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/10.0/sys/geom/geom_dev.c 254389 2013-08-15 22:52:39Z ken $");
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/bus.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/proc.h>
50 #include <sys/errno.h>
51 #include <sys/time.h>
52 #include <sys/disk.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>
59
60 struct g_dev_softc {
61 struct mtx sc_mtx;
62 struct cdev *sc_dev;
63 struct cdev *sc_alias;
64 int sc_open;
65 int sc_active;
66 };
67
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;
72
73 static struct cdevsw g_dev_cdevsw = {
74 .d_version = D_VERSION,
75 .d_open = g_dev_open,
76 .d_close = g_dev_close,
77 .d_read = physread,
78 .d_write = physwrite,
79 .d_ioctl = g_dev_ioctl,
80 .d_strategy = g_dev_strategy,
81 .d_name = "g_dev",
82 .d_flags = D_DISK | D_TRACKCLOSE,
83 };
84
85 static g_taste_t g_dev_taste;
86 static g_orphan_t g_dev_orphan;
87 static g_attrchanged_t g_dev_attrchanged;
88
89 static struct g_class g_dev_class = {
90 .name = "DEV",
91 .version = G_VERSION,
92 .taste = g_dev_taste,
93 .orphan = g_dev_orphan,
94 .attrchanged = g_dev_attrchanged
95 };
96
97 /*
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.
101 */
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)");
109
110 static void
111 g_dev_destroy(void *arg, int flags __unused)
112 {
113 struct g_consumer *cp;
114 struct g_geom *gp;
115 struct g_dev_softc *sc;
116
117 g_topology_assert();
118 cp = arg;
119 gp = cp->geom;
120 sc = cp->private;
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);
124 g_detach(cp);
125 g_destroy_consumer(cp);
126 g_destroy_geom(gp);
127 mtx_destroy(&sc->sc_mtx);
128 g_free(sc);
129 }
130
131 void
132 g_dev_print(void)
133 {
134 struct g_geom *gp;
135 char const *p = "";
136
137 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
138 printf("%s%s", p, gp->name);
139 p = " ";
140 }
141 printf("\n");
142 }
143
144 static void
145 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
146 {
147 struct g_dev_softc *sc;
148 struct cdev *dev;
149 char buf[SPECNAMELEN + 6];
150
151 sc = cp->private;
152 if (strcmp(attr, "GEOM::media") == 0) {
153 dev = sc->sc_dev;
154 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
155 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
156 dev = sc->sc_alias;
157 if (dev != NULL) {
158 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
159 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf,
160 M_WAITOK);
161 }
162 return;
163 }
164
165 if (strcmp(attr, "GEOM::physpath") != 0)
166 return;
167
168 if (g_access(cp, 1, 0, 0) == 0) {
169 char *physpath;
170 int error, physpath_len;
171
172 physpath_len = MAXPATHLEN;
173 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
174 error =
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;
180
181 dev = sc->sc_dev;
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);
188 sc->sc_alias = NULL;
189 }
190 g_free(physpath);
191 }
192 }
193
194 struct g_provider *
195 g_dev_getprovider(struct cdev *dev)
196 {
197 struct g_consumer *cp;
198
199 g_topology_assert();
200 if (dev == NULL)
201 return (NULL);
202 if (dev->si_devsw != &g_dev_cdevsw)
203 return (NULL);
204 cp = dev->si_drv2;
205 return (cp->provider);
206 }
207
208 static struct g_geom *
209 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
210 {
211 struct g_geom *gp;
212 struct g_consumer *cp;
213 struct g_dev_softc *sc;
214 int error, len;
215 struct cdev *dev, *adev;
216 char buf[64], *val;
217
218 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
219 g_topology_assert();
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);
224 cp->private = sc;
225 error = g_attach(cp, pp);
226 KASSERT(error == 0,
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);
230 if (error != 0) {
231 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
232 __func__, gp->name, error);
233 g_detach(cp);
234 g_destroy_consumer(cp);
235 g_destroy_geom(gp);
236 mtx_destroy(&sc->sc_mtx);
237 g_free(sc);
238 return (NULL);
239 }
240 dev->si_flags |= SI_UNMAPPED;
241 sc->sc_dev = dev;
242
243 /* Search for device alias name and create it if found. */
244 adev = NULL;
245 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
246 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
247 buf[14 + len] = 0;
248 val = getenv(buf);
249 if (val != NULL) {
250 snprintf(buf, sizeof(buf), "%s%s",
251 val, gp->name + len);
252 freeenv(val);
253 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
254 &adev, dev, "%s", buf);
255 adev->si_flags |= SI_UNMAPPED;
256 break;
257 }
258 }
259
260 dev->si_iosize_max = MAXPHYS;
261 dev->si_drv2 = cp;
262 if (adev != NULL) {
263 adev->si_iosize_max = MAXPHYS;
264 adev->si_drv2 = cp;
265 }
266
267 g_dev_attrchanged(cp, "GEOM::physpath");
268
269 return (gp);
270 }
271
272 static int
273 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
274 {
275 struct g_consumer *cp;
276 struct g_dev_softc *sc;
277 int error, r, w, e;
278
279 cp = dev->si_drv2;
280 if (cp == NULL)
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);
284
285 r = flags & FREAD ? 1 : 0;
286 w = flags & FWRITE ? 1 : 0;
287 #ifdef notyet
288 e = flags & O_EXCL ? 1 : 0;
289 #else
290 e = 0;
291 #endif
292 if (w) {
293 /*
294 * When running in very secure mode, do not allow
295 * opens for writing of any disks.
296 */
297 error = securelevel_ge(td->td_ucred, 2);
298 if (error)
299 return (error);
300 }
301 g_topology_lock();
302 error = g_access(cp, r, w, e);
303 g_topology_unlock();
304 if (error == 0) {
305 sc = cp->private;
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);
311 }
312 return(error);
313 }
314
315 static int
316 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
317 {
318 struct g_consumer *cp;
319 struct g_dev_softc *sc;
320 int error, r, w, e;
321
322 cp = dev->si_drv2;
323 if (cp == NULL)
324 return(ENXIO);
325 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
326 cp->geom->name, flags, fmt, td);
327
328 r = flags & FREAD ? -1 : 0;
329 w = flags & FWRITE ? -1 : 0;
330 #ifdef notyet
331 e = flags & O_EXCL ? -1 : 0;
332 #else
333 e = 0;
334 #endif
335 sc = cp->private;
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);
341 g_topology_lock();
342 error = g_access(cp, r, w, e);
343 g_topology_unlock();
344 return (error);
345 }
346
347 /*
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.
352 */
353 static int
354 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
355 {
356 struct g_consumer *cp;
357 struct g_provider *pp;
358 struct g_kerneldump kd;
359 off_t offset, length, chunk;
360 int i, error;
361 u_int u;
362
363 cp = dev->si_drv2;
364 pp = cp->provider;
365
366 error = 0;
367 KASSERT(cp->acr || cp->acw,
368 ("Consumer with zero access count in g_dev_ioctl"));
369
370 i = IOCPARM_LEN(cmd);
371 switch (cmd) {
372 case DIOCGSECTORSIZE:
373 *(u_int *)data = cp->provider->sectorsize;
374 if (*(u_int *)data == 0)
375 error = ENOENT;
376 break;
377 case DIOCGMEDIASIZE:
378 *(off_t *)data = cp->provider->mediasize;
379 if (*(off_t *)data == 0)
380 error = ENOENT;
381 break;
382 case DIOCGFWSECTORS:
383 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
384 if (error == 0 && *(u_int *)data == 0)
385 error = ENOENT;
386 break;
387 case DIOCGFWHEADS:
388 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
389 if (error == 0 && *(u_int *)data == 0)
390 error = ENOENT;
391 break;
392 case DIOCGFRONTSTUFF:
393 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
394 break;
395 case DIOCSKERNELDUMP:
396 u = *((u_int *)data);
397 if (!u) {
398 set_dumper(NULL, NULL);
399 error = 0;
400 break;
401 }
402 kd.offset = 0;
403 kd.length = OFF_MAX;
404 i = sizeof kd;
405 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
406 if (!error) {
407 error = set_dumper(&kd.di, devtoname(dev));
408 if (!error)
409 dev->si_flags |= SI_DUMPDEV;
410 }
411 break;
412 case DIOCGFLUSH:
413 error = g_io_flush(cp);
414 break;
415 case DIOCGDELETE:
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,
421 length);
422 error = EINVAL;
423 break;
424 }
425 while (length > 0) {
426 chunk = length;
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;
431 }
432 error = g_delete_data(cp, offset, chunk);
433 length -= chunk;
434 offset += chunk;
435 if (error)
436 break;
437 /*
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.
441 */
442 if (SIGPENDING(td))
443 break;
444 }
445 break;
446 case DIOCGIDENT:
447 error = g_io_getattr("GEOM::ident", cp, &i, data);
448 break;
449 case DIOCGPROVIDERNAME:
450 if (pp == NULL)
451 return (ENOENT);
452 strlcpy(data, pp->name, i);
453 break;
454 case DIOCGSTRIPESIZE:
455 *(off_t *)data = cp->provider->stripesize;
456 break;
457 case DIOCGSTRIPEOFFSET:
458 *(off_t *)data = cp->provider->stripeoffset;
459 break;
460 case DIOCGPHYSPATH:
461 error = g_io_getattr("GEOM::physpath", cp, &i, data);
462 if (error == 0 && *(char *)data == '\0')
463 error = ENOENT;
464 break;
465 default:
466 if (cp->provider->geom->ioctl != NULL) {
467 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
468 } else {
469 error = ENOIOCTL;
470 }
471 }
472
473 return (error);
474 }
475
476 static void
477 g_dev_done(struct bio *bp2)
478 {
479 struct g_consumer *cp;
480 struct g_dev_softc *sc;
481 struct bio *bp;
482 int destroy;
483
484 cp = bp2->bio_from;
485 sc = cp->private;
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",
490 bp2, bp->bio_error);
491 bp->bio_flags |= BIO_ERROR;
492 } else {
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);
495 }
496 bp->bio_resid = bp->bio_length - bp2->bio_completed;
497 bp->bio_completed = bp2->bio_completed;
498 g_destroy_bio(bp2);
499 destroy = 0;
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)
505 destroy = 1;
506 }
507 mtx_unlock(&sc->sc_mtx);
508 if (destroy)
509 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
510 biodone(bp);
511 }
512
513 static void
514 g_dev_strategy(struct bio *bp)
515 {
516 struct g_consumer *cp;
517 struct bio *bp2;
518 struct cdev *dev;
519 struct g_dev_softc *sc;
520
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));
526 dev = bp->bio_dev;
527 cp = dev->si_drv2;
528 sc = cp->private;
529 KASSERT(cp->acr || cp->acw,
530 ("Consumer with zero access count in g_dev_strategy"));
531 #ifdef INVARIANTS
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);
536 return;
537 }
538 #endif
539 mtx_lock(&sc->sc_mtx);
540 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
541 sc->sc_active++;
542 mtx_unlock(&sc->sc_mtx);
543
544 for (;;) {
545 /*
546 * XXX: This is not an ideal solution, but I belive it to
547 * XXX: deadlock safe, all things considered.
548 */
549 bp2 = g_clone_bio(bp);
550 if (bp2 != NULL)
551 break;
552 pause("gdstrat", hz / 10);
553 }
554 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
555 bp2->bio_done = g_dev_done;
556 g_trace(G_T_BIO,
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"));
563
564 }
565
566 /*
567 * g_dev_callback()
568 *
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().
573 */
574
575 static void
576 g_dev_callback(void *arg)
577 {
578 struct g_consumer *cp;
579 struct g_dev_softc *sc;
580 int destroy;
581
582 cp = arg;
583 sc = cp->private;
584 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
585
586 mtx_lock(&sc->sc_mtx);
587 sc->sc_dev = NULL;
588 sc->sc_alias = NULL;
589 destroy = (sc->sc_active == 0);
590 mtx_unlock(&sc->sc_mtx);
591 if (destroy)
592 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
593 }
594
595 /*
596 * g_dev_orphan()
597 *
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.
603 */
604
605 static void
606 g_dev_orphan(struct g_consumer *cp)
607 {
608 struct cdev *dev;
609 struct g_dev_softc *sc;
610
611 g_topology_assert();
612 sc = cp->private;
613 dev = sc->sc_dev;
614 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
615
616 /* Reset any dump-area set on this device */
617 if (dev->si_flags & SI_DUMPDEV)
618 set_dumper(NULL, NULL);
619
620 /* Destroy the struct cdev *so we get no more requests */
621 destroy_dev_sched_cb(dev, g_dev_callback, cp);
622 }
623
624 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
Cache object: 424342fc1cfcb6b14cfcbb7decb479c0
|