xref: /freebsd/sys/geom/geom_disk.c (revision 91ebf7d76582a73ac4c0a32964e4d3912010671e)
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("$FreeBSD$");
38 
39 #include "opt_geom.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/bio.h>
46 #include <sys/bus.h>
47 #include <sys/ctype.h>
48 #include <sys/fcntl.h>
49 #include <sys/malloc.h>
50 #include <sys/sbuf.h>
51 #include <sys/devicestat.h>
52 #include <machine/md_var.h>
53 
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <geom/geom.h>
57 #include <geom/geom_disk.h>
58 #include <geom/geom_int.h>
59 
60 #include <dev/led/led.h>
61 
62 #include <machine/bus.h>
63 
64 struct g_disk_softc {
65 	struct mtx		 done_mtx;
66 	struct disk		*dp;
67 	struct sysctl_ctx_list	sysctl_ctx;
68 	struct sysctl_oid	*sysctl_tree;
69 	char			led[64];
70 	uint32_t		state;
71 	struct mtx		 start_mtx;
72 };
73 
74 static g_access_t g_disk_access;
75 static g_start_t g_disk_start;
76 static g_ioctl_t g_disk_ioctl;
77 static g_dumpconf_t g_disk_dumpconf;
78 static g_provgone_t g_disk_providergone;
79 
80 static struct g_class g_disk_class = {
81 	.name = G_DISK_CLASS_NAME,
82 	.version = G_VERSION,
83 	.start = g_disk_start,
84 	.access = g_disk_access,
85 	.ioctl = g_disk_ioctl,
86 	.providergone = g_disk_providergone,
87 	.dumpconf = g_disk_dumpconf,
88 };
89 
90 SYSCTL_DECL(_kern_geom);
91 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0,
92     "GEOM_DISK stuff");
93 
94 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
95 
96 static int
97 g_disk_access(struct g_provider *pp, int r, int w, int e)
98 {
99 	struct disk *dp;
100 	struct g_disk_softc *sc;
101 	int error;
102 
103 	g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
104 	    pp->name, r, w, e);
105 	g_topology_assert();
106 	sc = pp->private;
107 	if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
108 		/*
109 		 * Allow decreasing access count even if disk is not
110 		 * available anymore.
111 		 */
112 		if (r <= 0 && w <= 0 && e <= 0)
113 			return (0);
114 		return (ENXIO);
115 	}
116 	r += pp->acr;
117 	w += pp->acw;
118 	e += pp->ace;
119 	error = 0;
120 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
121 		if (dp->d_open != NULL) {
122 			error = dp->d_open(dp);
123 			if (bootverbose && error != 0)
124 				printf("Opened disk %s -> %d\n",
125 				    pp->name, error);
126 			if (error != 0)
127 				return (error);
128 		}
129 		pp->mediasize = dp->d_mediasize;
130 		pp->sectorsize = dp->d_sectorsize;
131 		if (dp->d_maxsize == 0) {
132 			printf("WARNING: Disk drive %s%d has no d_maxsize\n",
133 			    dp->d_name, dp->d_unit);
134 			dp->d_maxsize = DFLTPHYS;
135 		}
136 		if (dp->d_delmaxsize == 0) {
137 			if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
138 				printf("WARNING: Disk drive %s%d has no "
139 				    "d_delmaxsize\n", dp->d_name, dp->d_unit);
140 			}
141 			dp->d_delmaxsize = dp->d_maxsize;
142 		}
143 		pp->stripeoffset = dp->d_stripeoffset;
144 		pp->stripesize = dp->d_stripesize;
145 		dp->d_flags |= DISKFLAG_OPEN;
146 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
147 		if (dp->d_close != NULL) {
148 			error = dp->d_close(dp);
149 			if (error != 0)
150 				printf("Closed disk %s -> %d\n",
151 				    pp->name, error);
152 		}
153 		sc->state = G_STATE_ACTIVE;
154 		if (sc->led[0] != 0)
155 			led_set(sc->led, "0");
156 		dp->d_flags &= ~DISKFLAG_OPEN;
157 	}
158 	return (error);
159 }
160 
161 static void
162 g_disk_kerneldump(struct bio *bp, struct disk *dp)
163 {
164 	struct g_kerneldump *gkd;
165 	struct g_geom *gp;
166 
167 	gkd = (struct g_kerneldump*)bp->bio_data;
168 	gp = bp->bio_to->geom;
169 	g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)",
170 		gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
171 	if (dp->d_dump == NULL) {
172 		g_io_deliver(bp, ENODEV);
173 		return;
174 	}
175 	gkd->di.dumper = dp->d_dump;
176 	gkd->di.priv = dp;
177 	gkd->di.blocksize = dp->d_sectorsize;
178 	gkd->di.maxiosize = dp->d_maxsize;
179 	gkd->di.mediaoffset = gkd->offset;
180 	if ((gkd->offset + gkd->length) > dp->d_mediasize)
181 		gkd->length = dp->d_mediasize - gkd->offset;
182 	gkd->di.mediasize = gkd->length;
183 	g_io_deliver(bp, 0);
184 }
185 
186 static void
187 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
188 {
189 	const char *cmd;
190 
191 	memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
192 	if (sc->led[0] != 0) {
193 		switch (sc->state) {
194 		case G_STATE_FAILED:
195 			cmd = "1";
196 			break;
197 		case G_STATE_REBUILD:
198 			cmd = "f5";
199 			break;
200 		case G_STATE_RESYNC:
201 			cmd = "f1";
202 			break;
203 		default:
204 			cmd = "0";
205 			break;
206 		}
207 		led_set(sc->led, cmd);
208 	}
209 	g_io_deliver(bp, 0);
210 }
211 
212 static void
213 g_disk_done(struct bio *bp)
214 {
215 	struct bintime now;
216 	struct bio *bp2;
217 	struct g_disk_softc *sc;
218 
219 	/* See "notes" for why we need a mutex here */
220 	/* XXX: will witness accept a mix of Giant/unGiant drivers here ? */
221 	bp2 = bp->bio_parent;
222 	sc = bp2->bio_to->private;
223 	bp->bio_completed = bp->bio_length - bp->bio_resid;
224 	binuptime(&now);
225 	mtx_lock(&sc->done_mtx);
226 	if (bp2->bio_error == 0)
227 		bp2->bio_error = bp->bio_error;
228 	bp2->bio_completed += bp->bio_completed;
229 
230 	switch (bp->bio_cmd) {
231 	case BIO_ZONE:
232 		bcopy(&bp->bio_zone, &bp2->bio_zone, sizeof(bp->bio_zone));
233 		/*FALLTHROUGH*/
234 	case BIO_READ:
235 	case BIO_WRITE:
236 	case BIO_DELETE:
237 	case BIO_FLUSH:
238 		devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
239 		break;
240 	default:
241 		break;
242 	}
243 	bp2->bio_inbed++;
244 	if (bp2->bio_children == bp2->bio_inbed) {
245 		mtx_unlock(&sc->done_mtx);
246 		bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
247 		g_io_deliver(bp2, bp2->bio_error);
248 	} else
249 		mtx_unlock(&sc->done_mtx);
250 	g_destroy_bio(bp);
251 }
252 
253 static int
254 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
255 {
256 	struct disk *dp;
257 	struct g_disk_softc *sc;
258 	int error;
259 
260 	sc = pp->private;
261 	dp = sc->dp;
262 
263 	if (dp->d_ioctl == NULL)
264 		return (ENOIOCTL);
265 	error = dp->d_ioctl(dp, cmd, data, fflag, td);
266 	return (error);
267 }
268 
269 static off_t
270 g_disk_maxsize(struct disk *dp, struct bio *bp)
271 {
272 	if (bp->bio_cmd == BIO_DELETE)
273 		return (dp->d_delmaxsize);
274 	return (dp->d_maxsize);
275 }
276 
277 static int
278 g_disk_maxsegs(struct disk *dp, struct bio *bp)
279 {
280 	return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1);
281 }
282 
283 static void
284 g_disk_advance(struct disk *dp, struct bio *bp, off_t off)
285 {
286 
287 	bp->bio_offset += off;
288 	bp->bio_length -= off;
289 
290 	if ((bp->bio_flags & BIO_VLIST) != 0) {
291 		bus_dma_segment_t *seg, *end;
292 
293 		seg = (bus_dma_segment_t *)bp->bio_data;
294 		end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
295 		off += bp->bio_ma_offset;
296 		while (off >= seg->ds_len) {
297 			KASSERT((seg != end),
298 			    ("vlist request runs off the end"));
299 			off -= seg->ds_len;
300 			seg++;
301 		}
302 		bp->bio_ma_offset = off;
303 		bp->bio_ma_n = end - seg;
304 		bp->bio_data = (void *)seg;
305 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
306 		bp->bio_ma += off / PAGE_SIZE;
307 		bp->bio_ma_offset += off;
308 		bp->bio_ma_offset %= PAGE_SIZE;
309 		bp->bio_ma_n -= off / PAGE_SIZE;
310 	} else {
311 		bp->bio_data += off;
312 	}
313 }
314 
315 static void
316 g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset,
317     off_t *plength, int *ppages)
318 {
319 	uintptr_t seg_page_base;
320 	uintptr_t seg_page_end;
321 	off_t offset;
322 	off_t length;
323 	int seg_pages;
324 
325 	offset = *poffset;
326 	length = *plength;
327 
328 	if (length > seg->ds_len - offset)
329 		length = seg->ds_len - offset;
330 
331 	seg_page_base = trunc_page(seg->ds_addr + offset);
332 	seg_page_end  = round_page(seg->ds_addr + offset + length);
333 	seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT;
334 
335 	if (seg_pages > *ppages) {
336 		seg_pages = *ppages;
337 		length = (seg_page_base + (seg_pages << PAGE_SHIFT)) -
338 		    (seg->ds_addr + offset);
339 	}
340 
341 	*poffset = 0;
342 	*plength -= length;
343 	*ppages -= seg_pages;
344 }
345 
346 static off_t
347 g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg)
348 {
349 	bus_dma_segment_t *seg, *end;
350 	off_t residual;
351 	off_t offset;
352 	int pages;
353 
354 	seg = (bus_dma_segment_t *)bp->bio_data;
355 	end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
356 	residual = bp->bio_length;
357 	offset = bp->bio_ma_offset;
358 	pages = g_disk_maxsegs(dp, bp);
359 	while (residual != 0 && pages != 0) {
360 		KASSERT((seg != end),
361 		    ("vlist limit runs off the end"));
362 		g_disk_seg_limit(seg, &offset, &residual, &pages);
363 		seg++;
364 	}
365 	if (pendseg != NULL)
366 		*pendseg = seg;
367 	return (residual);
368 }
369 
370 static bool
371 g_disk_limit(struct disk *dp, struct bio *bp)
372 {
373 	bool limited = false;
374 	off_t maxsz;
375 
376 	maxsz = g_disk_maxsize(dp, bp);
377 
378 	/*
379 	 * XXX: If we have a stripesize we should really use it here.
380 	 *      Care should be taken in the delete case if this is done
381 	 *      as deletes can be very sensitive to size given how they
382 	 *      are processed.
383 	 */
384 	if (bp->bio_length > maxsz) {
385 		bp->bio_length = maxsz;
386 		limited = true;
387 	}
388 
389 	if ((bp->bio_flags & BIO_VLIST) != 0) {
390 		bus_dma_segment_t *firstseg, *endseg;
391 		off_t residual;
392 
393 		firstseg = (bus_dma_segment_t*)bp->bio_data;
394 		residual = g_disk_vlist_limit(dp, bp, &endseg);
395 		if (residual != 0) {
396 			bp->bio_ma_n = endseg - firstseg;
397 			bp->bio_length -= residual;
398 			limited = true;
399 		}
400 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
401 		bp->bio_ma_n =
402 		    howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE);
403 	}
404 
405 	return (limited);
406 }
407 
408 static void
409 g_disk_start(struct bio *bp)
410 {
411 	struct bio *bp2, *bp3;
412 	struct disk *dp;
413 	struct g_disk_softc *sc;
414 	int error;
415 	off_t off;
416 
417 	sc = bp->bio_to->private;
418 	if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
419 		g_io_deliver(bp, ENXIO);
420 		return;
421 	}
422 	error = EJUSTRETURN;
423 	switch(bp->bio_cmd) {
424 	case BIO_DELETE:
425 		if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
426 			error = EOPNOTSUPP;
427 			break;
428 		}
429 		/* fall-through */
430 	case BIO_READ:
431 	case BIO_WRITE:
432 		KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 ||
433 		    (bp->bio_flags & BIO_UNMAPPED) == 0,
434 		    ("unmapped bio not supported by disk %s", dp->d_name));
435 		off = 0;
436 		bp3 = NULL;
437 		bp2 = g_clone_bio(bp);
438 		if (bp2 == NULL) {
439 			error = ENOMEM;
440 			break;
441 		}
442 		for (;;) {
443 			if (g_disk_limit(dp, bp2)) {
444 				off += bp2->bio_length;
445 
446 				/*
447 				 * To avoid a race, we need to grab the next bio
448 				 * before we schedule this one.  See "notes".
449 				 */
450 				bp3 = g_clone_bio(bp);
451 				if (bp3 == NULL)
452 					bp->bio_error = ENOMEM;
453 			}
454 			bp2->bio_done = g_disk_done;
455 			bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
456 			bp2->bio_bcount = bp2->bio_length;
457 			bp2->bio_disk = dp;
458 			mtx_lock(&sc->start_mtx);
459 			devstat_start_transaction_bio(dp->d_devstat, bp2);
460 			mtx_unlock(&sc->start_mtx);
461 			dp->d_strategy(bp2);
462 
463 			if (bp3 == NULL)
464 				break;
465 
466 			bp2 = bp3;
467 			bp3 = NULL;
468 			g_disk_advance(dp, bp2, off);
469 		}
470 		break;
471 	case BIO_GETATTR:
472 		/* Give the driver a chance to override */
473 		if (dp->d_getattr != NULL) {
474 			if (bp->bio_disk == NULL)
475 				bp->bio_disk = dp;
476 			error = dp->d_getattr(bp);
477 			if (error != -1)
478 				break;
479 			error = EJUSTRETURN;
480 		}
481 		if (g_handleattr_int(bp, "GEOM::candelete",
482 		    (dp->d_flags & DISKFLAG_CANDELETE) != 0))
483 			break;
484 		else if (g_handleattr_int(bp, "GEOM::fwsectors",
485 		    dp->d_fwsectors))
486 			break;
487 		else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
488 			break;
489 		else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0))
490 			break;
491 		else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
492 			break;
493 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
494 		    dp->d_hba_vendor))
495 			break;
496 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
497 		    dp->d_hba_device))
498 			break;
499 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
500 		    dp->d_hba_subvendor))
501 			break;
502 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
503 		    dp->d_hba_subdevice))
504 			break;
505 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
506 			g_disk_kerneldump(bp, dp);
507 		else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
508 			g_disk_setstate(bp, sc);
509 		else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
510 		    dp->d_rotation_rate))
511 			break;
512 		else
513 			error = ENOIOCTL;
514 		break;
515 	case BIO_FLUSH:
516 		g_trace(G_T_BIO, "g_disk_flushcache(%s)",
517 		    bp->bio_to->name);
518 		if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
519 			error = EOPNOTSUPP;
520 			break;
521 		}
522 		/*FALLTHROUGH*/
523 	case BIO_ZONE:
524 		if (bp->bio_cmd == BIO_ZONE) {
525 			if (!(dp->d_flags & DISKFLAG_CANZONE)) {
526 				error = EOPNOTSUPP;
527 				break;
528 			}
529 			g_trace(G_T_BIO, "g_disk_zone(%s)",
530 			    bp->bio_to->name);
531 		}
532 		bp2 = g_clone_bio(bp);
533 		if (bp2 == NULL) {
534 			g_io_deliver(bp, ENOMEM);
535 			return;
536 		}
537 		bp2->bio_done = g_disk_done;
538 		bp2->bio_disk = dp;
539 		mtx_lock(&sc->start_mtx);
540 		devstat_start_transaction_bio(dp->d_devstat, bp2);
541 		mtx_unlock(&sc->start_mtx);
542 		dp->d_strategy(bp2);
543 		break;
544 	default:
545 		error = EOPNOTSUPP;
546 		break;
547 	}
548 	if (error != EJUSTRETURN)
549 		g_io_deliver(bp, error);
550 	return;
551 }
552 
553 static void
554 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
555 {
556 	struct bio *bp;
557 	struct disk *dp;
558 	struct g_disk_softc *sc;
559 	char *buf;
560 	int res = 0;
561 
562 	sc = gp->softc;
563 	if (sc == NULL || (dp = sc->dp) == NULL)
564 		return;
565 	if (indent == NULL) {
566 		sbuf_printf(sb, " hd %u", dp->d_fwheads);
567 		sbuf_printf(sb, " sc %u", dp->d_fwsectors);
568 		return;
569 	}
570 	if (pp != NULL) {
571 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
572 		    indent, dp->d_fwheads);
573 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
574 		    indent, dp->d_fwsectors);
575 
576 		/*
577 		 * "rotationrate" is a little complicated, because the value
578 		 * returned by the drive might not be the RPM; 0 and 1 are
579 		 * special cases, and there's also a valid range.
580 		 */
581 		sbuf_printf(sb, "%s<rotationrate>", indent);
582 		if (dp->d_rotation_rate == 0)		/* Old drives don't */
583 			sbuf_printf(sb, "unknown");	/* report RPM. */
584 		else if (dp->d_rotation_rate == 1)	/* Since 0 is used */
585 			sbuf_printf(sb, "0");		/* above, SSDs use 1. */
586 		else if ((dp->d_rotation_rate >= 0x041) &&
587 		    (dp->d_rotation_rate <= 0xfffe))
588 			sbuf_printf(sb, "%u", dp->d_rotation_rate);
589 		else
590 			sbuf_printf(sb, "invalid");
591 		sbuf_printf(sb, "</rotationrate>\n");
592 		if (dp->d_getattr != NULL) {
593 			buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
594 			bp = g_alloc_bio();
595 			bp->bio_disk = dp;
596 			bp->bio_attribute = "GEOM::ident";
597 			bp->bio_length = DISK_IDENT_SIZE;
598 			bp->bio_data = buf;
599 			res = dp->d_getattr(bp);
600 			sbuf_printf(sb, "%s<ident>", indent);
601 			g_conf_printf_escaped(sb, "%s",
602 			    res == 0 ? buf: dp->d_ident);
603 			sbuf_printf(sb, "</ident>\n");
604 			bp->bio_attribute = "GEOM::lunid";
605 			bp->bio_length = DISK_IDENT_SIZE;
606 			bp->bio_data = buf;
607 			if (dp->d_getattr(bp) == 0) {
608 				sbuf_printf(sb, "%s<lunid>", indent);
609 				g_conf_printf_escaped(sb, "%s", buf);
610 				sbuf_printf(sb, "</lunid>\n");
611 			}
612 			bp->bio_attribute = "GEOM::lunname";
613 			bp->bio_length = DISK_IDENT_SIZE;
614 			bp->bio_data = buf;
615 			if (dp->d_getattr(bp) == 0) {
616 				sbuf_printf(sb, "%s<lunname>", indent);
617 				g_conf_printf_escaped(sb, "%s", buf);
618 				sbuf_printf(sb, "</lunname>\n");
619 			}
620 			g_destroy_bio(bp);
621 			g_free(buf);
622 		} else {
623 			sbuf_printf(sb, "%s<ident>", indent);
624 			g_conf_printf_escaped(sb, "%s", dp->d_ident);
625 			sbuf_printf(sb, "</ident>\n");
626 		}
627 		sbuf_printf(sb, "%s<descr>", indent);
628 		g_conf_printf_escaped(sb, "%s", dp->d_descr);
629 		sbuf_printf(sb, "</descr>\n");
630 	}
631 }
632 
633 static void
634 g_disk_resize(void *ptr, int flag)
635 {
636 	struct disk *dp;
637 	struct g_geom *gp;
638 	struct g_provider *pp;
639 
640 	if (flag == EV_CANCEL)
641 		return;
642 	g_topology_assert();
643 
644 	dp = ptr;
645 	gp = dp->d_geom;
646 
647 	if (dp->d_destroyed || gp == NULL)
648 		return;
649 
650 	LIST_FOREACH(pp, &gp->provider, provider) {
651 		if (pp->sectorsize != 0 &&
652 		    pp->sectorsize != dp->d_sectorsize)
653 			g_wither_provider(pp, ENXIO);
654 		else
655 			g_resize_provider(pp, dp->d_mediasize);
656 	}
657 }
658 
659 static void
660 g_disk_create(void *arg, int flag)
661 {
662 	struct g_geom *gp;
663 	struct g_provider *pp;
664 	struct disk *dp;
665 	struct g_disk_softc *sc;
666 	char tmpstr[80];
667 
668 	if (flag == EV_CANCEL)
669 		return;
670 	g_topology_assert();
671 	dp = arg;
672 
673 	mtx_pool_lock(mtxpool_sleep, dp);
674 	dp->d_init_level = DISK_INIT_START;
675 
676 	/*
677 	 * If the disk has already gone away, we can just stop here and
678 	 * call the user's callback to tell him we've cleaned things up.
679 	 */
680 	if (dp->d_goneflag != 0) {
681 		mtx_pool_unlock(mtxpool_sleep, dp);
682 		if (dp->d_gone != NULL)
683 			dp->d_gone(dp);
684 		return;
685 	}
686 	mtx_pool_unlock(mtxpool_sleep, dp);
687 
688 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
689 	mtx_init(&sc->start_mtx, "g_disk_start", NULL, MTX_DEF);
690 	mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
691 	sc->dp = dp;
692 	gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
693 	gp->softc = sc;
694 	pp = g_new_providerf(gp, "%s", gp->name);
695 	devstat_remove_entry(pp->stat);
696 	pp->stat = NULL;
697 	dp->d_devstat->id = pp;
698 	pp->mediasize = dp->d_mediasize;
699 	pp->sectorsize = dp->d_sectorsize;
700 	pp->stripeoffset = dp->d_stripeoffset;
701 	pp->stripesize = dp->d_stripesize;
702 	if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
703 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
704 	if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
705 		pp->flags |= G_PF_DIRECT_SEND;
706 	pp->flags |= G_PF_DIRECT_RECEIVE;
707 	if (bootverbose)
708 		printf("GEOM: new disk %s\n", gp->name);
709 	sysctl_ctx_init(&sc->sysctl_ctx);
710 	snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
711 	sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
712 		SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
713 		CTLFLAG_RD, 0, tmpstr);
714 	if (sc->sysctl_tree != NULL) {
715 		SYSCTL_ADD_STRING(&sc->sysctl_ctx,
716 		    SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
717 		    CTLFLAG_RWTUN, sc->led, sizeof(sc->led),
718 		    "LED name");
719 	}
720 	pp->private = sc;
721 	dp->d_geom = gp;
722 	g_error_provider(pp, 0);
723 
724 	mtx_pool_lock(mtxpool_sleep, dp);
725 	dp->d_init_level = DISK_INIT_DONE;
726 
727 	/*
728 	 * If the disk has gone away at this stage, start the withering
729 	 * process for it.
730 	 */
731 	if (dp->d_goneflag != 0) {
732 		mtx_pool_unlock(mtxpool_sleep, dp);
733 		g_wither_provider(pp, ENXIO);
734 		return;
735 	}
736 	mtx_pool_unlock(mtxpool_sleep, dp);
737 
738 }
739 
740 /*
741  * We get this callback after all of the consumers have gone away, and just
742  * before the provider is freed.  If the disk driver provided a d_gone
743  * callback, let them know that it is okay to free resources -- they won't
744  * be getting any more accesses from GEOM.
745  */
746 static void
747 g_disk_providergone(struct g_provider *pp)
748 {
749 	struct disk *dp;
750 	struct g_disk_softc *sc;
751 
752 	sc = (struct g_disk_softc *)pp->private;
753 	dp = sc->dp;
754 	if (dp != NULL && dp->d_gone != NULL)
755 		dp->d_gone(dp);
756 	if (sc->sysctl_tree != NULL) {
757 		sysctl_ctx_free(&sc->sysctl_ctx);
758 		sc->sysctl_tree = NULL;
759 	}
760 	if (sc->led[0] != 0) {
761 		led_set(sc->led, "0");
762 		sc->led[0] = 0;
763 	}
764 	pp->private = NULL;
765 	pp->geom->softc = NULL;
766 	mtx_destroy(&sc->done_mtx);
767 	mtx_destroy(&sc->start_mtx);
768 	g_free(sc);
769 }
770 
771 static void
772 g_disk_destroy(void *ptr, int flag)
773 {
774 	struct disk *dp;
775 	struct g_geom *gp;
776 	struct g_disk_softc *sc;
777 
778 	g_topology_assert();
779 	dp = ptr;
780 	gp = dp->d_geom;
781 	if (gp != NULL) {
782 		sc = gp->softc;
783 		if (sc != NULL)
784 			sc->dp = NULL;
785 		dp->d_geom = NULL;
786 		g_wither_geom(gp, ENXIO);
787 	}
788 
789 	g_free(dp);
790 }
791 
792 /*
793  * We only allow printable characters in disk ident,
794  * the rest is converted to 'x<HH>'.
795  */
796 static void
797 g_disk_ident_adjust(char *ident, size_t size)
798 {
799 	char *p, tmp[4], newid[DISK_IDENT_SIZE];
800 
801 	newid[0] = '\0';
802 	for (p = ident; *p != '\0'; p++) {
803 		if (isprint(*p)) {
804 			tmp[0] = *p;
805 			tmp[1] = '\0';
806 		} else {
807 			snprintf(tmp, sizeof(tmp), "x%02hhx",
808 			    *(unsigned char *)p);
809 		}
810 		if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
811 			break;
812 	}
813 	bzero(ident, size);
814 	strlcpy(ident, newid, size);
815 }
816 
817 struct disk *
818 disk_alloc(void)
819 {
820 
821 	return (g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO));
822 }
823 
824 void
825 disk_create(struct disk *dp, int version)
826 {
827 
828 	if (version != DISK_VERSION) {
829 		printf("WARNING: Attempt to add disk %s%d %s",
830 		    dp->d_name, dp->d_unit,
831 		    " using incompatible ABI version of disk(9)\n");
832 		printf("WARNING: Ignoring disk %s%d\n",
833 		    dp->d_name, dp->d_unit);
834 		return;
835 	}
836 	if (dp->d_flags & DISKFLAG_RESERVED) {
837 		printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
838 		    dp->d_name, dp->d_unit);
839 		printf("WARNING: Ignoring disk %s%d\n",
840 		    dp->d_name, dp->d_unit);
841 		return;
842 	}
843 	KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
844 	KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
845 	KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
846 	KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
847 	if (dp->d_devstat == NULL)
848 		dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
849 		    dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
850 		    DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
851 	dp->d_geom = NULL;
852 
853 	dp->d_init_level = DISK_INIT_NONE;
854 
855 	g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
856 	g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
857 }
858 
859 void
860 disk_destroy(struct disk *dp)
861 {
862 
863 	g_cancel_event(dp);
864 	dp->d_destroyed = 1;
865 	if (dp->d_devstat != NULL)
866 		devstat_remove_entry(dp->d_devstat);
867 	g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
868 }
869 
870 void
871 disk_gone(struct disk *dp)
872 {
873 	struct g_geom *gp;
874 	struct g_provider *pp;
875 
876 	mtx_pool_lock(mtxpool_sleep, dp);
877 	dp->d_goneflag = 1;
878 
879 	/*
880 	 * If we're still in the process of creating this disk (the
881 	 * g_disk_create() function is still queued, or is in
882 	 * progress), the init level will not yet be DISK_INIT_DONE.
883 	 *
884 	 * If that is the case, g_disk_create() will see d_goneflag
885 	 * and take care of cleaning things up.
886 	 *
887 	 * If the disk has already been created, we default to
888 	 * withering the provider as usual below.
889 	 *
890 	 * If the caller has not set a d_gone() callback, he will
891 	 * not be any worse off by returning here, because the geom
892 	 * has not been fully setup in any case.
893 	 */
894 	if (dp->d_init_level < DISK_INIT_DONE) {
895 		mtx_pool_unlock(mtxpool_sleep, dp);
896 		return;
897 	}
898 	mtx_pool_unlock(mtxpool_sleep, dp);
899 
900 	gp = dp->d_geom;
901 	if (gp != NULL) {
902 		pp = LIST_FIRST(&gp->provider);
903 		if (pp != NULL) {
904 			KASSERT(LIST_NEXT(pp, provider) == NULL,
905 			    ("geom %p has more than one provider", gp));
906 			g_wither_provider(pp, ENXIO);
907 		}
908 	}
909 }
910 
911 void
912 disk_attr_changed(struct disk *dp, const char *attr, int flag)
913 {
914 	struct g_geom *gp;
915 	struct g_provider *pp;
916 	char devnamebuf[128];
917 
918 	gp = dp->d_geom;
919 	if (gp != NULL)
920 		LIST_FOREACH(pp, &gp->provider, provider)
921 			(void)g_attr_changed(pp, attr, flag);
922 	snprintf(devnamebuf, sizeof(devnamebuf), "devname=%s%d", dp->d_name,
923 	    dp->d_unit);
924 	devctl_notify("GEOM", "disk", attr, devnamebuf);
925 }
926 
927 void
928 disk_media_changed(struct disk *dp, int flag)
929 {
930 	struct g_geom *gp;
931 	struct g_provider *pp;
932 
933 	gp = dp->d_geom;
934 	if (gp != NULL) {
935 		pp = LIST_FIRST(&gp->provider);
936 		if (pp != NULL) {
937 			KASSERT(LIST_NEXT(pp, provider) == NULL,
938 			    ("geom %p has more than one provider", gp));
939 			g_media_changed(pp, flag);
940 		}
941 	}
942 }
943 
944 void
945 disk_media_gone(struct disk *dp, int flag)
946 {
947 	struct g_geom *gp;
948 	struct g_provider *pp;
949 
950 	gp = dp->d_geom;
951 	if (gp != NULL) {
952 		pp = LIST_FIRST(&gp->provider);
953 		if (pp != NULL) {
954 			KASSERT(LIST_NEXT(pp, provider) == NULL,
955 			    ("geom %p has more than one provider", gp));
956 			g_media_gone(pp, flag);
957 		}
958 	}
959 }
960 
961 int
962 disk_resize(struct disk *dp, int flag)
963 {
964 
965 	if (dp->d_destroyed || dp->d_geom == NULL)
966 		return (0);
967 
968 	return (g_post_event(g_disk_resize, dp, flag, NULL));
969 }
970 
971 static void
972 g_kern_disks(void *p, int flag __unused)
973 {
974 	struct sbuf *sb;
975 	struct g_geom *gp;
976 	char *sp;
977 
978 	sb = p;
979 	sp = "";
980 	g_topology_assert();
981 	LIST_FOREACH(gp, &g_disk_class.geom, geom) {
982 		sbuf_printf(sb, "%s%s", sp, gp->name);
983 		sp = " ";
984 	}
985 	sbuf_finish(sb);
986 }
987 
988 static int
989 sysctl_disks(SYSCTL_HANDLER_ARGS)
990 {
991 	int error;
992 	struct sbuf *sb;
993 
994 	sb = sbuf_new_auto();
995 	g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
996 	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
997 	sbuf_delete(sb);
998 	return error;
999 }
1000 
1001 SYSCTL_PROC(_kern, OID_AUTO, disks,
1002     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
1003     sysctl_disks, "A", "names of available disks");
1004