FreeBSD/Linux Kernel Cross Reference
sys/geom/geom_dev.c
1 /*-
2 * Copyright (c) 2002 Poul-Henning Kamp
3 * Copyright (c) 2002 Networks Associates Technology, Inc.
4 * All rights reserved.
5 *
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.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
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
21 * permission.
22 *
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
33 * SUCH DAMAGE.
34 */
35
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD: releng/9.1/sys/geom/geom_dev.c 223089 2011-06-14 17:10:32Z gibbs $");
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/kernel.h>
43 #include <sys/conf.h>
44 #include <sys/ctype.h>
45 #include <sys/bio.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/proc.h>
49 #include <sys/errno.h>
50 #include <sys/time.h>
51 #include <sys/disk.h>
52 #include <sys/fcntl.h>
53 #include <sys/limits.h>
54 #include <geom/geom.h>
55 #include <geom/geom_int.h>
56 #include <machine/stdarg.h>
57
58 /*
59 * Use the consumer private field to reference a physdev alias (if any).
60 */
61 #define cp_alias_dev private
62
63 static d_open_t g_dev_open;
64 static d_close_t g_dev_close;
65 static d_strategy_t g_dev_strategy;
66 static d_ioctl_t g_dev_ioctl;
67
68 static struct cdevsw g_dev_cdevsw = {
69 .d_version = D_VERSION,
70 .d_open = g_dev_open,
71 .d_close = g_dev_close,
72 .d_read = physread,
73 .d_write = physwrite,
74 .d_ioctl = g_dev_ioctl,
75 .d_strategy = g_dev_strategy,
76 .d_name = "g_dev",
77 .d_flags = D_DISK | D_TRACKCLOSE,
78 };
79
80 static g_taste_t g_dev_taste;
81 static g_orphan_t g_dev_orphan;
82 static g_attrchanged_t g_dev_attrchanged;
83
84 static struct g_class g_dev_class = {
85 .name = "DEV",
86 .version = G_VERSION,
87 .taste = g_dev_taste,
88 .orphan = g_dev_orphan,
89 .attrchanged = g_dev_attrchanged
90 };
91
92 void
93 g_dev_print(void)
94 {
95 struct g_geom *gp;
96 char const *p = "";
97
98 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
99 printf("%s%s", p, gp->name);
100 p = " ";
101 }
102 printf("\n");
103 }
104
105 static void
106 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
107 {
108
109 if (strcmp(attr, "GEOM::physpath") != 0)
110 return;
111
112 if (g_access(cp, 1, 0, 0) == 0) {
113 char *physpath;
114 int error, physpath_len;
115
116 physpath_len = MAXPATHLEN;
117 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
118 error =
119 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
120 g_access(cp, -1, 0, 0);
121 if (error == 0 && strlen(physpath) != 0) {
122 struct cdev *dev;
123 struct cdev *old_alias_dev;
124 struct cdev **alias_devp;
125
126 dev = cp->geom->softc;
127 old_alias_dev = cp->cp_alias_dev;
128 alias_devp = (struct cdev **)&cp->cp_alias_dev;
129 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp,
130 dev, old_alias_dev, physpath);
131 } else if (cp->cp_alias_dev) {
132 destroy_dev((struct cdev *)cp->cp_alias_dev);
133 cp->cp_alias_dev = NULL;
134 }
135 g_free(physpath);
136 }
137 }
138
139 struct g_provider *
140 g_dev_getprovider(struct cdev *dev)
141 {
142 struct g_consumer *cp;
143
144 g_topology_assert();
145 if (dev == NULL)
146 return (NULL);
147 if (dev->si_devsw != &g_dev_cdevsw)
148 return (NULL);
149 cp = dev->si_drv2;
150 return (cp->provider);
151 }
152
153 static struct g_geom *
154 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
155 {
156 struct g_geom *gp;
157 struct g_consumer *cp;
158 int error, len;
159 struct cdev *dev, *adev;
160 char buf[64], *val;
161
162 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
163 g_topology_assert();
164 LIST_FOREACH(cp, &pp->consumers, consumers)
165 if (cp->geom->class == mp)
166 return (NULL);
167 gp = g_new_geomf(mp, pp->name);
168 cp = g_new_consumer(gp);
169 error = g_attach(cp, pp);
170 KASSERT(error == 0,
171 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
172 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
173 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
174 if (error != 0) {
175 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
176 __func__, gp->name, error);
177 g_detach(cp);
178 g_destroy_consumer(cp);
179 g_destroy_geom(gp);
180 return (NULL);
181 }
182
183 /* Search for device alias name and create it if found. */
184 adev = NULL;
185 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
186 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
187 buf[14 + len] = 0;
188 val = getenv(buf);
189 if (val != NULL) {
190 snprintf(buf, sizeof(buf), "%s%s",
191 val, gp->name + len);
192 freeenv(val);
193 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
194 &adev, dev, "%s", buf);
195 break;
196 }
197 }
198
199 if (pp->flags & G_PF_CANDELETE)
200 dev->si_flags |= SI_CANDELETE;
201 dev->si_iosize_max = MAXPHYS;
202 gp->softc = dev;
203 dev->si_drv1 = gp;
204 dev->si_drv2 = cp;
205 if (adev != NULL) {
206 if (pp->flags & G_PF_CANDELETE)
207 adev->si_flags |= SI_CANDELETE;
208 adev->si_iosize_max = MAXPHYS;
209 adev->si_drv1 = gp;
210 adev->si_drv2 = cp;
211 }
212
213 g_dev_attrchanged(cp, "GEOM::physpath");
214
215 return (gp);
216 }
217
218 static int
219 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
220 {
221 struct g_geom *gp;
222 struct g_consumer *cp;
223 int error, r, w, e;
224
225 gp = dev->si_drv1;
226 cp = dev->si_drv2;
227 if (gp == NULL || cp == NULL || gp->softc != dev)
228 return(ENXIO); /* g_dev_taste() not done yet */
229
230 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
231 gp->name, flags, fmt, td);
232
233 r = flags & FREAD ? 1 : 0;
234 w = flags & FWRITE ? 1 : 0;
235 #ifdef notyet
236 e = flags & O_EXCL ? 1 : 0;
237 #else
238 e = 0;
239 #endif
240 if (w) {
241 /*
242 * When running in very secure mode, do not allow
243 * opens for writing of any disks.
244 */
245 error = securelevel_ge(td->td_ucred, 2);
246 if (error)
247 return (error);
248 }
249 g_topology_lock();
250 if (dev->si_devsw == NULL)
251 error = ENXIO; /* We were orphaned */
252 else
253 error = g_access(cp, r, w, e);
254 g_topology_unlock();
255 return(error);
256 }
257
258 static int
259 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
260 {
261 struct g_geom *gp;
262 struct g_consumer *cp;
263 int error, r, w, e, i;
264
265 gp = dev->si_drv1;
266 cp = dev->si_drv2;
267 if (gp == NULL || cp == NULL)
268 return(ENXIO);
269 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
270 gp->name, flags, fmt, td);
271 r = flags & FREAD ? -1 : 0;
272 w = flags & FWRITE ? -1 : 0;
273 #ifdef notyet
274 e = flags & O_EXCL ? -1 : 0;
275 #else
276 e = 0;
277 #endif
278 g_topology_lock();
279 if (dev->si_devsw == NULL)
280 error = ENXIO; /* We were orphaned */
281 else
282 error = g_access(cp, r, w, e);
283 for (i = 0; i < 10 * hz;) {
284 if (cp->acr != 0 || cp->acw != 0)
285 break;
286 if (cp->nstart == cp->nend)
287 break;
288 pause("gdevwclose", hz / 10);
289 i += hz / 10;
290 }
291 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) {
292 printf("WARNING: Final close of geom_dev(%s) %s %s\n",
293 gp->name,
294 "still has outstanding I/O after 10 seconds.",
295 "Completing close anyway, panic may happen later.");
296 }
297 g_topology_unlock();
298 return (error);
299 }
300
301 /*
302 * XXX: Until we have unmessed the ioctl situation, there is a race against
303 * XXX: a concurrent orphanization. We cannot close it by holding topology
304 * XXX: since that would prevent us from doing our job, and stalling events
305 * XXX: will break (actually: stall) the BSD disklabel hacks.
306 */
307 static int
308 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
309 {
310 struct g_geom *gp;
311 struct g_consumer *cp;
312 struct g_provider *pp;
313 struct g_kerneldump kd;
314 off_t offset, length, chunk;
315 int i, error;
316 u_int u;
317
318 gp = dev->si_drv1;
319 cp = dev->si_drv2;
320 pp = cp->provider;
321
322 error = 0;
323 KASSERT(cp->acr || cp->acw,
324 ("Consumer with zero access count in g_dev_ioctl"));
325
326 i = IOCPARM_LEN(cmd);
327 switch (cmd) {
328 case DIOCGSECTORSIZE:
329 *(u_int *)data = cp->provider->sectorsize;
330 if (*(u_int *)data == 0)
331 error = ENOENT;
332 break;
333 case DIOCGMEDIASIZE:
334 *(off_t *)data = cp->provider->mediasize;
335 if (*(off_t *)data == 0)
336 error = ENOENT;
337 break;
338 case DIOCGFWSECTORS:
339 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
340 if (error == 0 && *(u_int *)data == 0)
341 error = ENOENT;
342 break;
343 case DIOCGFWHEADS:
344 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
345 if (error == 0 && *(u_int *)data == 0)
346 error = ENOENT;
347 break;
348 case DIOCGFRONTSTUFF:
349 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
350 break;
351 case DIOCSKERNELDUMP:
352 u = *((u_int *)data);
353 if (!u) {
354 set_dumper(NULL);
355 error = 0;
356 break;
357 }
358 kd.offset = 0;
359 kd.length = OFF_MAX;
360 i = sizeof kd;
361 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
362 if (!error) {
363 error = set_dumper(&kd.di);
364 if (!error)
365 dev->si_flags |= SI_DUMPDEV;
366 }
367 break;
368 case DIOCGFLUSH:
369 error = g_io_flush(cp);
370 break;
371 case DIOCGDELETE:
372 offset = ((off_t *)data)[0];
373 length = ((off_t *)data)[1];
374 if ((offset % cp->provider->sectorsize) != 0 ||
375 (length % cp->provider->sectorsize) != 0 || length <= 0) {
376 printf("%s: offset=%jd length=%jd\n", __func__, offset,
377 length);
378 error = EINVAL;
379 break;
380 }
381 while (length > 0) {
382 chunk = length;
383 if (chunk > 65536 * cp->provider->sectorsize)
384 chunk = 65536 * cp->provider->sectorsize;
385 error = g_delete_data(cp, offset, chunk);
386 length -= chunk;
387 offset += chunk;
388 if (error)
389 break;
390 /*
391 * Since the request size is unbounded, the service
392 * time is likewise. We make this ioctl interruptible
393 * by checking for signals for each bio.
394 */
395 if (SIGPENDING(td))
396 break;
397 }
398 break;
399 case DIOCGIDENT:
400 error = g_io_getattr("GEOM::ident", cp, &i, data);
401 break;
402 case DIOCGPROVIDERNAME:
403 if (pp == NULL)
404 return (ENOENT);
405 strlcpy(data, pp->name, i);
406 break;
407 case DIOCGSTRIPESIZE:
408 *(off_t *)data = cp->provider->stripesize;
409 break;
410 case DIOCGSTRIPEOFFSET:
411 *(off_t *)data = cp->provider->stripeoffset;
412 break;
413 case DIOCGPHYSPATH:
414 error = g_io_getattr("GEOM::physpath", cp, &i, data);
415 if (error == 0 && *(char *)data == '\0')
416 error = ENOENT;
417 break;
418 default:
419 if (cp->provider->geom->ioctl != NULL) {
420 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
421 } else {
422 error = ENOIOCTL;
423 }
424 }
425
426 return (error);
427 }
428
429 static void
430 g_dev_done(struct bio *bp2)
431 {
432 struct bio *bp;
433
434 bp = bp2->bio_parent;
435 bp->bio_error = bp2->bio_error;
436 if (bp->bio_error != 0) {
437 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
438 bp2, bp->bio_error);
439 bp->bio_flags |= BIO_ERROR;
440 } else {
441 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
442 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed);
443 }
444 bp->bio_resid = bp->bio_length - bp2->bio_completed;
445 bp->bio_completed = bp2->bio_completed;
446 g_destroy_bio(bp2);
447 biodone(bp);
448 }
449
450 static void
451 g_dev_strategy(struct bio *bp)
452 {
453 struct g_consumer *cp;
454 struct bio *bp2;
455 struct cdev *dev;
456
457 KASSERT(bp->bio_cmd == BIO_READ ||
458 bp->bio_cmd == BIO_WRITE ||
459 bp->bio_cmd == BIO_DELETE,
460 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
461 dev = bp->bio_dev;
462 cp = dev->si_drv2;
463 KASSERT(cp->acr || cp->acw,
464 ("Consumer with zero access count in g_dev_strategy"));
465 #ifdef INVARIANTS
466 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
467 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
468 bp->bio_resid = bp->bio_bcount;
469 biofinish(bp, NULL, EINVAL);
470 return;
471 }
472 #endif
473 for (;;) {
474 /*
475 * XXX: This is not an ideal solution, but I belive it to
476 * XXX: deadlock safe, all things considered.
477 */
478 bp2 = g_clone_bio(bp);
479 if (bp2 != NULL)
480 break;
481 pause("gdstrat", hz / 10);
482 }
483 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
484 bp2->bio_done = g_dev_done;
485 g_trace(G_T_BIO,
486 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
487 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
488 bp2->bio_data, bp2->bio_cmd);
489 g_io_request(bp2, cp);
490 KASSERT(cp->acr || cp->acw,
491 ("g_dev_strategy raced with g_dev_close and lost"));
492
493 }
494
495 /*
496 * g_dev_orphan()
497 *
498 * Called from below when the provider orphaned us.
499 * - Clear any dump settings.
500 * - Destroy the struct cdev *to prevent any more request from coming in. The
501 * provider is already marked with an error, so anything which comes in
502 * in the interrim will be returned immediately.
503 * - Wait for any outstanding I/O to finish.
504 * - Set our access counts to zero, whatever they were.
505 * - Detach and self-destruct.
506 */
507
508 static void
509 g_dev_orphan(struct g_consumer *cp)
510 {
511 struct g_geom *gp;
512 struct cdev *dev;
513
514 g_topology_assert();
515 gp = cp->geom;
516 dev = gp->softc;
517 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name);
518
519 /* Reset any dump-area set on this device */
520 if (dev->si_flags & SI_DUMPDEV)
521 set_dumper(NULL);
522
523 /* Destroy the struct cdev *so we get no more requests */
524 destroy_dev(dev);
525
526 /* Wait for the cows to come home */
527 while (cp->nstart != cp->nend)
528 pause("gdevorphan", hz / 10);
529
530 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
531 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
532
533 g_detach(cp);
534 g_destroy_consumer(cp);
535 g_destroy_geom(gp);
536 }
537
538 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
Cache object: 4371ec3a3484242e6be5530b004404a2
|