xref: /freebsd/sys/geom/geom_disk.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
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/msan.h>
53 #include <sys/sbuf.h>
54 #include <sys/devicestat.h>
55 
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <geom/geom.h>
59 #include <geom/geom_disk.h>
60 #include <geom/geom_int.h>
61 
62 #include <dev/led/led.h>
63 
64 #include <machine/bus.h>
65 
66 struct g_disk_softc {
67 	struct disk		*dp;
68 	struct devstat		*d_devstat;
69 	struct sysctl_ctx_list	sysctl_ctx;
70 	struct sysctl_oid	*sysctl_tree;
71 	char			led[64];
72 	uint32_t		state;
73 	struct mtx		 done_mtx;
74 	bool                    flush_notsup_succeed;
75 };
76 
77 static g_access_t g_disk_access;
78 static g_start_t g_disk_start;
79 static g_ioctl_t g_disk_ioctl;
80 static g_dumpconf_t g_disk_dumpconf;
81 static g_provgone_t g_disk_providergone;
82 
83 static int g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS);
84 
85 static struct g_class g_disk_class = {
86 	.name = G_DISK_CLASS_NAME,
87 	.version = G_VERSION,
88 	.start = g_disk_start,
89 	.access = g_disk_access,
90 	.ioctl = g_disk_ioctl,
91 	.providergone = g_disk_providergone,
92 	.dumpconf = g_disk_dumpconf,
93 };
94 
95 SYSCTL_DECL(_kern_geom);
96 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
97     "GEOM_DISK stuff");
98 
99 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
100 
101 static int
102 g_disk_access(struct g_provider *pp, int r, int w, int e)
103 {
104 	struct disk *dp;
105 	struct g_disk_softc *sc;
106 	int error;
107 
108 	g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
109 	    pp->name, r, w, e);
110 	g_topology_assert();
111 	sc = pp->private;
112 	if ((dp = sc->dp) == NULL || dp->d_destroyed) {
113 		/*
114 		 * Allow decreasing access count even if disk is not
115 		 * available anymore.
116 		 */
117 		if (r <= 0 && w <= 0 && e <= 0)
118 			return (0);
119 		return (ENXIO);
120 	}
121 	r += pp->acr;
122 	w += pp->acw;
123 	e += pp->ace;
124 	error = 0;
125 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
126 		/*
127 		 * It would be better to defer this decision to d_open if
128 		 * it was able to take flags.
129 		 */
130 		if (w > 0 && (dp->d_flags & DISKFLAG_WRITE_PROTECT) != 0)
131 			error = EROFS;
132 		if (error == 0 && dp->d_open != NULL)
133 			error = dp->d_open(dp);
134 		if (bootverbose && error != 0)
135 			printf("Opened disk %s -> %d\n", pp->name, error);
136 		if (error != 0)
137 			return (error);
138 		pp->sectorsize = dp->d_sectorsize;
139 		if (dp->d_maxsize == 0) {
140 			printf("WARNING: Disk drive %s%d has no d_maxsize\n",
141 			    dp->d_name, dp->d_unit);
142 			dp->d_maxsize = DFLTPHYS;
143 		}
144 		if (dp->d_delmaxsize == 0) {
145 			if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
146 				printf("WARNING: Disk drive %s%d has no "
147 				    "d_delmaxsize\n", dp->d_name, dp->d_unit);
148 			}
149 			dp->d_delmaxsize = dp->d_maxsize;
150 		}
151 		pp->stripeoffset = dp->d_stripeoffset;
152 		pp->stripesize = dp->d_stripesize;
153 		dp->d_flags |= DISKFLAG_OPEN;
154 		/*
155 		 * Do not invoke resize event when initial size was zero.
156 		 * Some disks report its size only after first opening.
157 		 */
158 		if (pp->mediasize == 0)
159 			pp->mediasize = dp->d_mediasize;
160 		else
161 			g_resize_provider(pp, dp->d_mediasize);
162 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
163 		if (dp->d_close != NULL) {
164 			error = dp->d_close(dp);
165 			if (error != 0)
166 				printf("Closed disk %s -> %d\n",
167 				    pp->name, error);
168 		}
169 		sc->state = G_STATE_ACTIVE;
170 		if (sc->led[0] != 0)
171 			led_set(sc->led, "0");
172 		dp->d_flags &= ~DISKFLAG_OPEN;
173 	}
174 	return (error);
175 }
176 
177 static void
178 g_disk_kerneldump(struct bio *bp, struct disk *dp)
179 {
180 	struct g_kerneldump *gkd;
181 	struct g_geom *gp;
182 
183 	gkd = (struct g_kerneldump*)bp->bio_data;
184 	gp = bp->bio_to->geom;
185 	g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)",
186 		gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
187 	if (dp->d_dump == NULL) {
188 		g_io_deliver(bp, ENODEV);
189 		return;
190 	}
191 	gkd->di.dumper = dp->d_dump;
192 	gkd->di.priv = dp;
193 	gkd->di.blocksize = dp->d_sectorsize;
194 	gkd->di.maxiosize = dp->d_maxsize;
195 	gkd->di.mediaoffset = gkd->offset;
196 	if ((gkd->offset + gkd->length) > dp->d_mediasize)
197 		gkd->length = dp->d_mediasize - gkd->offset;
198 	gkd->di.mediasize = gkd->length;
199 	g_io_deliver(bp, 0);
200 }
201 
202 static void
203 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
204 {
205 	const char *cmd;
206 
207 	memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
208 	if (sc->led[0] != 0) {
209 		switch (sc->state) {
210 		case G_STATE_FAILED:
211 			cmd = "1";
212 			break;
213 		case G_STATE_REBUILD:
214 			cmd = "f5";
215 			break;
216 		case G_STATE_RESYNC:
217 			cmd = "f1";
218 			break;
219 		default:
220 			cmd = "0";
221 			break;
222 		}
223 		led_set(sc->led, cmd);
224 	}
225 	g_io_deliver(bp, 0);
226 }
227 
228 static void
229 g_disk_done(struct bio *bp)
230 {
231 	struct bintime now;
232 	struct bio *bp2;
233 	struct g_disk_softc *sc;
234 
235 	/* See "notes" for why we need a mutex here */
236 	sc = bp->bio_caller1;
237 	bp2 = bp->bio_parent;
238 	binuptime(&now);
239 	mtx_lock(&sc->done_mtx);
240 	if (bp2->bio_error == 0)
241 		bp2->bio_error = bp->bio_error;
242 	bp2->bio_completed += bp->bio_length - bp->bio_resid;
243 
244 	if (bp->bio_cmd == BIO_READ)
245 		kmsan_check(bp2->bio_data, bp2->bio_completed, "g_disk_done");
246 
247 	switch (bp->bio_cmd) {
248 	case BIO_ZONE:
249 		bcopy(&bp->bio_zone, &bp2->bio_zone, sizeof(bp->bio_zone));
250 		/*FALLTHROUGH*/
251 	case BIO_READ:
252 	case BIO_WRITE:
253 	case BIO_DELETE:
254 	case BIO_FLUSH:
255 		devstat_end_transaction_bio_bt(sc->d_devstat, bp, &now);
256 		break;
257 	default:
258 		break;
259 	}
260 	bp2->bio_inbed++;
261 	if (bp2->bio_children == bp2->bio_inbed) {
262 		mtx_unlock(&sc->done_mtx);
263 		bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
264 		g_io_deliver(bp2, bp2->bio_error);
265 	} else
266 		mtx_unlock(&sc->done_mtx);
267 	g_destroy_bio(bp);
268 }
269 
270 static int
271 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
272 {
273 	struct disk *dp;
274 	struct g_disk_softc *sc;
275 
276 	sc = pp->private;
277 	dp = sc->dp;
278 	KASSERT(dp != NULL && !dp->d_destroyed,
279 	    ("g_disk_ioctl(%lx) on destroyed disk %s", cmd, pp->name));
280 
281 	if (dp->d_ioctl == NULL)
282 		return (ENOIOCTL);
283 	return (dp->d_ioctl(dp, cmd, data, fflag, td));
284 }
285 
286 static off_t
287 g_disk_maxsize(struct disk *dp, struct bio *bp)
288 {
289 	if (bp->bio_cmd == BIO_DELETE)
290 		return (dp->d_delmaxsize);
291 	return (dp->d_maxsize);
292 }
293 
294 static int
295 g_disk_maxsegs(struct disk *dp, struct bio *bp)
296 {
297 	return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1);
298 }
299 
300 static void
301 g_disk_advance(struct disk *dp, struct bio *bp, off_t off)
302 {
303 
304 	bp->bio_offset += off;
305 	bp->bio_length -= off;
306 
307 	if ((bp->bio_flags & BIO_VLIST) != 0) {
308 		bus_dma_segment_t *seg, *end;
309 
310 		seg = (bus_dma_segment_t *)bp->bio_data;
311 		end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
312 		off += bp->bio_ma_offset;
313 		while (off >= seg->ds_len) {
314 			KASSERT((seg != end),
315 			    ("vlist request runs off the end"));
316 			off -= seg->ds_len;
317 			seg++;
318 		}
319 		bp->bio_ma_offset = off;
320 		bp->bio_ma_n = end - seg;
321 		bp->bio_data = (void *)seg;
322 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
323 		bp->bio_ma += off / PAGE_SIZE;
324 		bp->bio_ma_offset += off;
325 		bp->bio_ma_offset %= PAGE_SIZE;
326 		bp->bio_ma_n -= off / PAGE_SIZE;
327 	} else {
328 		bp->bio_data += off;
329 	}
330 }
331 
332 static void
333 g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset,
334     off_t *plength, int *ppages)
335 {
336 	uintptr_t seg_page_base;
337 	uintptr_t seg_page_end;
338 	off_t offset;
339 	off_t length;
340 	int seg_pages;
341 
342 	offset = *poffset;
343 	length = *plength;
344 
345 	if (length > seg->ds_len - offset)
346 		length = seg->ds_len - offset;
347 
348 	seg_page_base = trunc_page(seg->ds_addr + offset);
349 	seg_page_end  = round_page(seg->ds_addr + offset + length);
350 	seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT;
351 
352 	if (seg_pages > *ppages) {
353 		seg_pages = *ppages;
354 		length = (seg_page_base + (seg_pages << PAGE_SHIFT)) -
355 		    (seg->ds_addr + offset);
356 	}
357 
358 	*poffset = 0;
359 	*plength -= length;
360 	*ppages -= seg_pages;
361 }
362 
363 static off_t
364 g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg)
365 {
366 	bus_dma_segment_t *seg, *end __diagused;
367 	off_t residual;
368 	off_t offset;
369 	int pages;
370 
371 	seg = (bus_dma_segment_t *)bp->bio_data;
372 	end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
373 	residual = bp->bio_length;
374 	offset = bp->bio_ma_offset;
375 	pages = g_disk_maxsegs(dp, bp);
376 	while (residual != 0 && pages != 0) {
377 		KASSERT((seg != end),
378 		    ("vlist limit runs off the end"));
379 		g_disk_seg_limit(seg, &offset, &residual, &pages);
380 		seg++;
381 	}
382 	if (pendseg != NULL)
383 		*pendseg = seg;
384 	return (residual);
385 }
386 
387 static bool
388 g_disk_limit(struct disk *dp, struct bio *bp)
389 {
390 	bool limited = false;
391 	off_t maxsz;
392 
393 	maxsz = g_disk_maxsize(dp, bp);
394 
395 	/*
396 	 * XXX: If we have a stripesize we should really use it here.
397 	 *      Care should be taken in the delete case if this is done
398 	 *      as deletes can be very sensitive to size given how they
399 	 *      are processed.
400 	 */
401 	if (bp->bio_length > maxsz) {
402 		bp->bio_length = maxsz;
403 		limited = true;
404 	}
405 
406 	if ((bp->bio_flags & BIO_VLIST) != 0) {
407 		bus_dma_segment_t *firstseg, *endseg;
408 		off_t residual;
409 
410 		firstseg = (bus_dma_segment_t*)bp->bio_data;
411 		residual = g_disk_vlist_limit(dp, bp, &endseg);
412 		if (residual != 0) {
413 			bp->bio_ma_n = endseg - firstseg;
414 			bp->bio_length -= residual;
415 			limited = true;
416 		}
417 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
418 		bp->bio_ma_n =
419 		    howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE);
420 	}
421 
422 	return (limited);
423 }
424 
425 static void
426 g_disk_start(struct bio *bp)
427 {
428 	struct bio *bp2, *bp3;
429 	struct disk *dp;
430 	struct g_disk_softc *sc;
431 	int error;
432 	off_t off;
433 
434 	biotrack(bp, __func__);
435 
436 	sc = bp->bio_to->private;
437 	dp = sc->dp;
438 	KASSERT(dp != NULL && !dp->d_destroyed,
439 	    ("g_disk_start(%p) on destroyed disk %s", bp, bp->bio_to->name));
440 	error = EJUSTRETURN;
441 	switch(bp->bio_cmd) {
442 	case BIO_DELETE:
443 		if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
444 			error = EOPNOTSUPP;
445 			break;
446 		}
447 		/* fall-through */
448 	case BIO_READ:
449 	case BIO_WRITE:
450 		KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 ||
451 		    (bp->bio_flags & BIO_UNMAPPED) == 0,
452 		    ("unmapped bio not supported by disk %s", dp->d_name));
453 
454 		if (bp->bio_cmd == BIO_WRITE)
455 			kmsan_check_bio(bp, "g_disk_start");
456 
457 		off = 0;
458 		bp3 = NULL;
459 		bp2 = g_clone_bio(bp);
460 		if (bp2 == NULL) {
461 			error = ENOMEM;
462 			break;
463 		}
464 		for (;;) {
465 			if (g_disk_limit(dp, bp2)) {
466 				off += bp2->bio_length;
467 
468 				/*
469 				 * To avoid a race, we need to grab the next bio
470 				 * before we schedule this one.  See "notes".
471 				 */
472 				bp3 = g_clone_bio(bp);
473 				if (bp3 == NULL)
474 					bp->bio_error = ENOMEM;
475 			}
476 			bp2->bio_done = g_disk_done;
477 			bp2->bio_caller1 = sc;
478 			bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
479 			bp2->bio_bcount = bp2->bio_length;
480 			bp2->bio_disk = dp;
481 			devstat_start_transaction_bio(dp->d_devstat, bp2);
482 			dp->d_strategy(bp2);
483 
484 			if (bp3 == NULL)
485 				break;
486 
487 			bp2 = bp3;
488 			bp3 = NULL;
489 			g_disk_advance(dp, bp2, off);
490 		}
491 		break;
492 	case BIO_GETATTR:
493 		/* Give the driver a chance to override */
494 		if (dp->d_getattr != NULL) {
495 			if (bp->bio_disk == NULL)
496 				bp->bio_disk = dp;
497 			error = dp->d_getattr(bp);
498 			if (error != -1)
499 				break;
500 			error = EJUSTRETURN;
501 		}
502 		if (g_handleattr_int(bp, "GEOM::candelete",
503 		    (dp->d_flags & DISKFLAG_CANDELETE) != 0))
504 			break;
505 		else if (g_handleattr_int(bp, "GEOM::fwsectors",
506 		    dp->d_fwsectors))
507 			break;
508 		else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
509 			break;
510 		else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
511 			break;
512 		else if (g_handleattr_str(bp, "GEOM::descr", dp->d_descr))
513 			break;
514 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
515 		    dp->d_hba_vendor))
516 			break;
517 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
518 		    dp->d_hba_device))
519 			break;
520 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
521 		    dp->d_hba_subvendor))
522 			break;
523 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
524 		    dp->d_hba_subdevice))
525 			break;
526 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
527 			g_disk_kerneldump(bp, dp);
528 		else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
529 			g_disk_setstate(bp, sc);
530 		else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
531 		    dp->d_rotation_rate))
532 			break;
533 		else if (g_handleattr_str(bp, "GEOM::attachment",
534 		    dp->d_attachment))
535 			break;
536 		else
537 			error = ENOIOCTL;
538 		break;
539 	case BIO_FLUSH:
540 		g_trace(G_T_BIO, "g_disk_flushcache(%s)",
541 		    bp->bio_to->name);
542 		if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
543 			error = (sc->flush_notsup_succeed) ? 0 : EOPNOTSUPP;
544 			break;
545 		}
546 		/*FALLTHROUGH*/
547 	case BIO_ZONE:
548 		if (bp->bio_cmd == BIO_ZONE) {
549 			if (!(dp->d_flags & DISKFLAG_CANZONE)) {
550 				error = EOPNOTSUPP;
551 				break;
552 			}
553 			g_trace(G_T_BIO, "g_disk_zone(%s)",
554 			    bp->bio_to->name);
555 		}
556 		bp2 = g_clone_bio(bp);
557 		if (bp2 == NULL) {
558 			g_io_deliver(bp, ENOMEM);
559 			return;
560 		}
561 		bp2->bio_done = g_disk_done;
562 		bp2->bio_caller1 = sc;
563 		bp2->bio_disk = dp;
564 		devstat_start_transaction_bio(dp->d_devstat, bp2);
565 		dp->d_strategy(bp2);
566 		break;
567 	case BIO_SPEEDUP:
568 		bp2 = g_clone_bio(bp);
569 		if (bp2 == NULL) {
570 			g_io_deliver(bp, ENOMEM);
571 			return;
572 		}
573 		bp2->bio_done = g_disk_done;
574 		bp2->bio_caller1 = sc;
575 		bp2->bio_disk = dp;
576 		dp->d_strategy(bp2);
577 		break;
578 	default:
579 		error = EOPNOTSUPP;
580 		break;
581 	}
582 	if (error != EJUSTRETURN)
583 		g_io_deliver(bp, error);
584 	return;
585 }
586 
587 static void
588 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
589 {
590 	struct bio *bp;
591 	struct disk *dp;
592 	struct g_disk_softc *sc;
593 	char *buf;
594 	int res = 0;
595 
596 	sc = gp->softc;
597 	if (sc == NULL || (dp = sc->dp) == NULL)
598 		return;
599 	if (indent == NULL) {
600 		sbuf_printf(sb, " hd %u", dp->d_fwheads);
601 		sbuf_printf(sb, " sc %u", dp->d_fwsectors);
602 		return;
603 	}
604 	if (pp != NULL) {
605 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
606 		    indent, dp->d_fwheads);
607 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
608 		    indent, dp->d_fwsectors);
609 
610 		/*
611 		 * "rotationrate" is a little complicated, because the value
612 		 * returned by the drive might not be the RPM; 0 and 1 are
613 		 * special cases, and there's also a valid range.
614 		 */
615 		sbuf_printf(sb, "%s<rotationrate>", indent);
616 		if (dp->d_rotation_rate == DISK_RR_UNKNOWN) /* Old drives */
617 			sbuf_cat(sb, "unknown");	/* don't report RPM. */
618 		else if (dp->d_rotation_rate == DISK_RR_NON_ROTATING)
619 			sbuf_cat(sb, "0");
620 		else if ((dp->d_rotation_rate >= DISK_RR_MIN) &&
621 		    (dp->d_rotation_rate <= DISK_RR_MAX))
622 			sbuf_printf(sb, "%u", dp->d_rotation_rate);
623 		else
624 			sbuf_cat(sb, "invalid");
625 		sbuf_cat(sb, "</rotationrate>\n");
626 		if (dp->d_getattr != NULL) {
627 			buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
628 			bp = g_alloc_bio();
629 			bp->bio_disk = dp;
630 			bp->bio_attribute = "GEOM::ident";
631 			bp->bio_length = DISK_IDENT_SIZE;
632 			bp->bio_data = buf;
633 			res = dp->d_getattr(bp);
634 			sbuf_printf(sb, "%s<ident>", indent);
635 			g_conf_cat_escaped(sb, res == 0 ? buf : dp->d_ident);
636 			sbuf_cat(sb, "</ident>\n");
637 			bp->bio_attribute = "GEOM::lunid";
638 			bp->bio_length = DISK_IDENT_SIZE;
639 			bp->bio_data = buf;
640 			if (dp->d_getattr(bp) == 0) {
641 				sbuf_printf(sb, "%s<lunid>", indent);
642 				g_conf_cat_escaped(sb, buf);
643 				sbuf_cat(sb, "</lunid>\n");
644 			}
645 			bp->bio_attribute = "GEOM::lunname";
646 			bp->bio_length = DISK_IDENT_SIZE;
647 			bp->bio_data = buf;
648 			if (dp->d_getattr(bp) == 0) {
649 				sbuf_printf(sb, "%s<lunname>", indent);
650 				g_conf_cat_escaped(sb, buf);
651 				sbuf_cat(sb, "</lunname>\n");
652 			}
653 			g_destroy_bio(bp);
654 			g_free(buf);
655 		} else {
656 			sbuf_printf(sb, "%s<ident>", indent);
657 			g_conf_cat_escaped(sb, dp->d_ident);
658 			sbuf_cat(sb, "</ident>\n");
659 		}
660 		sbuf_printf(sb, "%s<descr>", indent);
661 		g_conf_cat_escaped(sb, dp->d_descr);
662 		sbuf_cat(sb, "</descr>\n");
663 	}
664 }
665 
666 static void
667 g_disk_resize(void *ptr, int flag)
668 {
669 	struct disk *dp;
670 	struct g_geom *gp;
671 	struct g_provider *pp;
672 
673 	if (flag == EV_CANCEL)
674 		return;
675 	g_topology_assert();
676 
677 	dp = ptr;
678 	gp = dp->d_geom;
679 
680 	if (dp->d_destroyed || gp == NULL)
681 		return;
682 
683 	LIST_FOREACH(pp, &gp->provider, provider) {
684 		if (pp->sectorsize != 0 &&
685 		    pp->sectorsize != dp->d_sectorsize)
686 			g_wither_provider(pp, ENXIO);
687 		else
688 			g_resize_provider(pp, dp->d_mediasize);
689 	}
690 }
691 
692 static void
693 g_disk_create(void *arg, int flag)
694 {
695 	struct g_geom *gp;
696 	struct g_provider *pp;
697 	struct disk *dp;
698 	struct g_disk_softc *sc;
699 	struct disk_alias *dap;
700 	char tmpstr[80];
701 
702 	if (flag == EV_CANCEL)
703 		return;
704 	g_topology_assert();
705 	dp = arg;
706 
707 	mtx_pool_lock(mtxpool_sleep, dp);
708 	dp->d_init_level = DISK_INIT_START;
709 
710 	/*
711 	 * If the disk has already gone away, we can just stop here and
712 	 * call the user's callback to tell him we've cleaned things up.
713 	 */
714 	if (dp->d_goneflag != 0) {
715 		mtx_pool_unlock(mtxpool_sleep, dp);
716 		if (dp->d_gone != NULL)
717 			dp->d_gone(dp);
718 		return;
719 	}
720 	mtx_pool_unlock(mtxpool_sleep, dp);
721 
722 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
723 	mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
724 	sc->dp = dp;
725 	if (dp->d_devstat == NULL) {
726 		dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
727 		    dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
728 		    DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
729 	}
730 	sc->d_devstat = dp->d_devstat;
731 	gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
732 	gp->softc = sc;
733 	pp = g_new_providerf(gp, "%s", gp->name);
734 	LIST_FOREACH(dap, &dp->d_aliases, da_next)
735 		g_provider_add_alias(pp, "%s%d", dap->da_alias, dp->d_unit);
736 	devstat_remove_entry(pp->stat);
737 	pp->stat = NULL;
738 	dp->d_devstat->id = pp;
739 	pp->mediasize = dp->d_mediasize;
740 	pp->sectorsize = dp->d_sectorsize;
741 	pp->stripeoffset = dp->d_stripeoffset;
742 	pp->stripesize = dp->d_stripesize;
743 	if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
744 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
745 	if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
746 		pp->flags |= G_PF_DIRECT_SEND;
747 	pp->flags |= G_PF_DIRECT_RECEIVE;
748 	if (bootverbose)
749 		printf("GEOM: new disk %s\n", gp->name);
750 	sysctl_ctx_init(&sc->sysctl_ctx);
751 	snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
752 	sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
753 		SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
754 		CTLFLAG_RD | CTLFLAG_MPSAFE, 0, tmpstr);
755 	if (sc->sysctl_tree != NULL) {
756 		SYSCTL_ADD_STRING(&sc->sysctl_ctx,
757 		    SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
758 		    CTLFLAG_RWTUN, sc->led, sizeof(sc->led),
759 		    "LED name");
760 		SYSCTL_ADD_PROC(&sc->sysctl_ctx,
761 		    SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "flags",
762 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, dp, 0,
763 		    g_disk_sysctl_flags, "A", "Report disk flags");
764 		SYSCTL_ADD_BOOL(&sc->sysctl_ctx,
765 		    SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "flush_notsup_succeed",
766 		    CTLFLAG_RWTUN, &sc->flush_notsup_succeed, sizeof(sc->flush_notsup_succeed),
767 		    "Do not return EOPNOTSUPP if there is no cache to flush");
768 	}
769 	pp->private = sc;
770 	dp->d_geom = gp;
771 	g_error_provider(pp, 0);
772 
773 	mtx_pool_lock(mtxpool_sleep, dp);
774 	dp->d_init_level = DISK_INIT_DONE;
775 
776 	/*
777 	 * If the disk has gone away at this stage, start the withering
778 	 * process for it.
779 	 */
780 	if (dp->d_goneflag != 0) {
781 		mtx_pool_unlock(mtxpool_sleep, dp);
782 		g_wither_provider(pp, ENXIO);
783 		return;
784 	}
785 	mtx_pool_unlock(mtxpool_sleep, dp);
786 
787 }
788 
789 /*
790  * We get this callback after all of the consumers have gone away, and just
791  * before the provider is freed.  If the disk driver provided a d_gone
792  * callback, let them know that it is okay to free resources -- they won't
793  * be getting any more accesses from GEOM.
794  */
795 static void
796 g_disk_providergone(struct g_provider *pp)
797 {
798 	struct disk *dp;
799 	struct g_disk_softc *sc;
800 
801 	sc = (struct g_disk_softc *)pp->private;
802 	dp = sc->dp;
803 	if (dp != NULL && dp->d_gone != NULL)
804 		dp->d_gone(dp);
805 	if (sc->sysctl_tree != NULL) {
806 		sysctl_ctx_free(&sc->sysctl_ctx);
807 		sc->sysctl_tree = NULL;
808 	}
809 	if (sc->led[0] != 0) {
810 		led_set(sc->led, "0");
811 		sc->led[0] = 0;
812 	}
813 	pp->private = NULL;
814 	pp->geom->softc = NULL;
815 	mtx_destroy(&sc->done_mtx);
816 	g_free(sc);
817 }
818 
819 static void
820 g_disk_destroy(void *ptr, int flag)
821 {
822 	struct disk *dp;
823 	struct g_geom *gp;
824 	struct g_disk_softc *sc;
825 	struct disk_alias *dap, *daptmp;
826 
827 	g_topology_assert();
828 	dp = ptr;
829 	gp = dp->d_geom;
830 	if (gp != NULL) {
831 		sc = gp->softc;
832 		if (sc != NULL)
833 			sc->dp = NULL;
834 		dp->d_geom = NULL;
835 		g_wither_geom(gp, ENXIO);
836 	}
837 	LIST_FOREACH_SAFE(dap, &dp->d_aliases, da_next, daptmp)
838 		g_free(dap);
839 
840 	g_free(dp);
841 }
842 
843 /*
844  * We only allow printable characters in disk ident,
845  * the rest is converted to 'x<HH>'.
846  */
847 static void
848 g_disk_ident_adjust(char *ident, size_t size)
849 {
850 	char *p, tmp[4], newid[DISK_IDENT_SIZE];
851 
852 	newid[0] = '\0';
853 	for (p = ident; *p != '\0'; p++) {
854 		if (isprint(*p)) {
855 			tmp[0] = *p;
856 			tmp[1] = '\0';
857 		} else {
858 			snprintf(tmp, sizeof(tmp), "x%02hhx",
859 			    *(unsigned char *)p);
860 		}
861 		if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
862 			break;
863 	}
864 	bzero(ident, size);
865 	strlcpy(ident, newid, size);
866 }
867 
868 struct disk *
869 disk_alloc(void)
870 {
871 	struct disk *dp;
872 
873 	dp = g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO);
874 	LIST_INIT(&dp->d_aliases);
875 	dp->d_init_level = DISK_INIT_NONE;
876 	dp->d_cevent = g_alloc_event(M_WAITOK);
877 	dp->d_devent = g_alloc_event(M_WAITOK);
878 	return (dp);
879 }
880 
881 void
882 disk_create(struct disk *dp, int version)
883 {
884 
885 	if (version != DISK_VERSION) {
886 		printf("WARNING: Attempt to add disk %s%d %s",
887 		    dp->d_name, dp->d_unit,
888 		    " using incompatible ABI version of disk(9)\n");
889 		printf("WARNING: Ignoring disk %s%d\n",
890 		    dp->d_name, dp->d_unit);
891 		return;
892 	}
893 	if (dp->d_flags & DISKFLAG_RESERVED) {
894 		printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
895 		    dp->d_name, dp->d_unit);
896 		printf("WARNING: Ignoring disk %s%d\n",
897 		    dp->d_name, dp->d_unit);
898 		return;
899 	}
900 	KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
901 	KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
902 	KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
903 	KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
904 	g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
905 
906 	dp->d_init_level = DISK_INIT_CREATE;
907 
908 	KASSERT(dp->d_cevent != NULL,
909 	    ("Disk create for %p with event NULL", dp));
910 	g_post_event_ep(g_disk_create, dp, dp->d_cevent, dp, NULL);
911 }
912 
913 void
914 disk_destroy(struct disk *dp)
915 {
916 	struct disk_alias *dap, *daptmp;
917 
918 	/* If disk_create() was never called, just free the resources. */
919 	if (dp->d_init_level < DISK_INIT_CREATE) {
920 		if (dp->d_devstat != NULL)
921 			devstat_remove_entry(dp->d_devstat);
922 		LIST_FOREACH_SAFE(dap, &dp->d_aliases, da_next, daptmp)
923 			g_free(dap);
924 		g_free(dp->d_cevent);
925 		g_free(dp->d_devent);
926 		g_free(dp);
927 		return;
928 	}
929 
930 	KASSERT(dp->d_devent != NULL,
931 	    ("Disk destroy for %p with event NULL", dp));
932 	disk_gone(dp);
933 	dp->d_destroyed = 1;
934 	g_cancel_event(dp);
935 	if (dp->d_devstat != NULL)
936 		devstat_remove_entry(dp->d_devstat);
937 	g_post_event_ep(g_disk_destroy, dp, dp->d_devent, NULL);
938 }
939 
940 void
941 disk_add_alias(struct disk *dp, const char *name)
942 {
943 	struct disk_alias *dap;
944 
945 	dap = (struct disk_alias *)g_malloc(
946 		sizeof(struct disk_alias) + strlen(name) + 1, M_WAITOK);
947 	strcpy((char *)(dap + 1), name);
948 	dap->da_alias = (const char *)(dap + 1);
949 	LIST_INSERT_HEAD(&dp->d_aliases, dap, da_next);
950 }
951 
952 void
953 disk_gone(struct disk *dp)
954 {
955 	struct g_geom *gp;
956 	struct g_provider *pp;
957 
958 	mtx_pool_lock(mtxpool_sleep, dp);
959 
960 	/*
961 	 * Second wither call makes no sense, plus we can not access the list
962 	 * of providers without topology lock after calling wither once.
963 	 */
964 	if (dp->d_goneflag != 0) {
965 		mtx_pool_unlock(mtxpool_sleep, dp);
966 		return;
967 	}
968 
969 	dp->d_goneflag = 1;
970 
971 	/*
972 	 * If we're still in the process of creating this disk (the
973 	 * g_disk_create() function is still queued, or is in
974 	 * progress), the init level will not yet be DISK_INIT_DONE.
975 	 *
976 	 * If that is the case, g_disk_create() will see d_goneflag
977 	 * and take care of cleaning things up.
978 	 *
979 	 * If the disk has already been created, we default to
980 	 * withering the provider as usual below.
981 	 *
982 	 * If the caller has not set a d_gone() callback, he will
983 	 * not be any worse off by returning here, because the geom
984 	 * has not been fully setup in any case.
985 	 */
986 	if (dp->d_init_level < DISK_INIT_DONE) {
987 		mtx_pool_unlock(mtxpool_sleep, dp);
988 		return;
989 	}
990 	mtx_pool_unlock(mtxpool_sleep, dp);
991 
992 	gp = dp->d_geom;
993 	pp = LIST_FIRST(&gp->provider);
994 	if (pp != NULL) {
995 		KASSERT(LIST_NEXT(pp, provider) == NULL,
996 		    ("geom %p has more than one provider", gp));
997 		g_wither_provider(pp, ENXIO);
998 	}
999 }
1000 
1001 void
1002 disk_attr_changed(struct disk *dp, const char *attr, int flag)
1003 {
1004 	struct g_geom *gp = dp->d_geom;
1005 	struct g_provider *pp;
1006 	char devnamebuf[128];
1007 
1008 	if (gp == NULL)
1009 		return;
1010 	LIST_FOREACH(pp, &gp->provider, provider)
1011 		(void)g_attr_changed(pp, attr, flag);
1012 	snprintf(devnamebuf, sizeof(devnamebuf), "devname=%s%d", dp->d_name,
1013 	    dp->d_unit);
1014 	devctl_notify("GEOM", "disk", attr, devnamebuf);
1015 }
1016 
1017 void
1018 disk_media_changed(struct disk *dp, int flag)
1019 {
1020 	struct g_geom *gp = dp->d_geom;
1021 	struct g_provider *pp;
1022 
1023 	if (gp == NULL)
1024 		return;
1025 	pp = LIST_FIRST(&gp->provider);
1026 	if (pp != NULL) {
1027 		KASSERT(LIST_NEXT(pp, provider) == NULL,
1028 		    ("geom %p has more than one provider", gp));
1029 		g_media_changed(pp, flag);
1030 	}
1031 }
1032 
1033 void
1034 disk_media_gone(struct disk *dp, int flag)
1035 {
1036 	struct g_geom *gp = dp->d_geom;
1037 	struct g_provider *pp;
1038 
1039 	if (gp == NULL)
1040 		return;
1041 	pp = LIST_FIRST(&gp->provider);
1042 	if (pp != NULL) {
1043 		KASSERT(LIST_NEXT(pp, provider) == NULL,
1044 		    ("geom %p has more than one provider", gp));
1045 		g_media_gone(pp, flag);
1046 	}
1047 }
1048 
1049 int
1050 disk_resize(struct disk *dp, int flag)
1051 {
1052 
1053 	if (dp->d_destroyed || dp->d_geom == NULL)
1054 		return (0);
1055 
1056 	return (g_post_event(g_disk_resize, dp, flag, NULL));
1057 }
1058 
1059 static void
1060 g_kern_disks(void *p, int flag __unused)
1061 {
1062 	struct sbuf *sb;
1063 	struct g_geom *gp;
1064 	char *sp;
1065 
1066 	sb = p;
1067 	sp = "";
1068 	g_topology_assert();
1069 	LIST_FOREACH(gp, &g_disk_class.geom, geom) {
1070 		sbuf_printf(sb, "%s%s", sp, gp->name);
1071 		sp = " ";
1072 	}
1073 	sbuf_finish(sb);
1074 }
1075 
1076 static int
1077 g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS)
1078 {
1079 	struct disk *dp;
1080 	struct sbuf *sb;
1081 	int error;
1082 
1083 	sb = sbuf_new_auto();
1084 	dp = (struct disk *)arg1;
1085 	sbuf_printf(sb, "%b", dp->d_flags,
1086 		"\20"
1087 		"\2OPEN"
1088 		"\3CANDELETE"
1089 		"\4CANFLUSHCACHE"
1090 		"\5UNMAPPEDBIO"
1091 		"\6DIRECTCOMPLETION"
1092 		"\10CANZONE"
1093 		"\11WRITEPROTECT");
1094 
1095 	sbuf_finish(sb);
1096 	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1097 	sbuf_delete(sb);
1098 	return (error);
1099 }
1100 
1101 static int
1102 sysctl_disks(SYSCTL_HANDLER_ARGS)
1103 {
1104 	int error;
1105 	struct sbuf *sb;
1106 
1107 	sb = sbuf_new_auto();
1108 	g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
1109 	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1110 	sbuf_delete(sb);
1111 	return error;
1112 }
1113 
1114 SYSCTL_PROC(_kern, OID_AUTO, disks,
1115     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
1116     sysctl_disks, "A", "names of available disks");
1117