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;
172 char buf[SPECNAMELEN + 6];
178 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
179 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
180 devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
181 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
182 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
184 g_destroy_consumer(cp);
186 mtx_destroy(&sc->sc_mtx);
196 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
197 printf("%s%s", p, gp->name);
204 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
206 struct g_dev_softc *sc;
208 char buf[SPECNAMELEN + 6];
211 if (strcmp(attr, "GEOM::media") == 0) {
213 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
214 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
215 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
218 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
219 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf,
221 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf,
227 if (strcmp(attr, "GEOM::physpath") != 0)
230 if (g_access(cp, 1, 0, 0) == 0) {
232 int error, physpath_len;
234 physpath_len = MAXPATHLEN;
235 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
237 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 *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,
247 dev, old_alias_dev, physpath);
248 } else if (sc->sc_alias) {
249 destroy_dev((struct cdev *)sc->sc_alias);
257 g_dev_getprovider(struct cdev *dev)
259 struct g_consumer *cp;
264 if (dev->si_devsw != &g_dev_cdevsw)
267 return (cp->provider);
270 static struct g_geom *
271 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
274 struct g_consumer *cp;
275 struct g_dev_softc *sc;
277 struct cdev *dev, *adev;
278 char buf[SPECNAMELEN + 6], *val;
280 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
282 gp = g_new_geomf(mp, "%s", pp->name);
283 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
284 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
285 cp = g_new_consumer(gp);
287 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
288 error = g_attach(cp, pp);
290 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
291 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
292 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
294 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
295 __func__, gp->name, error);
297 g_destroy_consumer(cp);
299 mtx_destroy(&sc->sc_mtx);
303 dev->si_flags |= SI_UNMAPPED;
306 /* Search for device alias name and create it if found. */
308 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
309 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
313 snprintf(buf, sizeof(buf), "%s%s",
314 val, gp->name + len);
316 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
317 &adev, dev, "%s", buf);
318 adev->si_flags |= SI_UNMAPPED;
323 dev->si_iosize_max = MAXPHYS;
327 adev->si_iosize_max = MAXPHYS;
332 g_dev_attrchanged(cp, "GEOM::physpath");
333 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
334 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
340 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
342 struct g_consumer *cp;
343 struct g_dev_softc *sc;
348 return(ENXIO); /* g_dev_taste() not done yet */
349 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
350 cp->geom->name, flags, fmt, td);
352 r = flags & FREAD ? 1 : 0;
353 w = flags & FWRITE ? 1 : 0;
355 e = flags & O_EXCL ? 1 : 0;
361 * When running in very secure mode, do not allow
362 * opens for writing of any disks.
364 error = securelevel_ge(td->td_ucred, 2);
369 error = g_access(cp, r, w, e);
373 mtx_lock(&sc->sc_mtx);
374 if (sc->sc_open == 0 && sc->sc_active != 0)
375 wakeup(&sc->sc_active);
376 sc->sc_open += r + w + e;
377 mtx_unlock(&sc->sc_mtx);
383 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
385 struct g_consumer *cp;
386 struct g_dev_softc *sc;
392 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
393 cp->geom->name, flags, fmt, td);
395 r = flags & FREAD ? -1 : 0;
396 w = flags & FWRITE ? -1 : 0;
398 e = flags & O_EXCL ? -1 : 0;
404 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
405 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
406 * which would result in zero deltas, which in turn would cause
407 * panic in g_access(9).
409 * Note that we cannot zero the counters (ie. do "r = cp->acr"
410 * etc) instead, because the consumer might be opened in another
417 mtx_lock(&sc->sc_mtx);
418 sc->sc_open += r + w + e;
419 while (sc->sc_open == 0 && sc->sc_active != 0)
420 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
421 mtx_unlock(&sc->sc_mtx);
423 error = g_access(cp, r, w, e);
429 * XXX: Until we have unmessed the ioctl situation, there is a race against
430 * XXX: a concurrent orphanization. We cannot close it by holding topology
431 * XXX: since that would prevent us from doing our job, and stalling events
432 * XXX: will break (actually: stall) the BSD disklabel hacks.
435 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
437 struct g_consumer *cp;
438 struct g_provider *pp;
439 off_t offset, length, chunk;
446 KASSERT(cp->acr || cp->acw,
447 ("Consumer with zero access count in g_dev_ioctl"));
449 i = IOCPARM_LEN(cmd);
451 case DIOCGSECTORSIZE:
452 *(u_int *)data = cp->provider->sectorsize;
453 if (*(u_int *)data == 0)
457 *(off_t *)data = cp->provider->mediasize;
458 if (*(off_t *)data == 0)
462 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
463 if (error == 0 && *(u_int *)data == 0)
467 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
468 if (error == 0 && *(u_int *)data == 0)
471 case DIOCGFRONTSTUFF:
472 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
474 case DIOCSKERNELDUMP:
475 if (*(u_int *)data == 0)
476 error = g_dev_setdumpdev(NULL);
478 error = g_dev_setdumpdev(dev);
481 error = g_io_flush(cp);
484 offset = ((off_t *)data)[0];
485 length = ((off_t *)data)[1];
486 if ((offset % cp->provider->sectorsize) != 0 ||
487 (length % cp->provider->sectorsize) != 0 || length <= 0) {
488 printf("%s: offset=%jd length=%jd\n", __func__, offset,
495 if (g_dev_del_max_sectors != 0 && chunk >
496 g_dev_del_max_sectors * cp->provider->sectorsize) {
497 chunk = g_dev_del_max_sectors *
498 cp->provider->sectorsize;
500 error = g_delete_data(cp, offset, chunk);
506 * Since the request size can be large, the service
507 * time can be is likewise. We make this ioctl
508 * interruptible by checking for signals for each bio.
515 error = g_io_getattr("GEOM::ident", cp, &i, data);
517 case DIOCGPROVIDERNAME:
520 strlcpy(data, pp->name, i);
522 case DIOCGSTRIPESIZE:
523 *(off_t *)data = cp->provider->stripesize;
525 case DIOCGSTRIPEOFFSET:
526 *(off_t *)data = cp->provider->stripeoffset;
529 error = g_io_getattr("GEOM::physpath", cp, &i, data);
530 if (error == 0 && *(char *)data == '\0')
534 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
536 if (arg->len > sizeof(arg->value)) {
540 error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
544 if (cp->provider->geom->ioctl != NULL) {
545 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
555 g_dev_done(struct bio *bp2)
557 struct g_consumer *cp;
558 struct g_dev_softc *sc;
564 bp = bp2->bio_parent;
565 bp->bio_error = bp2->bio_error;
566 bp->bio_completed = bp2->bio_completed;
567 bp->bio_resid = bp->bio_length - bp2->bio_completed;
568 if (bp2->bio_error != 0) {
569 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
570 bp2, bp2->bio_error);
571 bp->bio_flags |= BIO_ERROR;
573 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
574 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
578 mtx_lock(&sc->sc_mtx);
579 if ((--sc->sc_active) == 0) {
580 if (sc->sc_open == 0)
581 wakeup(&sc->sc_active);
582 if (sc->sc_dev == NULL)
585 mtx_unlock(&sc->sc_mtx);
587 g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
592 g_dev_strategy(struct bio *bp)
594 struct g_consumer *cp;
597 struct g_dev_softc *sc;
599 KASSERT(bp->bio_cmd == BIO_READ ||
600 bp->bio_cmd == BIO_WRITE ||
601 bp->bio_cmd == BIO_DELETE ||
602 bp->bio_cmd == BIO_FLUSH,
603 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
607 KASSERT(cp->acr || cp->acw,
608 ("Consumer with zero access count in g_dev_strategy"));
610 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
611 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
612 bp->bio_resid = bp->bio_bcount;
613 biofinish(bp, NULL, EINVAL);
617 mtx_lock(&sc->sc_mtx);
618 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
620 mtx_unlock(&sc->sc_mtx);
624 * XXX: This is not an ideal solution, but I belive it to
625 * XXX: deadlock safe, all things considered.
627 bp2 = g_clone_bio(bp);
630 pause("gdstrat", hz / 10);
632 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
633 bp2->bio_done = g_dev_done;
635 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
636 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
637 bp2->bio_data, bp2->bio_cmd);
638 g_io_request(bp2, cp);
639 KASSERT(cp->acr || cp->acw,
640 ("g_dev_strategy raced with g_dev_close and lost"));
647 * Called by devfs when asynchronous device destruction is completed.
648 * - Mark that we have no attached device any more.
649 * - If there are no outstanding requests, schedule geom destruction.
650 * Otherwise destruction will be scheduled later by g_dev_done().
654 g_dev_callback(void *arg)
656 struct g_consumer *cp;
657 struct g_dev_softc *sc;
662 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
664 mtx_lock(&sc->sc_mtx);
667 destroy = (sc->sc_active == 0);
668 mtx_unlock(&sc->sc_mtx);
670 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
676 * Called from below when the provider orphaned us.
677 * - Clear any dump settings.
678 * - Request asynchronous device destruction to prevent any more requests
679 * from coming in. The provider is already marked with an error, so
680 * anything which comes in in the interrim will be returned immediately.
684 g_dev_orphan(struct g_consumer *cp)
687 struct g_dev_softc *sc;
692 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
694 /* Reset any dump-area set on this device */
695 if (dev->si_flags & SI_DUMPDEV)
696 set_dumper(NULL, NULL);
698 /* Destroy the struct cdev *so we get no more requests */
699 destroy_dev_sched_cb(dev, g_dev_callback, cp);
702 DECLARE_GEOM_CLASS(g_dev_class, g_dev);