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_taste_t g_dev_taste;
86 static g_orphan_t g_dev_orphan;
87 static g_attrchanged_t g_dev_attrchanged;
89 static struct g_class g_dev_class = {
93 .orphan = g_dev_orphan,
94 .attrchanged = g_dev_attrchanged
98 * We target 262144 (8 x 32768) sectors by default as this significantly
99 * increases the throughput on commonly used SSD's with a marginal
100 * increase in non-interruptible request latency.
102 static uint64_t g_dev_del_max_sectors = 262144;
103 SYSCTL_DECL(_kern_geom);
104 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
105 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
106 &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
107 "delete request sent to the provider. Larger requests are chunked "
108 "so they can be interrupted. (0 = disable chunking)");
111 g_dev_destroy(void *arg, int flags __unused)
113 struct g_consumer *cp;
115 struct g_dev_softc *sc;
121 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
122 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
123 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
125 g_destroy_consumer(cp);
127 mtx_destroy(&sc->sc_mtx);
137 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
138 printf("%s%s", p, gp->name);
145 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
147 struct g_dev_softc *sc;
149 char buf[SPECNAMELEN + 6];
152 if (strcmp(attr, "GEOM::media") == 0) {
154 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
155 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
158 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
159 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf,
165 if (strcmp(attr, "GEOM::physpath") != 0)
168 if (g_access(cp, 1, 0, 0) == 0) {
170 int error, physpath_len;
172 physpath_len = MAXPATHLEN;
173 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
175 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
176 g_access(cp, -1, 0, 0);
177 if (error == 0 && strlen(physpath) != 0) {
178 struct cdev *old_alias_dev;
179 struct cdev **alias_devp;
182 old_alias_dev = sc->sc_alias;
183 alias_devp = (struct cdev **)&sc->sc_alias;
184 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp,
185 dev, old_alias_dev, physpath);
186 } else if (sc->sc_alias) {
187 destroy_dev((struct cdev *)sc->sc_alias);
195 g_dev_getprovider(struct cdev *dev)
197 struct g_consumer *cp;
202 if (dev->si_devsw != &g_dev_cdevsw)
205 return (cp->provider);
208 static struct g_geom *
209 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
212 struct g_consumer *cp;
213 struct g_dev_softc *sc;
215 struct cdev *dev, *adev;
218 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
220 gp = g_new_geomf(mp, "%s", pp->name);
221 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
222 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
223 cp = g_new_consumer(gp);
225 error = g_attach(cp, pp);
227 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
228 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
229 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
231 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
232 __func__, gp->name, error);
234 g_destroy_consumer(cp);
236 mtx_destroy(&sc->sc_mtx);
240 dev->si_flags |= SI_UNMAPPED;
243 /* Search for device alias name and create it if found. */
245 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
246 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
250 snprintf(buf, sizeof(buf), "%s%s",
251 val, gp->name + len);
253 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
254 &adev, dev, "%s", buf);
255 adev->si_flags |= SI_UNMAPPED;
260 dev->si_iosize_max = MAXPHYS;
263 adev->si_iosize_max = MAXPHYS;
267 g_dev_attrchanged(cp, "GEOM::physpath");
273 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
275 struct g_consumer *cp;
276 struct g_dev_softc *sc;
281 return(ENXIO); /* g_dev_taste() not done yet */
282 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
283 cp->geom->name, flags, fmt, td);
285 r = flags & FREAD ? 1 : 0;
286 w = flags & FWRITE ? 1 : 0;
288 e = flags & O_EXCL ? 1 : 0;
294 * When running in very secure mode, do not allow
295 * opens for writing of any disks.
297 error = securelevel_ge(td->td_ucred, 2);
302 error = g_access(cp, r, w, e);
306 mtx_lock(&sc->sc_mtx);
307 if (sc->sc_open == 0 && sc->sc_active != 0)
308 wakeup(&sc->sc_active);
309 sc->sc_open += r + w + e;
310 mtx_unlock(&sc->sc_mtx);
316 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
318 struct g_consumer *cp;
319 struct g_dev_softc *sc;
325 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
326 cp->geom->name, flags, fmt, td);
328 r = flags & FREAD ? -1 : 0;
329 w = flags & FWRITE ? -1 : 0;
331 e = flags & O_EXCL ? -1 : 0;
336 mtx_lock(&sc->sc_mtx);
337 sc->sc_open += r + w + e;
338 while (sc->sc_open == 0 && sc->sc_active != 0)
339 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
340 mtx_unlock(&sc->sc_mtx);
342 error = g_access(cp, r, w, e);
348 * XXX: Until we have unmessed the ioctl situation, there is a race against
349 * XXX: a concurrent orphanization. We cannot close it by holding topology
350 * XXX: since that would prevent us from doing our job, and stalling events
351 * XXX: will break (actually: stall) the BSD disklabel hacks.
354 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
356 struct g_consumer *cp;
357 struct g_provider *pp;
358 struct g_kerneldump kd;
359 off_t offset, length, chunk;
367 KASSERT(cp->acr || cp->acw,
368 ("Consumer with zero access count in g_dev_ioctl"));
370 i = IOCPARM_LEN(cmd);
372 case DIOCGSECTORSIZE:
373 *(u_int *)data = cp->provider->sectorsize;
374 if (*(u_int *)data == 0)
378 *(off_t *)data = cp->provider->mediasize;
379 if (*(off_t *)data == 0)
383 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
384 if (error == 0 && *(u_int *)data == 0)
388 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
389 if (error == 0 && *(u_int *)data == 0)
392 case DIOCGFRONTSTUFF:
393 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
395 case DIOCSKERNELDUMP:
396 u = *((u_int *)data);
398 set_dumper(NULL, NULL);
405 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
407 error = set_dumper(&kd.di, devtoname(dev));
409 dev->si_flags |= SI_DUMPDEV;
413 error = g_io_flush(cp);
416 offset = ((off_t *)data)[0];
417 length = ((off_t *)data)[1];
418 if ((offset % cp->provider->sectorsize) != 0 ||
419 (length % cp->provider->sectorsize) != 0 || length <= 0) {
420 printf("%s: offset=%jd length=%jd\n", __func__, offset,
427 if (g_dev_del_max_sectors != 0 && chunk >
428 g_dev_del_max_sectors * cp->provider->sectorsize) {
429 chunk = g_dev_del_max_sectors *
430 cp->provider->sectorsize;
432 error = g_delete_data(cp, offset, chunk);
438 * Since the request size can be large, the service
439 * time can be is likewise. We make this ioctl
440 * interruptible by checking for signals for each bio.
447 error = g_io_getattr("GEOM::ident", cp, &i, data);
449 case DIOCGPROVIDERNAME:
452 strlcpy(data, pp->name, i);
454 case DIOCGSTRIPESIZE:
455 *(off_t *)data = cp->provider->stripesize;
457 case DIOCGSTRIPEOFFSET:
458 *(off_t *)data = cp->provider->stripeoffset;
461 error = g_io_getattr("GEOM::physpath", cp, &i, data);
462 if (error == 0 && *(char *)data == '\0')
466 if (cp->provider->geom->ioctl != NULL) {
467 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
477 g_dev_done(struct bio *bp2)
479 struct g_consumer *cp;
480 struct g_dev_softc *sc;
486 bp = bp2->bio_parent;
487 bp->bio_error = bp2->bio_error;
488 if (bp->bio_error != 0) {
489 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
491 bp->bio_flags |= BIO_ERROR;
493 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
494 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed);
496 bp->bio_resid = bp->bio_length - bp2->bio_completed;
497 bp->bio_completed = bp2->bio_completed;
500 mtx_lock(&sc->sc_mtx);
501 if ((--sc->sc_active) == 0) {
502 if (sc->sc_open == 0)
503 wakeup(&sc->sc_active);
504 if (sc->sc_dev == NULL)
507 mtx_unlock(&sc->sc_mtx);
509 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
514 g_dev_strategy(struct bio *bp)
516 struct g_consumer *cp;
519 struct g_dev_softc *sc;
521 KASSERT(bp->bio_cmd == BIO_READ ||
522 bp->bio_cmd == BIO_WRITE ||
523 bp->bio_cmd == BIO_DELETE ||
524 bp->bio_cmd == BIO_FLUSH,
525 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
529 KASSERT(cp->acr || cp->acw,
530 ("Consumer with zero access count in g_dev_strategy"));
532 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
533 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
534 bp->bio_resid = bp->bio_bcount;
535 biofinish(bp, NULL, EINVAL);
539 mtx_lock(&sc->sc_mtx);
540 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
542 mtx_unlock(&sc->sc_mtx);
546 * XXX: This is not an ideal solution, but I belive it to
547 * XXX: deadlock safe, all things considered.
549 bp2 = g_clone_bio(bp);
552 pause("gdstrat", hz / 10);
554 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
555 bp2->bio_done = g_dev_done;
557 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
558 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
559 bp2->bio_data, bp2->bio_cmd);
560 g_io_request(bp2, cp);
561 KASSERT(cp->acr || cp->acw,
562 ("g_dev_strategy raced with g_dev_close and lost"));
569 * Called by devfs when asynchronous device destruction is completed.
570 * - Mark that we have no attached device any more.
571 * - If there are no outstanding requests, schedule geom destruction.
572 * Otherwise destruction will be scheduled later by g_dev_done().
576 g_dev_callback(void *arg)
578 struct g_consumer *cp;
579 struct g_dev_softc *sc;
584 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
586 mtx_lock(&sc->sc_mtx);
589 destroy = (sc->sc_active == 0);
590 mtx_unlock(&sc->sc_mtx);
592 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
598 * Called from below when the provider orphaned us.
599 * - Clear any dump settings.
600 * - Request asynchronous device destruction to prevent any more requests
601 * from coming in. The provider is already marked with an error, so
602 * anything which comes in in the interrim will be returned immediately.
606 g_dev_orphan(struct g_consumer *cp)
609 struct g_dev_softc *sc;
614 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
616 /* Reset any dump-area set on this device */
617 if (dev->si_flags & SI_DUMPDEV)
618 set_dumper(NULL, NULL);
620 /* Destroy the struct cdev *so we get no more requests */
621 destroy_dev_sched_cb(dev, g_dev_callback, cp);
624 DECLARE_GEOM_CLASS(g_dev_class, g_dev);