xref: /freebsd/sys/geom/geom_disk.c (revision c0e5e7f3d2e1bac7ada43b35843e3a018777854e)
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/ctype.h>
47 #include <sys/fcntl.h>
48 #include <sys/malloc.h>
49 #include <sys/sbuf.h>
50 #include <sys/devicestat.h>
51 #include <machine/md_var.h>
52 
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <geom/geom.h>
56 #include <geom/geom_disk.h>
57 #include <geom/geom_int.h>
58 
59 #include <dev/led/led.h>
60 
61 #include <machine/bus.h>
62 
63 struct g_disk_softc {
64 	struct mtx		 done_mtx;
65 	struct disk		*dp;
66 	struct sysctl_ctx_list	sysctl_ctx;
67 	struct sysctl_oid	*sysctl_tree;
68 	char			led[64];
69 	uint32_t		state;
70 	struct mtx		 start_mtx;
71 };
72 
73 static g_access_t g_disk_access;
74 static g_start_t g_disk_start;
75 static g_ioctl_t g_disk_ioctl;
76 static g_dumpconf_t g_disk_dumpconf;
77 static g_provgone_t g_disk_providergone;
78 
79 static struct g_class g_disk_class = {
80 	.name = G_DISK_CLASS_NAME,
81 	.version = G_VERSION,
82 	.start = g_disk_start,
83 	.access = g_disk_access,
84 	.ioctl = g_disk_ioctl,
85 	.providergone = g_disk_providergone,
86 	.dumpconf = g_disk_dumpconf,
87 };
88 
89 SYSCTL_DECL(_kern_geom);
90 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0,
91     "GEOM_DISK stuff");
92 
93 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
94 
95 static int
96 g_disk_access(struct g_provider *pp, int r, int w, int e)
97 {
98 	struct disk *dp;
99 	struct g_disk_softc *sc;
100 	int error;
101 
102 	g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
103 	    pp->name, r, w, e);
104 	g_topology_assert();
105 	sc = pp->private;
106 	if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
107 		/*
108 		 * Allow decreasing access count even if disk is not
109 		 * avaliable anymore.
110 		 */
111 		if (r <= 0 && w <= 0 && e <= 0)
112 			return (0);
113 		return (ENXIO);
114 	}
115 	r += pp->acr;
116 	w += pp->acw;
117 	e += pp->ace;
118 	error = 0;
119 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
120 		if (dp->d_open != NULL) {
121 			error = dp->d_open(dp);
122 			if (bootverbose && error != 0)
123 				printf("Opened disk %s -> %d\n",
124 				    pp->name, error);
125 			if (error != 0)
126 				return (error);
127 		}
128 		pp->mediasize = dp->d_mediasize;
129 		pp->sectorsize = dp->d_sectorsize;
130 		if (dp->d_maxsize == 0) {
131 			printf("WARNING: Disk drive %s%d has no d_maxsize\n",
132 			    dp->d_name, dp->d_unit);
133 			dp->d_maxsize = DFLTPHYS;
134 		}
135 		if (dp->d_delmaxsize == 0) {
136 			if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
137 				printf("WARNING: Disk drive %s%d has no "
138 				    "d_delmaxsize\n", dp->d_name, dp->d_unit);
139 			}
140 			dp->d_delmaxsize = dp->d_maxsize;
141 		}
142 		pp->stripeoffset = dp->d_stripeoffset;
143 		pp->stripesize = dp->d_stripesize;
144 		dp->d_flags |= DISKFLAG_OPEN;
145 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
146 		if (dp->d_close != NULL) {
147 			error = dp->d_close(dp);
148 			if (error != 0)
149 				printf("Closed disk %s -> %d\n",
150 				    pp->name, error);
151 		}
152 		sc->state = G_STATE_ACTIVE;
153 		if (sc->led[0] != 0)
154 			led_set(sc->led, "0");
155 		dp->d_flags &= ~DISKFLAG_OPEN;
156 	}
157 	return (error);
158 }
159 
160 static void
161 g_disk_kerneldump(struct bio *bp, struct disk *dp)
162 {
163 	struct g_kerneldump *gkd;
164 	struct g_geom *gp;
165 
166 	gkd = (struct g_kerneldump*)bp->bio_data;
167 	gp = bp->bio_to->geom;
168 	g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)",
169 		gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
170 	if (dp->d_dump == NULL) {
171 		g_io_deliver(bp, ENODEV);
172 		return;
173 	}
174 	gkd->di.dumper = dp->d_dump;
175 	gkd->di.priv = dp;
176 	gkd->di.blocksize = dp->d_sectorsize;
177 	gkd->di.maxiosize = dp->d_maxsize;
178 	gkd->di.mediaoffset = gkd->offset;
179 	if ((gkd->offset + gkd->length) > dp->d_mediasize)
180 		gkd->length = dp->d_mediasize - gkd->offset;
181 	gkd->di.mediasize = gkd->length;
182 	g_io_deliver(bp, 0);
183 }
184 
185 static void
186 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
187 {
188 	const char *cmd;
189 
190 	memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
191 	if (sc->led[0] != 0) {
192 		switch (sc->state) {
193 		case G_STATE_FAILED:
194 			cmd = "1";
195 			break;
196 		case G_STATE_REBUILD:
197 			cmd = "f5";
198 			break;
199 		case G_STATE_RESYNC:
200 			cmd = "f1";
201 			break;
202 		default:
203 			cmd = "0";
204 			break;
205 		}
206 		led_set(sc->led, cmd);
207 	}
208 	g_io_deliver(bp, 0);
209 }
210 
211 static void
212 g_disk_done(struct bio *bp)
213 {
214 	struct bintime now;
215 	struct bio *bp2;
216 	struct g_disk_softc *sc;
217 
218 	/* See "notes" for why we need a mutex here */
219 	/* XXX: will witness accept a mix of Giant/unGiant drivers here ? */
220 	bp2 = bp->bio_parent;
221 	sc = bp2->bio_to->private;
222 	bp->bio_completed = bp->bio_length - bp->bio_resid;
223 	binuptime(&now);
224 	mtx_lock(&sc->done_mtx);
225 	if (bp2->bio_error == 0)
226 		bp2->bio_error = bp->bio_error;
227 	bp2->bio_completed += bp->bio_completed;
228 	switch (bp->bio_cmd) {
229 	case BIO_READ:
230 	case BIO_WRITE:
231 	case BIO_DELETE:
232 	case BIO_FLUSH:
233 		devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
234 		break;
235 	default:
236 		break;
237 	}
238 	bp2->bio_inbed++;
239 	if (bp2->bio_children == bp2->bio_inbed) {
240 		mtx_unlock(&sc->done_mtx);
241 		bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
242 		g_io_deliver(bp2, bp2->bio_error);
243 	} else
244 		mtx_unlock(&sc->done_mtx);
245 	g_destroy_bio(bp);
246 }
247 
248 static int
249 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
250 {
251 	struct disk *dp;
252 	struct g_disk_softc *sc;
253 	int error;
254 
255 	sc = pp->private;
256 	dp = sc->dp;
257 
258 	if (dp->d_ioctl == NULL)
259 		return (ENOIOCTL);
260 	error = dp->d_ioctl(dp, cmd, data, fflag, td);
261 	return (error);
262 }
263 
264 static off_t
265 g_disk_maxsize(struct disk *dp, struct bio *bp)
266 {
267 	if (bp->bio_cmd == BIO_DELETE)
268 		return (dp->d_delmaxsize);
269 	return (dp->d_maxsize);
270 }
271 
272 static int
273 g_disk_maxsegs(struct disk *dp, struct bio *bp)
274 {
275 	return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1);
276 }
277 
278 static void
279 g_disk_advance(struct disk *dp, struct bio *bp, off_t off)
280 {
281 
282 	bp->bio_offset += off;
283 	bp->bio_length -= off;
284 
285 	if ((bp->bio_flags & BIO_VLIST) != 0) {
286 		bus_dma_segment_t *seg, *end;
287 
288 		seg = (bus_dma_segment_t *)bp->bio_data;
289 		end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
290 		off += bp->bio_ma_offset;
291 		while (off >= seg->ds_len) {
292 			KASSERT((seg != end),
293 			    ("vlist request runs off the end"));
294 			off -= seg->ds_len;
295 			seg++;
296 		}
297 		bp->bio_ma_offset = off;
298 		bp->bio_ma_n = end - seg;
299 		bp->bio_data = (void *)seg;
300 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
301 		bp->bio_ma += off / PAGE_SIZE;
302 		bp->bio_ma_offset += off;
303 		bp->bio_ma_offset %= PAGE_SIZE;
304 		bp->bio_ma_n -= off / PAGE_SIZE;
305 	} else {
306 		bp->bio_data += off;
307 	}
308 }
309 
310 static void
311 g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset,
312     off_t *plength, int *ppages)
313 {
314 	uintptr_t seg_page_base;
315 	uintptr_t seg_page_end;
316 	off_t offset;
317 	off_t length;
318 	int seg_pages;
319 
320 	offset = *poffset;
321 	length = *plength;
322 
323 	if (length > seg->ds_len - offset)
324 		length = seg->ds_len - offset;
325 
326 	seg_page_base = trunc_page(seg->ds_addr + offset);
327 	seg_page_end  = round_page(seg->ds_addr + offset + length);
328 	seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT;
329 
330 	if (seg_pages > *ppages) {
331 		seg_pages = *ppages;
332 		length = (seg_page_base + (seg_pages << PAGE_SHIFT)) -
333 		    (seg->ds_addr + offset);
334 	}
335 
336 	*poffset = 0;
337 	*plength -= length;
338 	*ppages -= seg_pages;
339 }
340 
341 static off_t
342 g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg)
343 {
344 	bus_dma_segment_t *seg, *end;
345 	off_t residual;
346 	off_t offset;
347 	int pages;
348 
349 	seg = (bus_dma_segment_t *)bp->bio_data;
350 	end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
351 	residual = bp->bio_length;
352 	offset = bp->bio_ma_offset;
353 	pages = g_disk_maxsegs(dp, bp);
354 	while (residual != 0 && pages != 0) {
355 		KASSERT((seg != end),
356 		    ("vlist limit runs off the end"));
357 		g_disk_seg_limit(seg, &offset, &residual, &pages);
358 		seg++;
359 	}
360 	if (pendseg != NULL)
361 		*pendseg = seg;
362 	return (residual);
363 }
364 
365 static bool
366 g_disk_limit(struct disk *dp, struct bio *bp)
367 {
368 	bool limited = false;
369 	off_t maxsz;
370 
371 	maxsz = g_disk_maxsize(dp, bp);
372 
373 	/*
374 	 * XXX: If we have a stripesize we should really use it here.
375 	 *      Care should be taken in the delete case if this is done
376 	 *      as deletes can be very sensitive to size given how they
377 	 *      are processed.
378 	 */
379 	if (bp->bio_length > maxsz) {
380 		bp->bio_length = maxsz;
381 		limited = true;
382 	}
383 
384 	if ((bp->bio_flags & BIO_VLIST) != 0) {
385 		bus_dma_segment_t *firstseg, *endseg;
386 		off_t residual;
387 
388 		firstseg = (bus_dma_segment_t*)bp->bio_data;
389 		residual = g_disk_vlist_limit(dp, bp, &endseg);
390 		if (residual != 0) {
391 			bp->bio_ma_n = endseg - firstseg;
392 			bp->bio_length -= residual;
393 			limited = true;
394 		}
395 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
396 		bp->bio_ma_n =
397 		    howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE);
398 	}
399 
400 	return (limited);
401 }
402 
403 static void
404 g_disk_start(struct bio *bp)
405 {
406 	struct bio *bp2, *bp3;
407 	struct disk *dp;
408 	struct g_disk_softc *sc;
409 	int error;
410 	off_t off;
411 
412 	sc = bp->bio_to->private;
413 	if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
414 		g_io_deliver(bp, ENXIO);
415 		return;
416 	}
417 	error = EJUSTRETURN;
418 	switch(bp->bio_cmd) {
419 	case BIO_DELETE:
420 		if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
421 			error = EOPNOTSUPP;
422 			break;
423 		}
424 		/* fall-through */
425 	case BIO_READ:
426 	case BIO_WRITE:
427 		KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 ||
428 		    (bp->bio_flags & BIO_UNMAPPED) == 0,
429 		    ("unmapped bio not supported by disk %s", dp->d_name));
430 		off = 0;
431 		bp3 = NULL;
432 		bp2 = g_clone_bio(bp);
433 		if (bp2 == NULL) {
434 			error = ENOMEM;
435 			break;
436 		}
437 		for (;;) {
438 			if (g_disk_limit(dp, bp2)) {
439 				off += bp2->bio_length;
440 
441 				/*
442 				 * To avoid a race, we need to grab the next bio
443 				 * before we schedule this one.  See "notes".
444 				 */
445 				bp3 = g_clone_bio(bp);
446 				if (bp3 == NULL)
447 					bp->bio_error = ENOMEM;
448 			}
449 			bp2->bio_done = g_disk_done;
450 			bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
451 			bp2->bio_bcount = bp2->bio_length;
452 			bp2->bio_disk = dp;
453 			mtx_lock(&sc->start_mtx);
454 			devstat_start_transaction_bio(dp->d_devstat, bp2);
455 			mtx_unlock(&sc->start_mtx);
456 			dp->d_strategy(bp2);
457 
458 			if (bp3 == NULL)
459 				break;
460 
461 			bp2 = bp3;
462 			bp3 = NULL;
463 			g_disk_advance(dp, bp2, off);
464 		}
465 		break;
466 	case BIO_GETATTR:
467 		/* Give the driver a chance to override */
468 		if (dp->d_getattr != NULL) {
469 			if (bp->bio_disk == NULL)
470 				bp->bio_disk = dp;
471 			error = dp->d_getattr(bp);
472 			if (error != -1)
473 				break;
474 			error = EJUSTRETURN;
475 		}
476 		if (g_handleattr_int(bp, "GEOM::candelete",
477 		    (dp->d_flags & DISKFLAG_CANDELETE) != 0))
478 			break;
479 		else if (g_handleattr_int(bp, "GEOM::fwsectors",
480 		    dp->d_fwsectors))
481 			break;
482 		else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
483 			break;
484 		else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0))
485 			break;
486 		else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
487 			break;
488 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
489 		    dp->d_hba_vendor))
490 			break;
491 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
492 		    dp->d_hba_device))
493 			break;
494 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
495 		    dp->d_hba_subvendor))
496 			break;
497 		else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
498 		    dp->d_hba_subdevice))
499 			break;
500 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
501 			g_disk_kerneldump(bp, dp);
502 		else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
503 			g_disk_setstate(bp, sc);
504 		else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
505 		    dp->d_rotation_rate))
506 			break;
507 		else
508 			error = ENOIOCTL;
509 		break;
510 	case BIO_FLUSH:
511 		g_trace(G_T_BIO, "g_disk_flushcache(%s)",
512 		    bp->bio_to->name);
513 		if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
514 			error = EOPNOTSUPP;
515 			break;
516 		}
517 		bp2 = g_clone_bio(bp);
518 		if (bp2 == NULL) {
519 			g_io_deliver(bp, ENOMEM);
520 			return;
521 		}
522 		bp2->bio_done = g_disk_done;
523 		bp2->bio_disk = dp;
524 		mtx_lock(&sc->start_mtx);
525 		devstat_start_transaction_bio(dp->d_devstat, bp2);
526 		mtx_unlock(&sc->start_mtx);
527 		dp->d_strategy(bp2);
528 		break;
529 	default:
530 		error = EOPNOTSUPP;
531 		break;
532 	}
533 	if (error != EJUSTRETURN)
534 		g_io_deliver(bp, error);
535 	return;
536 }
537 
538 static void
539 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
540 {
541 	struct bio *bp;
542 	struct disk *dp;
543 	struct g_disk_softc *sc;
544 	char *buf;
545 	int res = 0;
546 
547 	sc = gp->softc;
548 	if (sc == NULL || (dp = sc->dp) == NULL)
549 		return;
550 	if (indent == NULL) {
551 		sbuf_printf(sb, " hd %u", dp->d_fwheads);
552 		sbuf_printf(sb, " sc %u", dp->d_fwsectors);
553 		return;
554 	}
555 	if (pp != NULL) {
556 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
557 		    indent, dp->d_fwheads);
558 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
559 		    indent, dp->d_fwsectors);
560 
561 		/*
562 		 * "rotationrate" is a little complicated, because the value
563 		 * returned by the drive might not be the RPM; 0 and 1 are
564 		 * special cases, and there's also a valid range.
565 		 */
566 		sbuf_printf(sb, "%s<rotationrate>", indent);
567 		if (dp->d_rotation_rate == 0)		/* Old drives don't */
568 			sbuf_printf(sb, "unknown");	/* report RPM. */
569 		else if (dp->d_rotation_rate == 1)	/* Since 0 is used */
570 			sbuf_printf(sb, "0");		/* above, SSDs use 1. */
571 		else if ((dp->d_rotation_rate >= 0x041) &&
572 		    (dp->d_rotation_rate <= 0xfffe))
573 			sbuf_printf(sb, "%u", dp->d_rotation_rate);
574 		else
575 			sbuf_printf(sb, "invalid");
576 		sbuf_printf(sb, "</rotationrate>\n");
577 		if (dp->d_getattr != NULL) {
578 			buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
579 			bp = g_alloc_bio();
580 			bp->bio_disk = dp;
581 			bp->bio_attribute = "GEOM::ident";
582 			bp->bio_length = DISK_IDENT_SIZE;
583 			bp->bio_data = buf;
584 			res = dp->d_getattr(bp);
585 			sbuf_printf(sb, "%s<ident>", indent);
586 			g_conf_printf_escaped(sb, "%s",
587 			    res == 0 ? buf: dp->d_ident);
588 			sbuf_printf(sb, "</ident>\n");
589 			bp->bio_attribute = "GEOM::lunid";
590 			bp->bio_length = DISK_IDENT_SIZE;
591 			bp->bio_data = buf;
592 			if (dp->d_getattr(bp) == 0) {
593 				sbuf_printf(sb, "%s<lunid>", indent);
594 				g_conf_printf_escaped(sb, "%s", buf);
595 				sbuf_printf(sb, "</lunid>\n");
596 			}
597 			bp->bio_attribute = "GEOM::lunname";
598 			bp->bio_length = DISK_IDENT_SIZE;
599 			bp->bio_data = buf;
600 			if (dp->d_getattr(bp) == 0) {
601 				sbuf_printf(sb, "%s<lunname>", indent);
602 				g_conf_printf_escaped(sb, "%s", buf);
603 				sbuf_printf(sb, "</lunname>\n");
604 			}
605 			g_destroy_bio(bp);
606 			g_free(buf);
607 		} else {
608 			sbuf_printf(sb, "%s<ident>", indent);
609 			g_conf_printf_escaped(sb, "%s", dp->d_ident);
610 			sbuf_printf(sb, "</ident>\n");
611 		}
612 		sbuf_printf(sb, "%s<descr>", indent);
613 		g_conf_printf_escaped(sb, "%s", dp->d_descr);
614 		sbuf_printf(sb, "</descr>\n");
615 	}
616 }
617 
618 static void
619 g_disk_resize(void *ptr, int flag)
620 {
621 	struct disk *dp;
622 	struct g_geom *gp;
623 	struct g_provider *pp;
624 
625 	if (flag == EV_CANCEL)
626 		return;
627 	g_topology_assert();
628 
629 	dp = ptr;
630 	gp = dp->d_geom;
631 
632 	if (dp->d_destroyed || gp == NULL)
633 		return;
634 
635 	LIST_FOREACH(pp, &gp->provider, provider) {
636 		if (pp->sectorsize != 0 &&
637 		    pp->sectorsize != dp->d_sectorsize)
638 			g_wither_provider(pp, ENXIO);
639 		else
640 			g_resize_provider(pp, dp->d_mediasize);
641 	}
642 }
643 
644 static void
645 g_disk_create(void *arg, int flag)
646 {
647 	struct g_geom *gp;
648 	struct g_provider *pp;
649 	struct disk *dp;
650 	struct g_disk_softc *sc;
651 	char tmpstr[80];
652 
653 	if (flag == EV_CANCEL)
654 		return;
655 	g_topology_assert();
656 	dp = arg;
657 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
658 	mtx_init(&sc->start_mtx, "g_disk_start", NULL, MTX_DEF);
659 	mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
660 	sc->dp = dp;
661 	gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
662 	gp->softc = sc;
663 	pp = g_new_providerf(gp, "%s", gp->name);
664 	devstat_remove_entry(pp->stat);
665 	pp->stat = NULL;
666 	dp->d_devstat->id = pp;
667 	pp->mediasize = dp->d_mediasize;
668 	pp->sectorsize = dp->d_sectorsize;
669 	pp->stripeoffset = dp->d_stripeoffset;
670 	pp->stripesize = dp->d_stripesize;
671 	if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
672 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
673 	if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
674 		pp->flags |= G_PF_DIRECT_SEND;
675 	pp->flags |= G_PF_DIRECT_RECEIVE;
676 	if (bootverbose)
677 		printf("GEOM: new disk %s\n", gp->name);
678 	sysctl_ctx_init(&sc->sysctl_ctx);
679 	snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
680 	sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
681 		SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
682 		CTLFLAG_RD, 0, tmpstr);
683 	if (sc->sysctl_tree != NULL) {
684 		SYSCTL_ADD_STRING(&sc->sysctl_ctx,
685 		    SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
686 		    CTLFLAG_RWTUN, sc->led, sizeof(sc->led),
687 		    "LED name");
688 	}
689 	pp->private = sc;
690 	dp->d_geom = gp;
691 	g_error_provider(pp, 0);
692 }
693 
694 /*
695  * We get this callback after all of the consumers have gone away, and just
696  * before the provider is freed.  If the disk driver provided a d_gone
697  * callback, let them know that it is okay to free resources -- they won't
698  * be getting any more accesses from GEOM.
699  */
700 static void
701 g_disk_providergone(struct g_provider *pp)
702 {
703 	struct disk *dp;
704 	struct g_disk_softc *sc;
705 
706 	sc = (struct g_disk_softc *)pp->private;
707 	dp = sc->dp;
708 	if (dp != NULL && dp->d_gone != NULL)
709 		dp->d_gone(dp);
710 	if (sc->sysctl_tree != NULL) {
711 		sysctl_ctx_free(&sc->sysctl_ctx);
712 		sc->sysctl_tree = NULL;
713 	}
714 	if (sc->led[0] != 0) {
715 		led_set(sc->led, "0");
716 		sc->led[0] = 0;
717 	}
718 	pp->private = NULL;
719 	pp->geom->softc = NULL;
720 	mtx_destroy(&sc->done_mtx);
721 	mtx_destroy(&sc->start_mtx);
722 	g_free(sc);
723 }
724 
725 static void
726 g_disk_destroy(void *ptr, int flag)
727 {
728 	struct disk *dp;
729 	struct g_geom *gp;
730 	struct g_disk_softc *sc;
731 
732 	g_topology_assert();
733 	dp = ptr;
734 	gp = dp->d_geom;
735 	if (gp != NULL) {
736 		sc = gp->softc;
737 		if (sc != NULL)
738 			sc->dp = NULL;
739 		dp->d_geom = NULL;
740 		g_wither_geom(gp, ENXIO);
741 	}
742 	g_free(dp);
743 }
744 
745 /*
746  * We only allow printable characters in disk ident,
747  * the rest is converted to 'x<HH>'.
748  */
749 static void
750 g_disk_ident_adjust(char *ident, size_t size)
751 {
752 	char *p, tmp[4], newid[DISK_IDENT_SIZE];
753 
754 	newid[0] = '\0';
755 	for (p = ident; *p != '\0'; p++) {
756 		if (isprint(*p)) {
757 			tmp[0] = *p;
758 			tmp[1] = '\0';
759 		} else {
760 			snprintf(tmp, sizeof(tmp), "x%02hhx",
761 			    *(unsigned char *)p);
762 		}
763 		if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
764 			break;
765 	}
766 	bzero(ident, size);
767 	strlcpy(ident, newid, size);
768 }
769 
770 struct disk *
771 disk_alloc(void)
772 {
773 
774 	return (g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO));
775 }
776 
777 void
778 disk_create(struct disk *dp, int version)
779 {
780 
781 	if (version != DISK_VERSION) {
782 		printf("WARNING: Attempt to add disk %s%d %s",
783 		    dp->d_name, dp->d_unit,
784 		    " using incompatible ABI version of disk(9)\n");
785 		printf("WARNING: Ignoring disk %s%d\n",
786 		    dp->d_name, dp->d_unit);
787 		return;
788 	}
789 	if (dp->d_flags & DISKFLAG_RESERVED) {
790 		printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
791 		    dp->d_name, dp->d_unit);
792 		printf("WARNING: Ignoring disk %s%d\n",
793 		    dp->d_name, dp->d_unit);
794 		return;
795 	}
796 	KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
797 	KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
798 	KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
799 	KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
800 	if (dp->d_devstat == NULL)
801 		dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
802 		    dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
803 		    DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
804 	dp->d_geom = NULL;
805 	g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
806 	g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
807 }
808 
809 void
810 disk_destroy(struct disk *dp)
811 {
812 
813 	g_cancel_event(dp);
814 	dp->d_destroyed = 1;
815 	if (dp->d_devstat != NULL)
816 		devstat_remove_entry(dp->d_devstat);
817 	g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
818 }
819 
820 void
821 disk_gone(struct disk *dp)
822 {
823 	struct g_geom *gp;
824 	struct g_provider *pp;
825 
826 	gp = dp->d_geom;
827 	if (gp != NULL) {
828 		pp = LIST_FIRST(&gp->provider);
829 		if (pp != NULL) {
830 			KASSERT(LIST_NEXT(pp, provider) == NULL,
831 			    ("geom %p has more than one provider", gp));
832 			g_wither_provider(pp, ENXIO);
833 		}
834 	}
835 }
836 
837 void
838 disk_attr_changed(struct disk *dp, const char *attr, int flag)
839 {
840 	struct g_geom *gp;
841 	struct g_provider *pp;
842 
843 	gp = dp->d_geom;
844 	if (gp != NULL)
845 		LIST_FOREACH(pp, &gp->provider, provider)
846 			(void)g_attr_changed(pp, attr, flag);
847 }
848 
849 void
850 disk_media_changed(struct disk *dp, int flag)
851 {
852 	struct g_geom *gp;
853 	struct g_provider *pp;
854 
855 	gp = dp->d_geom;
856 	if (gp != NULL) {
857 		pp = LIST_FIRST(&gp->provider);
858 		if (pp != NULL) {
859 			KASSERT(LIST_NEXT(pp, provider) == NULL,
860 			    ("geom %p has more than one provider", gp));
861 			g_media_changed(pp, flag);
862 		}
863 	}
864 }
865 
866 void
867 disk_media_gone(struct disk *dp, int flag)
868 {
869 	struct g_geom *gp;
870 	struct g_provider *pp;
871 
872 	gp = dp->d_geom;
873 	if (gp != NULL) {
874 		pp = LIST_FIRST(&gp->provider);
875 		if (pp != NULL) {
876 			KASSERT(LIST_NEXT(pp, provider) == NULL,
877 			    ("geom %p has more than one provider", gp));
878 			g_media_gone(pp, flag);
879 		}
880 	}
881 }
882 
883 int
884 disk_resize(struct disk *dp, int flag)
885 {
886 
887 	if (dp->d_destroyed || dp->d_geom == NULL)
888 		return (0);
889 
890 	return (g_post_event(g_disk_resize, dp, flag, NULL));
891 }
892 
893 static void
894 g_kern_disks(void *p, int flag __unused)
895 {
896 	struct sbuf *sb;
897 	struct g_geom *gp;
898 	char *sp;
899 
900 	sb = p;
901 	sp = "";
902 	g_topology_assert();
903 	LIST_FOREACH(gp, &g_disk_class.geom, geom) {
904 		sbuf_printf(sb, "%s%s", sp, gp->name);
905 		sp = " ";
906 	}
907 	sbuf_finish(sb);
908 }
909 
910 static int
911 sysctl_disks(SYSCTL_HANDLER_ARGS)
912 {
913 	int error;
914 	struct sbuf *sb;
915 
916 	sb = sbuf_new_auto();
917 	g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
918 	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
919 	sbuf_delete(sb);
920 	return error;
921 }
922 
923 SYSCTL_PROC(_kern, OID_AUTO, disks,
924     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
925     sysctl_disks, "A", "names of available disks");
926