xref: /freebsd/sys/geom/geom_disk.c (revision 6be3386466ab79a84b48429ae66244f21526d3df)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2002 Poul-Henning Kamp
5  * Copyright (c) 2002 Networks Associates Technology, Inc.
6  * All rights reserved.
7  *
8  * This software was developed for the FreeBSD Project by Poul-Henning Kamp
9  * and NAI Labs, the Security Research Division of Network Associates, Inc.
10  * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
11  * DARPA CHATS research program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. The names of the authors may not be used to endorse or promote
22  *    products derived from this software without specific prior written
23  *    permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_geom.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/bio.h>
48 #include <sys/ctype.h>
49 #include <sys/devctl.h>
50 #include <sys/fcntl.h>
51 #include <sys/malloc.h>
52 #include <sys/sbuf.h>
53 #include <sys/devicestat.h>
54 
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <geom/geom.h>
58 #include <geom/geom_disk.h>
59 #include <geom/geom_int.h>
60 
61 #include <dev/led/led.h>
62 
63 #include <machine/bus.h>
64 
65 struct g_disk_softc {
66 	struct disk		*dp;
67 	struct devstat		*d_devstat;
68 	struct sysctl_ctx_list	sysctl_ctx;
69 	struct sysctl_oid	*sysctl_tree;
70 	char			led[64];
71 	uint32_t		state;
72 	struct mtx		 done_mtx;
73 };
74 
75 static g_access_t g_disk_access;
76 static g_start_t g_disk_start;
77 static g_ioctl_t g_disk_ioctl;
78 static g_dumpconf_t g_disk_dumpconf;
79 static g_provgone_t g_disk_providergone;
80 
81 static int g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS);
82 
83 static struct g_class g_disk_class = {
84 	.name = G_DISK_CLASS_NAME,
85 	.version = G_VERSION,
86 	.start = g_disk_start,
87 	.access = g_disk_access,
88 	.ioctl = g_disk_ioctl,
89 	.providergone = g_disk_providergone,
90 	.dumpconf = g_disk_dumpconf,
91 };
92 
93 SYSCTL_DECL(_kern_geom);
94 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
95     "GEOM_DISK stuff");
96 
97 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
98 
99 static int
100 g_disk_access(struct g_provider *pp, int r, int w, int e)
101 {
102 	struct disk *dp;
103 	struct g_disk_softc *sc;
104 	int error;
105 
106 	g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
107 	    pp->name, r, w, e);
108 	g_topology_assert();
109 	sc = pp->private;
110 	if ((dp = sc->dp) == NULL || dp->d_destroyed) {
111 		/*
112 		 * Allow decreasing access count even if disk is not
113 		 * available anymore.
114 		 */
115 		if (r <= 0 && w <= 0 && e <= 0)
116 			return (0);
117 		return (ENXIO);
118 	}
119 	r += pp->acr;
120 	w += pp->acw;
121 	e += pp->ace;
122 	error = 0;
123 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
124 		/*
125 		 * It would be better to defer this decision to d_open if
126 		 * it was able to take flags.
127 		 */
128 		if (w > 0 && (dp->d_flags & DISKFLAG_WRITE_PROTECT) != 0)
129 			error = EROFS;
130 		if (error == 0 && dp->d_open != NULL)
131 			error = dp->d_open(dp);
132 		if (bootverbose && error != 0)
133 			printf("Opened disk %s -> %d\n", pp->name, error);
134 		if (error != 0)
135 			return (error);
136 		pp->sectorsize = dp->d_sectorsize;
137 		if (dp->d_maxsize == 0) {
138 			printf("WARNING: Disk drive %s%d has no d_maxsize\n",
139 			    dp->d_name, dp->d_unit);
140 			dp->d_maxsize = DFLTPHYS;
141 		}
142 		if (dp->d_delmaxsize == 0) {
143 			if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
144 				printf("WARNING: Disk drive %s%d has no "
145 				    "d_delmaxsize\n", dp->d_name, dp->d_unit);
146 			}
147 			dp->d_delmaxsize = dp->d_maxsize;
148 		}
149 		pp->stripeoffset = dp->d_stripeoffset;
150 		pp->stripesize = dp->d_stripesize;
151 		dp->d_flags |= DISKFLAG_OPEN;
152 		/*
153 		 * Do not invoke resize event when initial size was zero.
154 		 * Some disks report its size only after first opening.
155 		 */
156 		if (pp->mediasize == 0)
157 			pp->mediasize = dp->d_mediasize;
158 		else
159 			g_resize_provider(pp, dp->d_mediasize);
160 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
161 		if (dp->d_close != NULL) {
162 			error = dp->d_close(dp);
163 			if (error != 0)
164 				printf("Closed disk %s -> %d\n",
165 				    pp->name, error);
166 		}
167 		sc->state = G_STATE_ACTIVE;
168 		if (sc->led[0] != 0)
169 			led_set(sc->led, "0");
170 		dp->d_flags &= ~DISKFLAG_OPEN;
171 	}
172 	return (error);
173 }
174 
175 static void
176 g_disk_kerneldump(struct bio *bp, struct disk *dp)
177 {
178 	struct g_kerneldump *gkd;
179 	struct g_geom *gp;
180 
181 	gkd = (struct g_kerneldump*)bp->bio_data;
182 	gp = bp->bio_to->geom;
183 	g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)",
184 		gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
185 	if (dp->d_dump == NULL) {
186 		g_io_deliver(bp, ENODEV);
187 		return;
188 	}
189 	gkd->di.dumper = dp->d_dump;
190 	gkd->di.priv = dp;
191 	gkd->di.blocksize = dp->d_sectorsize;
192 	gkd->di.maxiosize = dp->d_maxsize;
193 	gkd->di.mediaoffset = gkd->offset;
194 	if ((gkd->offset + gkd->length) > dp->d_mediasize)
195 		gkd->length = dp->d_mediasize - gkd->offset;
196 	gkd->di.mediasize = gkd->length;
197 	g_io_deliver(bp, 0);
198 }
199 
200 static void
201 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
202 {
203 	const char *cmd;
204 
205 	memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
206 	if (sc->led[0] != 0) {
207 		switch (sc->state) {
208 		case G_STATE_FAILED:
209 			cmd = "1";
210 			break;
211 		case G_STATE_REBUILD:
212 			cmd = "f5";
213 			break;
214 		case G_STATE_RESYNC:
215 			cmd = "f1";
216 			break;
217 		default:
218 			cmd = "0";
219 			break;
220 		}
221 		led_set(sc->led, cmd);
222 	}
223 	g_io_deliver(bp, 0);
224 }
225 
226 static void
227 g_disk_done(struct bio *bp)
228 {
229 	struct bintime now;
230 	struct bio *bp2;
231 	struct g_disk_softc *sc;
232 
233 	/* See "notes" for why we need a mutex here */
234 	sc = bp->bio_caller1;
235 	bp2 = bp->bio_parent;
236 	binuptime(&now);
237 	mtx_lock(&sc->done_mtx);
238 	if (bp2->bio_error == 0)
239 		bp2->bio_error = bp->bio_error;
240 	bp2->bio_completed += bp->bio_length - bp->bio_resid;
241 
242 	switch (bp->bio_cmd) {
243 	case BIO_ZONE:
244 		bcopy(&bp->bio_zone, &bp2->bio_zone, sizeof(bp->bio_zone));
245 		/*FALLTHROUGH*/
246 	case BIO_READ:
247 	case BIO_WRITE:
248 	case BIO_DELETE:
249 	case BIO_FLUSH:
250 		devstat_end_transaction_bio_bt(sc->d_devstat, bp, &now);
251 		break;
252 	default:
253 		break;
254 	}
255 	bp2->bio_inbed++;
256 	if (bp2->bio_children == bp2->bio_inbed) {
257 		mtx_unlock(&sc->done_mtx);
258 		bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
259 		g_io_deliver(bp2, bp2->bio_error);
260 	} else
261 		mtx_unlock(&sc->done_mtx);
262 	g_destroy_bio(bp);
263 }
264 
265 static int
266 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
267 {
268 	struct disk *dp;
269 	struct g_disk_softc *sc;
270 
271 	sc = pp->private;
272 	dp = sc->dp;
273 	KASSERT(dp != NULL && !dp->d_destroyed,
274 	    ("g_disk_ioctl(%lx) on destroyed disk %s", cmd, pp->name));
275 
276 	if (dp->d_ioctl == NULL)
277 		return (ENOIOCTL);
278 	return (dp->d_ioctl(dp, cmd, data, fflag, td));
279 }
280 
281 static off_t
282 g_disk_maxsize(struct disk *dp, struct bio *bp)
283 {
284 	if (bp->bio_cmd == BIO_DELETE)
285 		return (dp->d_delmaxsize);
286 	return (dp->d_maxsize);
287 }
288 
289 static int
290 g_disk_maxsegs(struct disk *dp, struct bio *bp)
291 {
292 	return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1);
293 }
294 
295 static void
296 g_disk_advance(struct disk *dp, struct bio *bp, off_t off)
297 {
298 
299 	bp->bio_offset += off;
300 	bp->bio_length -= off;
301 
302 	if ((bp->bio_flags & BIO_VLIST) != 0) {
303 		bus_dma_segment_t *seg, *end;
304 
305 		seg = (bus_dma_segment_t *)bp->bio_data;
306 		end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
307 		off += bp->bio_ma_offset;
308 		while (off >= seg->ds_len) {
309 			KASSERT((seg != end),
310 			    ("vlist request runs off the end"));
311 			off -= seg->ds_len;
312 			seg++;
313 		}
314 		bp->bio_ma_offset = off;
315 		bp->bio_ma_n = end - seg;
316 		bp->bio_data = (void *)seg;
317 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
318 		bp->bio_ma += off / PAGE_SIZE;
319 		bp->bio_ma_offset += off;
320 		bp->bio_ma_offset %= PAGE_SIZE;
321 		bp->bio_ma_n -= off / PAGE_SIZE;
322 	} else {
323 		bp->bio_data += off;
324 	}
325 }
326 
327 static void
328 g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset,
329     off_t *plength, int *ppages)
330 {
331 	uintptr_t seg_page_base;
332 	uintptr_t seg_page_end;
333 	off_t offset;
334 	off_t length;
335 	int seg_pages;
336 
337 	offset = *poffset;
338 	length = *plength;
339 
340 	if (length > seg->ds_len - offset)
341 		length = seg->ds_len - offset;
342 
343 	seg_page_base = trunc_page(seg->ds_addr + offset);
344 	seg_page_end  = round_page(seg->ds_addr + offset + length);
345 	seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT;
346 
347 	if (seg_pages > *ppages) {
348 		seg_pages = *ppages;
349 		length = (seg_page_base + (seg_pages << PAGE_SHIFT)) -
350 		    (seg->ds_addr + offset);
351 	}
352 
353 	*poffset = 0;
354 	*plength -= length;
355 	*ppages -= seg_pages;
356 }
357 
358 static off_t
359 g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg)
360 {
361 	bus_dma_segment_t *seg, *end;
362 	off_t residual;
363 	off_t offset;
364 	int pages;
365 
366 	seg = (bus_dma_segment_t *)bp->bio_data;
367 	end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
368 	residual = bp->bio_length;
369 	offset = bp->bio_ma_offset;
370 	pages = g_disk_maxsegs(dp, bp);
371 	while (residual != 0 && pages != 0) {
372 		KASSERT((seg != end),
373 		    ("vlist limit runs off the end"));
374 		g_disk_seg_limit(seg, &offset, &residual, &pages);
375 		seg++;
376 	}
377 	if (pendseg != NULL)
378 		*pendseg = seg;
379 	return (residual);
380 }
381 
382 static bool
383 g_disk_limit(struct disk *dp, struct bio *bp)
384 {
385 	bool limited = false;
386 	off_t maxsz;
387 
388 	maxsz = g_disk_maxsize(dp, bp);
389 
390 	/*
391 	 * XXX: If we have a stripesize we should really use it here.
392 	 *      Care should be taken in the delete case if this is done
393 	 *      as deletes can be very sensitive to size given how they
394 	 *      are processed.
395 	 */
396 	if (bp->bio_length > maxsz) {
397 		bp->bio_length = maxsz;
398 		limited = true;
399 	}
400 
401 	if ((bp->bio_flags & BIO_VLIST) != 0) {
402 		bus_dma_segment_t *firstseg, *endseg;
403 		off_t residual;
404 
405 		firstseg = (bus_dma_segment_t*)bp->bio_data;
406 		residual = g_disk_vlist_limit(dp, bp, &endseg);
407 		if (residual != 0) {
408 			bp->bio_ma_n = endseg - firstseg;
409 			bp->bio_length -= residual;
410 			limited = true;
411 		}
412 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
413 		bp->bio_ma_n =
414 		    howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE);
415 	}
416 
417 	return (limited);
418 }
419 
420 static void
421 g_disk_start(struct bio *bp)
422 {
423 	struct bio *bp2, *bp3;
424 	struct disk *dp;
425 	struct g_disk_softc *sc;
426 	int error;
427 	off_t off;
428 
429 	biotrack(bp, __func__);
430 
431 	sc = bp->bio_to->private;
432 	dp = sc->dp;
433 	KASSERT(dp != NULL && !dp->d_destroyed,
434 	    ("g_disk_start(%p) on destroyed disk %s", bp, bp->bio_to->name));
435 	error = EJUSTRETURN;
436 	switch(bp->bio_cmd) {
437 	case BIO_DELETE:
438 		if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
439 			error = EOPNOTSUPP;
440 			break;
441 		}
442 		/* fall-through */
443 	case BIO_READ:
444 	case BIO_WRITE:
445 		KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 ||
446 		    (bp->bio_flags & BIO_UNMAPPED) == 0,
447 		    ("unmapped bio not supported by disk %s", dp->d_name));
448 		off = 0;
449 		bp3 = NULL;
450 		bp2 = g_clone_bio(bp);
451 		if (bp2 == NULL) {
452 			error = ENOMEM;
453 			break;
454 		}
455 		for (;;) {
456 			if (g_disk_limit(dp, bp2)) {
457 				off += bp2->bio_length;
458 
459 				/*
460 				 * To avoid a race, we need to grab the next bio
461 				 * before we schedule this one.  See "notes".
462 				 */
463 				bp3 = g_clone_bio(bp);
464 				if (bp3 == NULL)
465 					bp->bio_error = ENOMEM;
466 			}
467 			bp2->bio_done = g_disk_done;
468 			bp2->bio_caller1 = sc;
469 			bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
470 			bp2->bio_bcount = bp2->bio_length;
471 			bp2->bio_disk = dp;
472 			devstat_start_transaction_bio(dp->d_devstat, bp2);
473 			dp->d_strategy(bp2);
474 
475 			if (bp3 == NULL)
476 				break;
477 
478 			bp2 = bp3;
479 			bp3 = NULL;
480 			g_disk_advance(dp, bp2, off);
481 		}
482 		break;
483 	case BIO_GETATTR:
484 		/* Give the driver a chance to override */
485 		if (dp->d_getattr != NULL) {
486 			if (bp->bio_disk == NULL)
487 				bp->bio_disk = dp;
488 			error = dp->d_getattr(bp);
489 			if (error != -1)
490 				break;
491 			error = EJUSTRETURN;
492 		}
493 		if (g_handleattr_int(bp, "GEOM::candelete",
494 		    (dp->d_flags & DISKFLAG_CANDELETE) != 0))
495 			break;
496 		else if (g_handleattr_int(bp, "GEOM::fwsectors",
497 		    dp->d_fwsectors))
498 			break;
499 		else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
500 			break;
501 		else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
502 			break;
503 		else if (g_handleattr_str(bp, "GEOM::descr", dp->d_descr))
504 			break;
505 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
506 		    dp->d_hba_vendor))
507 			break;
508 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
509 		    dp->d_hba_device))
510 			break;
511 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
512 		    dp->d_hba_subvendor))
513 			break;
514 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
515 		    dp->d_hba_subdevice))
516 			break;
517 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
518 			g_disk_kerneldump(bp, dp);
519 		else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
520 			g_disk_setstate(bp, sc);
521 		else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
522 		    dp->d_rotation_rate))
523 			break;
524 		else if (g_handleattr_str(bp, "GEOM::attachment",
525 		    dp->d_attachment))
526 			break;
527 		else
528 			error = ENOIOCTL;
529 		break;
530 	case BIO_FLUSH:
531 		g_trace(G_T_BIO, "g_disk_flushcache(%s)",
532 		    bp->bio_to->name);
533 		if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
534 			error = EOPNOTSUPP;
535 			break;
536 		}
537 		/*FALLTHROUGH*/
538 	case BIO_ZONE:
539 		if (bp->bio_cmd == BIO_ZONE) {
540 			if (!(dp->d_flags & DISKFLAG_CANZONE)) {
541 				error = EOPNOTSUPP;
542 				break;
543 			}
544 			g_trace(G_T_BIO, "g_disk_zone(%s)",
545 			    bp->bio_to->name);
546 		}
547 		bp2 = g_clone_bio(bp);
548 		if (bp2 == NULL) {
549 			g_io_deliver(bp, ENOMEM);
550 			return;
551 		}
552 		bp2->bio_done = g_disk_done;
553 		bp2->bio_caller1 = sc;
554 		bp2->bio_disk = dp;
555 		devstat_start_transaction_bio(dp->d_devstat, bp2);
556 		dp->d_strategy(bp2);
557 		break;
558 	case BIO_SPEEDUP:
559 		bp2 = g_clone_bio(bp);
560 		if (bp2 == NULL) {
561 			g_io_deliver(bp, ENOMEM);
562 			return;
563 		}
564 		bp2->bio_done = g_disk_done;
565 		bp2->bio_caller1 = sc;
566 		bp2->bio_disk = dp;
567 		dp->d_strategy(bp2);
568 		break;
569 	default:
570 		error = EOPNOTSUPP;
571 		break;
572 	}
573 	if (error != EJUSTRETURN)
574 		g_io_deliver(bp, error);
575 	return;
576 }
577 
578 static void
579 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
580 {
581 	struct bio *bp;
582 	struct disk *dp;
583 	struct g_disk_softc *sc;
584 	char *buf;
585 	int res = 0;
586 
587 	sc = gp->softc;
588 	if (sc == NULL || (dp = sc->dp) == NULL)
589 		return;
590 	if (indent == NULL) {
591 		sbuf_printf(sb, " hd %u", dp->d_fwheads);
592 		sbuf_printf(sb, " sc %u", dp->d_fwsectors);
593 		return;
594 	}
595 	if (pp != NULL) {
596 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
597 		    indent, dp->d_fwheads);
598 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
599 		    indent, dp->d_fwsectors);
600 
601 		/*
602 		 * "rotationrate" is a little complicated, because the value
603 		 * returned by the drive might not be the RPM; 0 and 1 are
604 		 * special cases, and there's also a valid range.
605 		 */
606 		sbuf_printf(sb, "%s<rotationrate>", indent);
607 		if (dp->d_rotation_rate == DISK_RR_UNKNOWN) /* Old drives */
608 			sbuf_cat(sb, "unknown");	/* don't report RPM. */
609 		else if (dp->d_rotation_rate == DISK_RR_NON_ROTATING)
610 			sbuf_cat(sb, "0");
611 		else if ((dp->d_rotation_rate >= DISK_RR_MIN) &&
612 		    (dp->d_rotation_rate <= DISK_RR_MAX))
613 			sbuf_printf(sb, "%u", dp->d_rotation_rate);
614 		else
615 			sbuf_cat(sb, "invalid");
616 		sbuf_cat(sb, "</rotationrate>\n");
617 		if (dp->d_getattr != NULL) {
618 			buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
619 			bp = g_alloc_bio();
620 			bp->bio_disk = dp;
621 			bp->bio_attribute = "GEOM::ident";
622 			bp->bio_length = DISK_IDENT_SIZE;
623 			bp->bio_data = buf;
624 			res = dp->d_getattr(bp);
625 			sbuf_printf(sb, "%s<ident>", indent);
626 			g_conf_cat_escaped(sb, res == 0 ? buf : dp->d_ident);
627 			sbuf_cat(sb, "</ident>\n");
628 			bp->bio_attribute = "GEOM::lunid";
629 			bp->bio_length = DISK_IDENT_SIZE;
630 			bp->bio_data = buf;
631 			if (dp->d_getattr(bp) == 0) {
632 				sbuf_printf(sb, "%s<lunid>", indent);
633 				g_conf_cat_escaped(sb, buf);
634 				sbuf_cat(sb, "</lunid>\n");
635 			}
636 			bp->bio_attribute = "GEOM::lunname";
637 			bp->bio_length = DISK_IDENT_SIZE;
638 			bp->bio_data = buf;
639 			if (dp->d_getattr(bp) == 0) {
640 				sbuf_printf(sb, "%s<lunname>", indent);
641 				g_conf_cat_escaped(sb, buf);
642 				sbuf_cat(sb, "</lunname>\n");
643 			}
644 			g_destroy_bio(bp);
645 			g_free(buf);
646 		} else {
647 			sbuf_printf(sb, "%s<ident>", indent);
648 			g_conf_cat_escaped(sb, dp->d_ident);
649 			sbuf_cat(sb, "</ident>\n");
650 		}
651 		sbuf_printf(sb, "%s<descr>", indent);
652 		g_conf_cat_escaped(sb, dp->d_descr);
653 		sbuf_cat(sb, "</descr>\n");
654 	}
655 }
656 
657 static void
658 g_disk_resize(void *ptr, int flag)
659 {
660 	struct disk *dp;
661 	struct g_geom *gp;
662 	struct g_provider *pp;
663 
664 	if (flag == EV_CANCEL)
665 		return;
666 	g_topology_assert();
667 
668 	dp = ptr;
669 	gp = dp->d_geom;
670 
671 	if (dp->d_destroyed || gp == NULL)
672 		return;
673 
674 	LIST_FOREACH(pp, &gp->provider, provider) {
675 		if (pp->sectorsize != 0 &&
676 		    pp->sectorsize != dp->d_sectorsize)
677 			g_wither_provider(pp, ENXIO);
678 		else
679 			g_resize_provider(pp, dp->d_mediasize);
680 	}
681 }
682 
683 static void
684 g_disk_create(void *arg, int flag)
685 {
686 	struct g_geom *gp;
687 	struct g_provider *pp;
688 	struct disk *dp;
689 	struct g_disk_softc *sc;
690 	struct disk_alias *dap;
691 	char tmpstr[80];
692 
693 	if (flag == EV_CANCEL)
694 		return;
695 	g_topology_assert();
696 	dp = arg;
697 
698 	mtx_pool_lock(mtxpool_sleep, dp);
699 	dp->d_init_level = DISK_INIT_START;
700 
701 	/*
702 	 * If the disk has already gone away, we can just stop here and
703 	 * call the user's callback to tell him we've cleaned things up.
704 	 */
705 	if (dp->d_goneflag != 0) {
706 		mtx_pool_unlock(mtxpool_sleep, dp);
707 		if (dp->d_gone != NULL)
708 			dp->d_gone(dp);
709 		return;
710 	}
711 	mtx_pool_unlock(mtxpool_sleep, dp);
712 
713 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
714 	mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
715 	sc->dp = dp;
716 	sc->d_devstat = dp->d_devstat;
717 	gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
718 	gp->softc = sc;
719 	pp = g_new_providerf(gp, "%s", gp->name);
720 	LIST_FOREACH(dap, &dp->d_aliases, da_next)
721 		g_provider_add_alias(pp, "%s%d", dap->da_alias, dp->d_unit);
722 	devstat_remove_entry(pp->stat);
723 	pp->stat = NULL;
724 	dp->d_devstat->id = pp;
725 	pp->mediasize = dp->d_mediasize;
726 	pp->sectorsize = dp->d_sectorsize;
727 	pp->stripeoffset = dp->d_stripeoffset;
728 	pp->stripesize = dp->d_stripesize;
729 	if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
730 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
731 	if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
732 		pp->flags |= G_PF_DIRECT_SEND;
733 	pp->flags |= G_PF_DIRECT_RECEIVE;
734 	if (bootverbose)
735 		printf("GEOM: new disk %s\n", gp->name);
736 	sysctl_ctx_init(&sc->sysctl_ctx);
737 	snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
738 	sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
739 		SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
740 		CTLFLAG_RD | CTLFLAG_MPSAFE, 0, tmpstr);
741 	if (sc->sysctl_tree != NULL) {
742 		SYSCTL_ADD_STRING(&sc->sysctl_ctx,
743 		    SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
744 		    CTLFLAG_RWTUN, sc->led, sizeof(sc->led),
745 		    "LED name");
746 		SYSCTL_ADD_PROC(&sc->sysctl_ctx,
747 		    SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "flags",
748 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, dp, 0,
749 		    g_disk_sysctl_flags, "A", "Report disk flags");
750 	}
751 	pp->private = sc;
752 	dp->d_geom = gp;
753 	g_error_provider(pp, 0);
754 
755 	mtx_pool_lock(mtxpool_sleep, dp);
756 	dp->d_init_level = DISK_INIT_DONE;
757 
758 	/*
759 	 * If the disk has gone away at this stage, start the withering
760 	 * process for it.
761 	 */
762 	if (dp->d_goneflag != 0) {
763 		mtx_pool_unlock(mtxpool_sleep, dp);
764 		g_wither_provider(pp, ENXIO);
765 		return;
766 	}
767 	mtx_pool_unlock(mtxpool_sleep, dp);
768 
769 }
770 
771 /*
772  * We get this callback after all of the consumers have gone away, and just
773  * before the provider is freed.  If the disk driver provided a d_gone
774  * callback, let them know that it is okay to free resources -- they won't
775  * be getting any more accesses from GEOM.
776  */
777 static void
778 g_disk_providergone(struct g_provider *pp)
779 {
780 	struct disk *dp;
781 	struct g_disk_softc *sc;
782 
783 	sc = (struct g_disk_softc *)pp->private;
784 	dp = sc->dp;
785 	if (dp != NULL && dp->d_gone != NULL)
786 		dp->d_gone(dp);
787 	if (sc->sysctl_tree != NULL) {
788 		sysctl_ctx_free(&sc->sysctl_ctx);
789 		sc->sysctl_tree = NULL;
790 	}
791 	if (sc->led[0] != 0) {
792 		led_set(sc->led, "0");
793 		sc->led[0] = 0;
794 	}
795 	pp->private = NULL;
796 	pp->geom->softc = NULL;
797 	mtx_destroy(&sc->done_mtx);
798 	g_free(sc);
799 }
800 
801 static void
802 g_disk_destroy(void *ptr, int flag)
803 {
804 	struct disk *dp;
805 	struct g_geom *gp;
806 	struct g_disk_softc *sc;
807 	struct disk_alias *dap, *daptmp;
808 
809 	g_topology_assert();
810 	dp = ptr;
811 	gp = dp->d_geom;
812 	if (gp != NULL) {
813 		sc = gp->softc;
814 		if (sc != NULL)
815 			sc->dp = NULL;
816 		dp->d_geom = NULL;
817 		g_wither_geom(gp, ENXIO);
818 	}
819 	LIST_FOREACH_SAFE(dap, &dp->d_aliases, da_next, daptmp)
820 		g_free(dap);
821 
822 	g_free(dp);
823 }
824 
825 /*
826  * We only allow printable characters in disk ident,
827  * the rest is converted to 'x<HH>'.
828  */
829 static void
830 g_disk_ident_adjust(char *ident, size_t size)
831 {
832 	char *p, tmp[4], newid[DISK_IDENT_SIZE];
833 
834 	newid[0] = '\0';
835 	for (p = ident; *p != '\0'; p++) {
836 		if (isprint(*p)) {
837 			tmp[0] = *p;
838 			tmp[1] = '\0';
839 		} else {
840 			snprintf(tmp, sizeof(tmp), "x%02hhx",
841 			    *(unsigned char *)p);
842 		}
843 		if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
844 			break;
845 	}
846 	bzero(ident, size);
847 	strlcpy(ident, newid, size);
848 }
849 
850 struct disk *
851 disk_alloc(void)
852 {
853 	struct disk *dp;
854 
855 	dp = g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO);
856 	LIST_INIT(&dp->d_aliases);
857 	return (dp);
858 }
859 
860 void
861 disk_create(struct disk *dp, int version)
862 {
863 
864 	if (version != DISK_VERSION) {
865 		printf("WARNING: Attempt to add disk %s%d %s",
866 		    dp->d_name, dp->d_unit,
867 		    " using incompatible ABI version of disk(9)\n");
868 		printf("WARNING: Ignoring disk %s%d\n",
869 		    dp->d_name, dp->d_unit);
870 		return;
871 	}
872 	if (dp->d_flags & DISKFLAG_RESERVED) {
873 		printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
874 		    dp->d_name, dp->d_unit);
875 		printf("WARNING: Ignoring disk %s%d\n",
876 		    dp->d_name, dp->d_unit);
877 		return;
878 	}
879 	KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
880 	KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
881 	KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
882 	KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
883 	if (dp->d_devstat == NULL)
884 		dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
885 		    dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
886 		    DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
887 	dp->d_geom = NULL;
888 
889 	dp->d_init_level = DISK_INIT_NONE;
890 
891 	g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
892 	g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
893 }
894 
895 void
896 disk_destroy(struct disk *dp)
897 {
898 
899 	disk_gone(dp);
900 	dp->d_destroyed = 1;
901 	g_cancel_event(dp);
902 	if (dp->d_devstat != NULL)
903 		devstat_remove_entry(dp->d_devstat);
904 	g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
905 }
906 
907 void
908 disk_add_alias(struct disk *dp, const char *name)
909 {
910 	struct disk_alias *dap;
911 
912 	dap = (struct disk_alias *)g_malloc(
913 		sizeof(struct disk_alias) + strlen(name) + 1, M_WAITOK);
914 	strcpy((char *)(dap + 1), name);
915 	dap->da_alias = (const char *)(dap + 1);
916 	LIST_INSERT_HEAD(&dp->d_aliases, dap, da_next);
917 }
918 
919 void
920 disk_gone(struct disk *dp)
921 {
922 	struct g_geom *gp;
923 	struct g_provider *pp;
924 
925 	mtx_pool_lock(mtxpool_sleep, dp);
926 
927 	/*
928 	 * Second wither call makes no sense, plus we can not access the list
929 	 * of providers without topology lock after calling wither once.
930 	 */
931 	if (dp->d_goneflag != 0) {
932 		mtx_pool_unlock(mtxpool_sleep, dp);
933 		return;
934 	}
935 
936 	dp->d_goneflag = 1;
937 
938 	/*
939 	 * If we're still in the process of creating this disk (the
940 	 * g_disk_create() function is still queued, or is in
941 	 * progress), the init level will not yet be DISK_INIT_DONE.
942 	 *
943 	 * If that is the case, g_disk_create() will see d_goneflag
944 	 * and take care of cleaning things up.
945 	 *
946 	 * If the disk has already been created, we default to
947 	 * withering the provider as usual below.
948 	 *
949 	 * If the caller has not set a d_gone() callback, he will
950 	 * not be any worse off by returning here, because the geom
951 	 * has not been fully setup in any case.
952 	 */
953 	if (dp->d_init_level < DISK_INIT_DONE) {
954 		mtx_pool_unlock(mtxpool_sleep, dp);
955 		return;
956 	}
957 	mtx_pool_unlock(mtxpool_sleep, dp);
958 
959 	gp = dp->d_geom;
960 	pp = LIST_FIRST(&gp->provider);
961 	if (pp != NULL) {
962 		KASSERT(LIST_NEXT(pp, provider) == NULL,
963 		    ("geom %p has more than one provider", gp));
964 		g_wither_provider(pp, ENXIO);
965 	}
966 }
967 
968 void
969 disk_attr_changed(struct disk *dp, const char *attr, int flag)
970 {
971 	struct g_geom *gp;
972 	struct g_provider *pp;
973 	char devnamebuf[128];
974 
975 	gp = dp->d_geom;
976 	if (gp != NULL)
977 		LIST_FOREACH(pp, &gp->provider, provider)
978 			(void)g_attr_changed(pp, attr, flag);
979 	snprintf(devnamebuf, sizeof(devnamebuf), "devname=%s%d", dp->d_name,
980 	    dp->d_unit);
981 	devctl_notify("GEOM", "disk", attr, devnamebuf);
982 }
983 
984 void
985 disk_media_changed(struct disk *dp, int flag)
986 {
987 	struct g_geom *gp;
988 	struct g_provider *pp;
989 
990 	gp = dp->d_geom;
991 	if (gp != NULL) {
992 		pp = LIST_FIRST(&gp->provider);
993 		if (pp != NULL) {
994 			KASSERT(LIST_NEXT(pp, provider) == NULL,
995 			    ("geom %p has more than one provider", gp));
996 			g_media_changed(pp, flag);
997 		}
998 	}
999 }
1000 
1001 void
1002 disk_media_gone(struct disk *dp, int flag)
1003 {
1004 	struct g_geom *gp;
1005 	struct g_provider *pp;
1006 
1007 	gp = dp->d_geom;
1008 	if (gp != NULL) {
1009 		pp = LIST_FIRST(&gp->provider);
1010 		if (pp != NULL) {
1011 			KASSERT(LIST_NEXT(pp, provider) == NULL,
1012 			    ("geom %p has more than one provider", gp));
1013 			g_media_gone(pp, flag);
1014 		}
1015 	}
1016 }
1017 
1018 int
1019 disk_resize(struct disk *dp, int flag)
1020 {
1021 
1022 	if (dp->d_destroyed || dp->d_geom == NULL)
1023 		return (0);
1024 
1025 	return (g_post_event(g_disk_resize, dp, flag, NULL));
1026 }
1027 
1028 static void
1029 g_kern_disks(void *p, int flag __unused)
1030 {
1031 	struct sbuf *sb;
1032 	struct g_geom *gp;
1033 	char *sp;
1034 
1035 	sb = p;
1036 	sp = "";
1037 	g_topology_assert();
1038 	LIST_FOREACH(gp, &g_disk_class.geom, geom) {
1039 		sbuf_printf(sb, "%s%s", sp, gp->name);
1040 		sp = " ";
1041 	}
1042 	sbuf_finish(sb);
1043 }
1044 
1045 static int
1046 g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS)
1047 {
1048 	struct disk *dp;
1049 	struct sbuf *sb;
1050 	int error;
1051 
1052 	sb = sbuf_new_auto();
1053 	dp = (struct disk *)arg1;
1054 	sbuf_printf(sb, "%b", dp->d_flags,
1055 		"\20"
1056 		"\2OPEN"
1057 		"\3CANDELETE"
1058 		"\4CANFLUSHCACHE"
1059 		"\5UNMAPPEDBIO"
1060 		"\6DIRECTCOMPLETION"
1061 		"\10CANZONE"
1062 		"\11WRITEPROTECT");
1063 
1064 	sbuf_finish(sb);
1065 	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1066 	sbuf_delete(sb);
1067 	return (error);
1068 }
1069 
1070 static int
1071 sysctl_disks(SYSCTL_HANDLER_ARGS)
1072 {
1073 	int error;
1074 	struct sbuf *sb;
1075 
1076 	sb = sbuf_new_auto();
1077 	g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
1078 	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1079 	sbuf_delete(sb);
1080 	return error;
1081 }
1082 
1083 SYSCTL_PROC(_kern, OID_AUTO, disks,
1084     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
1085     sysctl_disks, "A", "names of available disks");
1086