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;
66 #define SC_A_DESTROY (1 << 31)
67 #define SC_A_OPEN (1 << 30)
68 #define SC_A_ACTIVE (SC_A_OPEN - 1)
71 static d_open_t g_dev_open;
72 static d_close_t g_dev_close;
73 static d_strategy_t g_dev_strategy;
74 static d_ioctl_t g_dev_ioctl;
76 static struct cdevsw g_dev_cdevsw = {
77 .d_version = D_VERSION,
79 .d_close = g_dev_close,
82 .d_ioctl = g_dev_ioctl,
83 .d_strategy = g_dev_strategy,
85 .d_flags = D_DISK | D_TRACKCLOSE,
88 static g_init_t g_dev_init;
89 static g_fini_t g_dev_fini;
90 static g_taste_t g_dev_taste;
91 static g_orphan_t g_dev_orphan;
92 static g_attrchanged_t g_dev_attrchanged;
94 static struct g_class g_dev_class = {
100 .orphan = g_dev_orphan,
101 .attrchanged = g_dev_attrchanged
105 * We target 262144 (8 x 32768) sectors by default as this significantly
106 * increases the throughput on commonly used SSD's with a marginal
107 * increase in non-interruptible request latency.
109 static uint64_t g_dev_del_max_sectors = 262144;
110 SYSCTL_DECL(_kern_geom);
111 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
112 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
113 &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
114 "delete request sent to the provider. Larger requests are chunked "
115 "so they can be interrupted. (0 = disable chunking)");
117 static char *dumpdev = NULL;
119 g_dev_init(struct g_class *mp)
122 dumpdev = kern_getenv("dumpdev");
126 g_dev_fini(struct g_class *mp)
134 g_dev_setdumpdev(struct cdev *dev, struct thread *td)
136 struct g_kerneldump kd;
137 struct g_consumer *cp;
141 return (set_dumper(NULL, NULL, td));
147 error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
149 error = set_dumper(&kd.di, devtoname(dev), td);
151 dev->si_flags |= SI_DUMPDEV;
157 init_dumpdev(struct cdev *dev)
159 struct g_consumer *cp;
160 const char *devprefix = "/dev/", *devname;
167 len = strlen(devprefix);
168 devname = devtoname(dev);
169 if (strcmp(devname, dumpdev) != 0 &&
170 (strncmp(dumpdev, devprefix, len) != 0 ||
171 strcmp(devname, dumpdev + len) != 0))
174 cp = (struct g_consumer *)dev->si_drv2;
175 error = g_access(cp, 1, 0, 0);
179 error = g_dev_setdumpdev(dev, curthread);
185 (void)g_access(cp, -1, 0, 0);
191 g_dev_destroy(void *arg, int flags __unused)
193 struct g_consumer *cp;
195 struct g_dev_softc *sc;
196 char buf[SPECNAMELEN + 6];
202 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
203 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
204 devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
205 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
206 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
208 g_destroy_consumer(cp);
210 mtx_destroy(&sc->sc_mtx);
220 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
221 printf("%s%s", p, gp->name);
228 g_dev_set_physpath(struct g_consumer *cp)
230 struct g_dev_softc *sc;
232 int error, physpath_len;
234 if (g_access(cp, 1, 0, 0) != 0)
238 physpath_len = MAXPATHLEN;
239 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
240 error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
241 g_access(cp, -1, 0, 0);
242 if (error == 0 && strlen(physpath) != 0) {
243 struct cdev *dev, *old_alias_dev;
244 struct cdev **alias_devp;
247 old_alias_dev = sc->sc_alias;
248 alias_devp = (struct cdev **)&sc->sc_alias;
249 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
250 old_alias_dev, physpath);
251 } else if (sc->sc_alias) {
252 destroy_dev((struct cdev *)sc->sc_alias);
259 g_dev_set_media(struct g_consumer *cp)
261 struct g_dev_softc *sc;
263 char buf[SPECNAMELEN + 6];
267 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
268 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
269 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
272 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
273 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
274 devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
279 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
282 if (strcmp(attr, "GEOM::media") == 0) {
287 if (strcmp(attr, "GEOM::physpath") == 0) {
288 g_dev_set_physpath(cp);
294 g_dev_getprovider(struct cdev *dev)
296 struct g_consumer *cp;
301 if (dev->si_devsw != &g_dev_cdevsw)
304 return (cp->provider);
307 static struct g_geom *
308 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
311 struct g_consumer *cp;
312 struct g_dev_softc *sc;
315 char buf[SPECNAMELEN + 6];
317 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
319 gp = g_new_geomf(mp, "%s", pp->name);
320 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
321 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
322 cp = g_new_consumer(gp);
324 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
325 error = g_attach(cp, pp);
327 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
328 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
329 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
331 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
332 __func__, gp->name, error);
334 g_destroy_consumer(cp);
336 mtx_destroy(&sc->sc_mtx);
340 dev->si_flags |= SI_UNMAPPED;
343 dev->si_iosize_max = MAXPHYS;
345 error = init_dumpdev(dev);
347 printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
348 __func__, gp->name, error);
350 g_dev_attrchanged(cp, "GEOM::physpath");
351 snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
352 devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
358 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
360 struct g_consumer *cp;
361 struct g_dev_softc *sc;
366 return (ENXIO); /* g_dev_taste() not done yet */
367 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
368 cp->geom->name, flags, fmt, td);
370 r = flags & FREAD ? 1 : 0;
371 w = flags & FWRITE ? 1 : 0;
373 e = flags & O_EXCL ? 1 : 0;
379 * This happens on attempt to open a device node with O_EXEC.
386 * When running in very secure mode, do not allow
387 * opens for writing of any disks.
389 error = securelevel_ge(td->td_ucred, 2);
394 error = g_access(cp, r, w, e);
398 mtx_lock(&sc->sc_mtx);
399 if (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
400 wakeup(&sc->sc_active);
401 sc->sc_open += r + w + e;
402 if (sc->sc_open == 0)
403 atomic_clear_int(&sc->sc_active, SC_A_OPEN);
405 atomic_set_int(&sc->sc_active, SC_A_OPEN);
406 mtx_unlock(&sc->sc_mtx);
412 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
414 struct g_consumer *cp;
415 struct g_dev_softc *sc;
421 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
422 cp->geom->name, flags, fmt, td);
424 r = flags & FREAD ? -1 : 0;
425 w = flags & FWRITE ? -1 : 0;
427 e = flags & O_EXCL ? -1 : 0;
433 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
434 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
435 * which would result in zero deltas, which in turn would cause
436 * panic in g_access(9).
438 * Note that we cannot zero the counters (ie. do "r = cp->acr"
439 * etc) instead, because the consumer might be opened in another
446 mtx_lock(&sc->sc_mtx);
447 sc->sc_open += r + w + e;
448 if (sc->sc_open == 0)
449 atomic_clear_int(&sc->sc_active, SC_A_OPEN);
451 atomic_set_int(&sc->sc_active, SC_A_OPEN);
452 while (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
453 msleep(&sc->sc_active, &sc->sc_mtx, 0, "g_dev_close", hz / 10);
454 mtx_unlock(&sc->sc_mtx);
456 error = g_access(cp, r, w, e);
462 * XXX: Until we have unmessed the ioctl situation, there is a race against
463 * XXX: a concurrent orphanization. We cannot close it by holding topology
464 * XXX: since that would prevent us from doing our job, and stalling events
465 * XXX: will break (actually: stall) the BSD disklabel hacks.
468 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
470 struct g_consumer *cp;
471 struct g_provider *pp;
472 off_t offset, length, chunk, odd;
479 KASSERT(cp->acr || cp->acw,
480 ("Consumer with zero access count in g_dev_ioctl"));
482 i = IOCPARM_LEN(cmd);
484 case DIOCGSECTORSIZE:
485 *(u_int *)data = cp->provider->sectorsize;
486 if (*(u_int *)data == 0)
490 *(off_t *)data = cp->provider->mediasize;
491 if (*(off_t *)data == 0)
495 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
496 if (error == 0 && *(u_int *)data == 0)
500 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
501 if (error == 0 && *(u_int *)data == 0)
504 case DIOCGFRONTSTUFF:
505 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
507 case DIOCSKERNELDUMP:
508 if (*(u_int *)data == 0)
509 error = g_dev_setdumpdev(NULL, td);
511 error = g_dev_setdumpdev(dev, td);
514 error = g_io_flush(cp);
517 offset = ((off_t *)data)[0];
518 length = ((off_t *)data)[1];
519 if ((offset % cp->provider->sectorsize) != 0 ||
520 (length % cp->provider->sectorsize) != 0 || length <= 0) {
521 printf("%s: offset=%jd length=%jd\n", __func__, offset,
528 if (g_dev_del_max_sectors != 0 && chunk >
529 g_dev_del_max_sectors * cp->provider->sectorsize) {
530 chunk = g_dev_del_max_sectors *
531 cp->provider->sectorsize;
532 if (cp->provider->stripesize > 0) {
533 odd = (offset + chunk +
534 cp->provider->stripeoffset) %
535 cp->provider->stripesize;
540 error = g_delete_data(cp, offset, chunk);
546 * Since the request size can be large, the service
547 * time can be is likewise. We make this ioctl
548 * interruptible by checking for signals for each bio.
555 error = g_io_getattr("GEOM::ident", cp, &i, data);
557 case DIOCGPROVIDERNAME:
560 strlcpy(data, pp->name, i);
562 case DIOCGSTRIPESIZE:
563 *(off_t *)data = cp->provider->stripesize;
565 case DIOCGSTRIPEOFFSET:
566 *(off_t *)data = cp->provider->stripeoffset;
569 error = g_io_getattr("GEOM::physpath", cp, &i, data);
570 if (error == 0 && *(char *)data == '\0')
574 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
576 if (arg->len > sizeof(arg->value)) {
580 error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
584 struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
585 struct disk_zone_rep_entry *new_entries, *old_entries;
586 struct disk_zone_report *rep;
594 if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
595 rep = &zone_args->zone_params.report;
596 #define MAXENTRIES (MAXPHYS / sizeof(struct disk_zone_rep_entry))
597 if (rep->entries_allocated > MAXENTRIES)
598 rep->entries_allocated = MAXENTRIES;
599 alloc_size = rep->entries_allocated *
600 sizeof(struct disk_zone_rep_entry);
602 new_entries = g_malloc(alloc_size,
604 old_entries = rep->entries;
605 rep->entries = new_entries;
607 error = g_io_zonecmd(zone_args, cp);
608 if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES &&
609 alloc_size != 0 && error == 0)
610 error = copyout(new_entries, old_entries, alloc_size);
611 if (old_entries != NULL && rep != NULL)
612 rep->entries = old_entries;
613 if (new_entries != NULL)
618 if (cp->provider->geom->ioctl != NULL) {
619 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
629 g_dev_done(struct bio *bp2)
631 struct g_consumer *cp;
632 struct g_dev_softc *sc;
638 bp = bp2->bio_parent;
639 bp->bio_error = bp2->bio_error;
640 bp->bio_completed = bp2->bio_completed;
641 bp->bio_resid = bp->bio_length - bp2->bio_completed;
642 if (bp2->bio_cmd == BIO_ZONE)
643 bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
645 if (bp2->bio_error != 0) {
646 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
647 bp2, bp2->bio_error);
648 bp->bio_flags |= BIO_ERROR;
650 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
651 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
654 active = atomic_fetchadd_int(&sc->sc_active, -1) - 1;
655 if ((active & SC_A_ACTIVE) == 0) {
656 if ((active & SC_A_OPEN) == 0)
657 wakeup(&sc->sc_active);
658 if (active & SC_A_DESTROY)
659 g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
665 g_dev_strategy(struct bio *bp)
667 struct g_consumer *cp;
670 struct g_dev_softc *sc;
672 KASSERT(bp->bio_cmd == BIO_READ ||
673 bp->bio_cmd == BIO_WRITE ||
674 bp->bio_cmd == BIO_DELETE ||
675 bp->bio_cmd == BIO_FLUSH ||
676 bp->bio_cmd == BIO_ZONE,
677 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
681 KASSERT(cp->acr || cp->acw,
682 ("Consumer with zero access count in g_dev_strategy"));
684 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
685 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
686 bp->bio_resid = bp->bio_bcount;
687 biofinish(bp, NULL, EINVAL);
691 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
692 atomic_add_int(&sc->sc_active, 1);
696 * XXX: This is not an ideal solution, but I believe it to
697 * XXX: deadlock safely, all things considered.
699 bp2 = g_clone_bio(bp);
702 pause("gdstrat", hz / 10);
704 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
705 bp2->bio_done = g_dev_done;
707 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
708 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
709 bp2->bio_data, bp2->bio_cmd);
710 g_io_request(bp2, cp);
711 KASSERT(cp->acr || cp->acw,
712 ("g_dev_strategy raced with g_dev_close and lost"));
719 * Called by devfs when asynchronous device destruction is completed.
720 * - Mark that we have no attached device any more.
721 * - If there are no outstanding requests, schedule geom destruction.
722 * Otherwise destruction will be scheduled later by g_dev_done().
726 g_dev_callback(void *arg)
728 struct g_consumer *cp;
729 struct g_dev_softc *sc;
734 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
738 active = atomic_fetchadd_int(&sc->sc_active, SC_A_DESTROY);
739 if ((active & SC_A_ACTIVE) == 0)
740 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
746 * Called from below when the provider orphaned us.
747 * - Clear any dump settings.
748 * - Request asynchronous device destruction to prevent any more requests
749 * from coming in. The provider is already marked with an error, so
750 * anything which comes in the interim will be returned immediately.
754 g_dev_orphan(struct g_consumer *cp)
757 struct g_dev_softc *sc;
762 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
764 /* Reset any dump-area set on this device */
765 if (dev->si_flags & SI_DUMPDEV)
766 (void)set_dumper(NULL, NULL, curthread);
768 /* Destroy the struct cdev *so we get no more requests */
770 destroy_dev_sched_cb(dev, g_dev_callback, cp);
773 DECLARE_GEOM_CLASS(g_dev_class, g_dev);