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>
46 #include <sys/mutex.h>
48 #include <sys/errno.h>
51 #include <sys/fcntl.h>
52 #include <sys/limits.h>
53 #include <sys/sysctl.h>
54 #include <geom/geom.h>
55 #include <geom/geom_int.h>
57 static d_open_t g_dev_open;
58 static d_close_t g_dev_close;
59 static d_strategy_t g_dev_strategy;
60 static d_ioctl_t g_dev_ioctl;
62 static struct cdevsw g_dev_cdevsw = {
63 .d_version = D_VERSION,
65 .d_close = g_dev_close,
68 .d_ioctl = g_dev_ioctl,
69 .d_strategy = g_dev_strategy,
71 .d_flags = D_DISK | D_TRACKCLOSE,
74 static g_taste_t g_dev_taste;
75 static g_orphan_t g_dev_orphan;
77 static struct g_class g_dev_class = {
81 .orphan = g_dev_orphan,
85 * We target 262144 (8 x 32768) sectors by default as this significantly
86 * increases the throughput on commonly used SSD's with a marginal
87 * increase in non-interruptible request latency.
89 static uint64_t g_dev_del_max_sectors = 262144;
90 SYSCTL_DECL(_kern_geom);
91 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
92 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
93 &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
94 "delete request sent to the provider. Larger requests are chunked "
95 "so they can be interrupted. (0 = disable chunking)");
103 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
104 printf("%s%s", p, gp->name);
111 g_dev_getprovider(struct cdev *dev)
113 struct g_consumer *cp;
118 if (dev->si_devsw != &g_dev_cdevsw)
121 return (cp->provider);
125 static struct g_geom *
126 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
129 struct g_consumer *cp;
131 struct cdev *dev, *adev;
134 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
136 LIST_FOREACH(cp, &pp->consumers, consumers)
137 if (cp->geom->class == mp)
139 gp = g_new_geomf(mp, pp->name);
140 cp = g_new_consumer(gp);
141 error = g_attach(cp, pp);
143 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
144 dev = make_dev(&g_dev_cdevsw, 0,
145 UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
147 /* Search for device alias name and create it if found. */
149 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
150 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
154 snprintf(buf, sizeof(buf), "%s%s",
155 val, gp->name + len);
157 adev = make_dev_alias(dev, buf);
162 if (pp->flags & G_PF_CANDELETE)
163 dev->si_flags |= SI_CANDELETE;
164 dev->si_iosize_max = MAXPHYS;
169 if (pp->flags & G_PF_CANDELETE)
170 adev->si_flags |= SI_CANDELETE;
171 adev->si_iosize_max = MAXPHYS;
179 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
182 struct g_consumer *cp;
187 if (gp == NULL || cp == NULL || gp->softc != dev)
188 return(ENXIO); /* g_dev_taste() not done yet */
190 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
191 gp->name, flags, fmt, td);
193 r = flags & FREAD ? 1 : 0;
194 w = flags & FWRITE ? 1 : 0;
196 e = flags & O_EXCL ? 1 : 0;
202 * When running in very secure mode, do not allow
203 * opens for writing of any disks.
205 error = securelevel_ge(td->td_ucred, 2);
210 if (dev->si_devsw == NULL)
211 error = ENXIO; /* We were orphaned */
213 error = g_access(cp, r, w, e);
219 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
222 struct g_consumer *cp;
223 int error, r, w, e, i;
227 if (gp == NULL || cp == NULL)
229 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
230 gp->name, flags, fmt, td);
231 r = flags & FREAD ? -1 : 0;
232 w = flags & FWRITE ? -1 : 0;
234 e = flags & O_EXCL ? -1 : 0;
239 if (dev->si_devsw == NULL)
240 error = ENXIO; /* We were orphaned */
242 error = g_access(cp, r, w, e);
243 for (i = 0; i < 10 * hz;) {
244 if (cp->acr != 0 || cp->acw != 0)
246 if (cp->nstart == cp->nend)
248 pause("gdevwclose", hz / 10);
251 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) {
252 printf("WARNING: Final close of geom_dev(%s) %s %s\n",
254 "still has outstanding I/O after 10 seconds.",
255 "Completing close anyway, panic may happen later.");
262 * XXX: Until we have unmessed the ioctl situation, there is a race against
263 * XXX: a concurrent orphanization. We cannot close it by holding topology
264 * XXX: since that would prevent us from doing our job, and stalling events
265 * XXX: will break (actually: stall) the BSD disklabel hacks.
268 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
271 struct g_consumer *cp;
272 struct g_provider *pp;
273 struct g_kerneldump kd;
274 off_t offset, length, chunk;
283 KASSERT(cp->acr || cp->acw,
284 ("Consumer with zero access count in g_dev_ioctl"));
286 i = IOCPARM_LEN(cmd);
288 case DIOCGSECTORSIZE:
289 *(u_int *)data = cp->provider->sectorsize;
290 if (*(u_int *)data == 0)
294 *(off_t *)data = cp->provider->mediasize;
295 if (*(off_t *)data == 0)
299 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
300 if (error == 0 && *(u_int *)data == 0)
304 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
305 if (error == 0 && *(u_int *)data == 0)
308 case DIOCGFRONTSTUFF:
309 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
311 case DIOCSKERNELDUMP:
312 u = *((u_int *)data);
321 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
323 error = set_dumper(&kd.di);
325 dev->si_flags |= SI_DUMPDEV;
329 error = g_io_flush(cp);
332 offset = ((off_t *)data)[0];
333 length = ((off_t *)data)[1];
334 if ((offset % cp->provider->sectorsize) != 0 ||
335 (length % cp->provider->sectorsize) != 0 || length <= 0) {
336 printf("%s: offset=%jd length=%jd\n", __func__, offset,
343 if (g_dev_del_max_sectors != 0 && chunk >
344 g_dev_del_max_sectors * cp->provider->sectorsize) {
345 chunk = g_dev_del_max_sectors *
346 cp->provider->sectorsize;
348 error = g_delete_data(cp, offset, chunk);
354 * Since the request size can be large, the service
355 * time can be is likewise. We make this ioctl
356 * interruptible by checking for signals for each bio.
363 error = g_io_getattr("GEOM::ident", cp, &i, data);
365 case DIOCGPROVIDERNAME:
368 strlcpy(data, pp->name, i);
370 case DIOCGSTRIPESIZE:
371 *(off_t *)data = cp->provider->stripesize;
373 case DIOCGSTRIPEOFFSET:
374 *(off_t *)data = cp->provider->stripeoffset;
377 if (cp->provider->geom->ioctl != NULL) {
378 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
388 g_dev_done(struct bio *bp2)
392 bp = bp2->bio_parent;
393 bp->bio_error = bp2->bio_error;
394 if (bp->bio_error != 0) {
395 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
397 bp->bio_flags |= BIO_ERROR;
399 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
400 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed);
402 bp->bio_resid = bp->bio_length - bp2->bio_completed;
403 bp->bio_completed = bp2->bio_completed;
409 g_dev_strategy(struct bio *bp)
411 struct g_consumer *cp;
415 KASSERT(bp->bio_cmd == BIO_READ ||
416 bp->bio_cmd == BIO_WRITE ||
417 bp->bio_cmd == BIO_DELETE,
418 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
421 KASSERT(cp->acr || cp->acw,
422 ("Consumer with zero access count in g_dev_strategy"));
424 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
425 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
426 bp->bio_resid = bp->bio_bcount;
427 biofinish(bp, NULL, EINVAL);
433 * XXX: This is not an ideal solution, but I belive it to
434 * XXX: deadlock safe, all things considered.
436 bp2 = g_clone_bio(bp);
439 pause("gdstrat", hz / 10);
441 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
442 bp2->bio_done = g_dev_done;
444 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
445 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
446 bp2->bio_data, bp2->bio_cmd);
447 g_io_request(bp2, cp);
448 KASSERT(cp->acr || cp->acw,
449 ("g_dev_strategy raced with g_dev_close and lost"));
456 * Called from below when the provider orphaned us.
457 * - Clear any dump settings.
458 * - Destroy the struct cdev *to prevent any more request from coming in. The
459 * provider is already marked with an error, so anything which comes in
460 * in the interrim will be returned immediately.
461 * - Wait for any outstanding I/O to finish.
462 * - Set our access counts to zero, whatever they were.
463 * - Detach and self-destruct.
467 g_dev_orphan(struct g_consumer *cp)
475 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name);
477 /* Reset any dump-area set on this device */
478 if (dev->si_flags & SI_DUMPDEV)
481 /* Destroy the struct cdev *so we get no more requests */
484 /* Wait for the cows to come home */
485 while (cp->nstart != cp->nend)
486 pause("gdevorphan", hz / 10);
488 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
489 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
492 g_destroy_consumer(cp);
496 DECLARE_GEOM_CLASS(g_dev_class, g_dev);