_strategy raced with g_dev_close and lost"));
625 * Called by devfs when asynchronous device destruction is completed.
626 * - Mark that we have no attached device any more.
627 * - If there are no outstanding requests, schedule geom destruction.
628 * Otherwise destruction will be scheduled later by g_dev_done().
632 g_dev_callback(void *arg)
634 struct g_consumer *cp;
635 struct g_dev_softc *sc;
640 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
642 mtx_lock(&sc->sc_mtx);
645 destroy = (sc->sc_active == 0);
646 mtx_unlock(&sc->sc_mtx);
648 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
654 * Called from below when the provider orphaned us.
655 * - Clear any dump settings.
656 * - Request asynchronous device destruction to prevent any more requests
657 * from coming in. The provider is already marked with an error, so
658 * anything which comes in in the interrim will be returned immediately.
662 g_dev_orphan(struct g_consumer *cp)
665 struct g_dev_softc *sc;
670 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
672 /* Reset any dump-area set on this device */
673 if (dev->si_flags & SI_DUMPDEV)
674 set_dumper(NULL, NULL);
676 /* Destroy the struct cdev *so we get no more requests */
677 destroy_dev_sched_cb(dev, g_dev_callback, cp);
680 DECLARE_GEOM_CLASS(g_dev_class, g_dev);