xref: /freebsd/sys/dev/md/md.c (revision 9162f64b58d01ec01481d60b6cdc06ffd8e8c7fc)
1 /*-
2  * ----------------------------------------------------------------------------
3  * "THE BEER-WARE LICENSE" (Revision 42):
4  * <phk@FreeBSD.ORG> wrote this file.  As long as you retain this notice you
5  * can do whatever you want with this stuff. If we meet some day, and you think
6  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
7  * ----------------------------------------------------------------------------
8  *
9  * $FreeBSD$
10  *
11  */
12 
13 /*-
14  * The following functions are based in the vn(4) driver: mdstart_swap(),
15  * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(),
16  * and as such under the following copyright:
17  *
18  * Copyright (c) 1988 University of Utah.
19  * Copyright (c) 1990, 1993
20  *	The Regents of the University of California.  All rights reserved.
21  *
22  * This code is derived from software contributed to Berkeley by
23  * the Systems Programming Group of the University of Utah Computer
24  * Science Department.
25  *
26  * Redistribution and use in source and binary forms, with or without
27  * modification, are permitted provided that the following conditions
28  * are met:
29  * 1. Redistributions of source code must retain the above copyright
30  *    notice, this list of conditions and the following disclaimer.
31  * 2. Redistributions in binary form must reproduce the above copyright
32  *    notice, this list of conditions and the following disclaimer in the
33  *    documentation and/or other materials provided with the distribution.
34  * 4. Neither the name of the University nor the names of its contributors
35  *    may be used to endorse or promote products derived from this software
36  *    without specific prior written permission.
37  *
38  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
39  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
40  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
41  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
42  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
43  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
44  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
45  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
46  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
47  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
48  * SUCH DAMAGE.
49  *
50  * from: Utah Hdr: vn.c 1.13 94/04/02
51  *
52  *	from: @(#)vn.c	8.6 (Berkeley) 4/1/94
53  * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
54  */
55 
56 #include "opt_geom.h"
57 #include "opt_md.h"
58 
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/bio.h>
62 #include <sys/conf.h>
63 #include <sys/devicestat.h>
64 #include <sys/fcntl.h>
65 #include <sys/kernel.h>
66 #include <sys/kthread.h>
67 #include <sys/limits.h>
68 #include <sys/linker.h>
69 #include <sys/lock.h>
70 #include <sys/malloc.h>
71 #include <sys/mdioctl.h>
72 #include <sys/mount.h>
73 #include <sys/mutex.h>
74 #include <sys/sx.h>
75 #include <sys/namei.h>
76 #include <sys/proc.h>
77 #include <sys/queue.h>
78 #include <sys/sched.h>
79 #include <sys/sf_buf.h>
80 #include <sys/sysctl.h>
81 #include <sys/vnode.h>
82 
83 #include <geom/geom.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/swap_pager.h>
90 #include <vm/uma.h>
91 
92 #define MD_MODVER 1
93 
94 #define MD_SHUTDOWN	0x10000		/* Tell worker thread to terminate. */
95 #define	MD_EXITING	0x20000		/* Worker thread is exiting. */
96 
97 #ifndef MD_NSECT
98 #define MD_NSECT (10000 * 2)
99 #endif
100 
101 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
102 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
103 
104 static int md_debug;
105 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, "");
106 
107 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE)
108 /*
109  * Preloaded image gets put here.
110  * Applications that patch the object with the image can determine
111  * the size looking at the start and end markers (strings),
112  * so we want them contiguous.
113  */
114 static struct {
115 	u_char start[MD_ROOT_SIZE*1024];
116 	u_char end[128];
117 } mfs_root = {
118 	.start = "MFS Filesystem goes here",
119 	.end = "MFS Filesystem had better STOP here",
120 };
121 #endif
122 
123 static g_init_t g_md_init;
124 static g_fini_t g_md_fini;
125 static g_start_t g_md_start;
126 static g_access_t g_md_access;
127 static void g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
128     struct g_consumer *cp __unused, struct g_provider *pp);
129 
130 static int	mdunits;
131 static struct cdev *status_dev = 0;
132 static struct sx md_sx;
133 
134 static d_ioctl_t mdctlioctl;
135 
136 static struct cdevsw mdctl_cdevsw = {
137 	.d_version =	D_VERSION,
138 	.d_ioctl =	mdctlioctl,
139 	.d_name =	MD_NAME,
140 };
141 
142 struct g_class g_md_class = {
143 	.name = "MD",
144 	.version = G_VERSION,
145 	.init = g_md_init,
146 	.fini = g_md_fini,
147 	.start = g_md_start,
148 	.access = g_md_access,
149 	.dumpconf = g_md_dumpconf,
150 };
151 
152 DECLARE_GEOM_CLASS(g_md_class, g_md);
153 
154 
155 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list);
156 
157 #define NINDIR	(PAGE_SIZE / sizeof(uintptr_t))
158 #define NMASK	(NINDIR-1)
159 static int nshift;
160 
161 struct indir {
162 	uintptr_t	*array;
163 	u_int		total;
164 	u_int		used;
165 	u_int		shift;
166 };
167 
168 struct md_s {
169 	int unit;
170 	LIST_ENTRY(md_s) list;
171 	struct bio_queue_head bio_queue;
172 	struct mtx queue_mtx;
173 	struct cdev *dev;
174 	enum md_types type;
175 	off_t mediasize;
176 	unsigned sectorsize;
177 	unsigned opencount;
178 	unsigned fwheads;
179 	unsigned fwsectors;
180 	unsigned flags;
181 	char name[20];
182 	struct proc *procp;
183 	struct g_geom *gp;
184 	struct g_provider *pp;
185 	int (*start)(struct md_s *sc, struct bio *bp);
186 	struct devstat *devstat;
187 
188 	/* MD_MALLOC related fields */
189 	struct indir *indir;
190 	uma_zone_t uma;
191 
192 	/* MD_PRELOAD related fields */
193 	u_char *pl_ptr;
194 	size_t pl_len;
195 
196 	/* MD_VNODE related fields */
197 	struct vnode *vnode;
198 	char file[PATH_MAX];
199 	struct ucred *cred;
200 
201 	/* MD_SWAP related fields */
202 	vm_object_t object;
203 };
204 
205 static struct indir *
206 new_indir(u_int shift)
207 {
208 	struct indir *ip;
209 
210 	ip = malloc(sizeof *ip, M_MD, M_NOWAIT | M_ZERO);
211 	if (ip == NULL)
212 		return (NULL);
213 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
214 	    M_MDSECT, M_NOWAIT | M_ZERO);
215 	if (ip->array == NULL) {
216 		free(ip, M_MD);
217 		return (NULL);
218 	}
219 	ip->total = NINDIR;
220 	ip->shift = shift;
221 	return (ip);
222 }
223 
224 static void
225 del_indir(struct indir *ip)
226 {
227 
228 	free(ip->array, M_MDSECT);
229 	free(ip, M_MD);
230 }
231 
232 static void
233 destroy_indir(struct md_s *sc, struct indir *ip)
234 {
235 	int i;
236 
237 	for (i = 0; i < NINDIR; i++) {
238 		if (!ip->array[i])
239 			continue;
240 		if (ip->shift)
241 			destroy_indir(sc, (struct indir*)(ip->array[i]));
242 		else if (ip->array[i] > 255)
243 			uma_zfree(sc->uma, (void *)(ip->array[i]));
244 	}
245 	del_indir(ip);
246 }
247 
248 /*
249  * This function does the math and allocates the top level "indir" structure
250  * for a device of "size" sectors.
251  */
252 
253 static struct indir *
254 dimension(off_t size)
255 {
256 	off_t rcnt;
257 	struct indir *ip;
258 	int i, layer;
259 
260 	rcnt = size;
261 	layer = 0;
262 	while (rcnt > NINDIR) {
263 		rcnt /= NINDIR;
264 		layer++;
265 	}
266 	/* figure out log2(NINDIR) */
267 	for (i = NINDIR, nshift = -1; i; nshift++)
268 		i >>= 1;
269 
270 	/*
271 	 * XXX: the top layer is probably not fully populated, so we allocate
272 	 * too much space for ip->array in here.
273 	 */
274 	ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
275 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
276 	    M_MDSECT, M_WAITOK | M_ZERO);
277 	ip->total = NINDIR;
278 	ip->shift = layer * nshift;
279 	return (ip);
280 }
281 
282 /*
283  * Read a given sector
284  */
285 
286 static uintptr_t
287 s_read(struct indir *ip, off_t offset)
288 {
289 	struct indir *cip;
290 	int idx;
291 	uintptr_t up;
292 
293 	if (md_debug > 1)
294 		printf("s_read(%jd)\n", (intmax_t)offset);
295 	up = 0;
296 	for (cip = ip; cip != NULL;) {
297 		if (cip->shift) {
298 			idx = (offset >> cip->shift) & NMASK;
299 			up = cip->array[idx];
300 			cip = (struct indir *)up;
301 			continue;
302 		}
303 		idx = offset & NMASK;
304 		return (cip->array[idx]);
305 	}
306 	return (0);
307 }
308 
309 /*
310  * Write a given sector, prune the tree if the value is 0
311  */
312 
313 static int
314 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
315 {
316 	struct indir *cip, *lip[10];
317 	int idx, li;
318 	uintptr_t up;
319 
320 	if (md_debug > 1)
321 		printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
322 	up = 0;
323 	li = 0;
324 	cip = ip;
325 	for (;;) {
326 		lip[li++] = cip;
327 		if (cip->shift) {
328 			idx = (offset >> cip->shift) & NMASK;
329 			up = cip->array[idx];
330 			if (up != 0) {
331 				cip = (struct indir *)up;
332 				continue;
333 			}
334 			/* Allocate branch */
335 			cip->array[idx] =
336 			    (uintptr_t)new_indir(cip->shift - nshift);
337 			if (cip->array[idx] == 0)
338 				return (ENOSPC);
339 			cip->used++;
340 			up = cip->array[idx];
341 			cip = (struct indir *)up;
342 			continue;
343 		}
344 		/* leafnode */
345 		idx = offset & NMASK;
346 		up = cip->array[idx];
347 		if (up != 0)
348 			cip->used--;
349 		cip->array[idx] = ptr;
350 		if (ptr != 0)
351 			cip->used++;
352 		break;
353 	}
354 	if (cip->used != 0 || li == 1)
355 		return (0);
356 	li--;
357 	while (cip->used == 0 && cip != ip) {
358 		li--;
359 		idx = (offset >> lip[li]->shift) & NMASK;
360 		up = lip[li]->array[idx];
361 		KASSERT(up == (uintptr_t)cip, ("md screwed up"));
362 		del_indir(cip);
363 		lip[li]->array[idx] = 0;
364 		lip[li]->used--;
365 		cip = lip[li];
366 	}
367 	return (0);
368 }
369 
370 
371 static int
372 g_md_access(struct g_provider *pp, int r, int w, int e)
373 {
374 	struct md_s *sc;
375 
376 	sc = pp->geom->softc;
377 	if (sc == NULL) {
378 		if (r <= 0 && w <= 0 && e <= 0)
379 			return (0);
380 		return (ENXIO);
381 	}
382 	r += pp->acr;
383 	w += pp->acw;
384 	e += pp->ace;
385 	if ((sc->flags & MD_READONLY) != 0 && w > 0)
386 		return (EROFS);
387 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
388 		sc->opencount = 1;
389 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
390 		sc->opencount = 0;
391 	}
392 	return (0);
393 }
394 
395 static void
396 g_md_start(struct bio *bp)
397 {
398 	struct md_s *sc;
399 
400 	sc = bp->bio_to->geom->softc;
401 	if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
402 		devstat_start_transaction_bio(sc->devstat, bp);
403 	mtx_lock(&sc->queue_mtx);
404 	bioq_disksort(&sc->bio_queue, bp);
405 	mtx_unlock(&sc->queue_mtx);
406 	wakeup(sc);
407 }
408 
409 static int
410 mdstart_malloc(struct md_s *sc, struct bio *bp)
411 {
412 	int i, error;
413 	u_char *dst;
414 	off_t secno, nsec, uc;
415 	uintptr_t sp, osp;
416 
417 	switch (bp->bio_cmd) {
418 	case BIO_READ:
419 	case BIO_WRITE:
420 	case BIO_DELETE:
421 		break;
422 	default:
423 		return (EOPNOTSUPP);
424 	}
425 
426 	nsec = bp->bio_length / sc->sectorsize;
427 	secno = bp->bio_offset / sc->sectorsize;
428 	dst = bp->bio_data;
429 	error = 0;
430 	while (nsec--) {
431 		osp = s_read(sc->indir, secno);
432 		if (bp->bio_cmd == BIO_DELETE) {
433 			if (osp != 0)
434 				error = s_write(sc->indir, secno, 0);
435 		} else if (bp->bio_cmd == BIO_READ) {
436 			if (osp == 0)
437 				bzero(dst, sc->sectorsize);
438 			else if (osp <= 255)
439 				for (i = 0; i < sc->sectorsize; i++)
440 					dst[i] = osp;
441 			else
442 				bcopy((void *)osp, dst, sc->sectorsize);
443 			osp = 0;
444 		} else if (bp->bio_cmd == BIO_WRITE) {
445 			if (sc->flags & MD_COMPRESS) {
446 				uc = dst[0];
447 				for (i = 1; i < sc->sectorsize; i++)
448 					if (dst[i] != uc)
449 						break;
450 			} else {
451 				i = 0;
452 				uc = 0;
453 			}
454 			if (i == sc->sectorsize) {
455 				if (osp != uc)
456 					error = s_write(sc->indir, secno, uc);
457 			} else {
458 				if (osp <= 255) {
459 					sp = (uintptr_t)uma_zalloc(sc->uma,
460 					    M_NOWAIT);
461 					if (sp == 0) {
462 						error = ENOSPC;
463 						break;
464 					}
465 					bcopy(dst, (void *)sp, sc->sectorsize);
466 					error = s_write(sc->indir, secno, sp);
467 				} else {
468 					bcopy(dst, (void *)osp, sc->sectorsize);
469 					osp = 0;
470 				}
471 			}
472 		} else {
473 			error = EOPNOTSUPP;
474 		}
475 		if (osp > 255)
476 			uma_zfree(sc->uma, (void*)osp);
477 		if (error != 0)
478 			break;
479 		secno++;
480 		dst += sc->sectorsize;
481 	}
482 	bp->bio_resid = 0;
483 	return (error);
484 }
485 
486 static int
487 mdstart_preload(struct md_s *sc, struct bio *bp)
488 {
489 
490 	switch (bp->bio_cmd) {
491 	case BIO_READ:
492 		bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data,
493 		    bp->bio_length);
494 		break;
495 	case BIO_WRITE:
496 		bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset,
497 		    bp->bio_length);
498 		break;
499 	}
500 	bp->bio_resid = 0;
501 	return (0);
502 }
503 
504 static int
505 mdstart_vnode(struct md_s *sc, struct bio *bp)
506 {
507 	int error, vfslocked;
508 	struct uio auio;
509 	struct iovec aiov;
510 	struct mount *mp;
511 	struct vnode *vp;
512 	struct thread *td;
513 
514 	switch (bp->bio_cmd) {
515 	case BIO_READ:
516 	case BIO_WRITE:
517 	case BIO_FLUSH:
518 		break;
519 	default:
520 		return (EOPNOTSUPP);
521 	}
522 
523 	td = curthread;
524 	vp = sc->vnode;
525 
526 	/*
527 	 * VNODE I/O
528 	 *
529 	 * If an error occurs, we set BIO_ERROR but we do not set
530 	 * B_INVAL because (for a write anyway), the buffer is
531 	 * still valid.
532 	 */
533 
534 	if (bp->bio_cmd == BIO_FLUSH) {
535 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
536 		(void) vn_start_write(vp, &mp, V_WAIT);
537 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
538 		error = VOP_FSYNC(vp, MNT_WAIT, td);
539 		VOP_UNLOCK(vp, 0);
540 		vn_finished_write(mp);
541 		VFS_UNLOCK_GIANT(vfslocked);
542 		return (error);
543 	}
544 
545 	bzero(&auio, sizeof(auio));
546 
547 	aiov.iov_base = bp->bio_data;
548 	aiov.iov_len = bp->bio_length;
549 	auio.uio_iov = &aiov;
550 	auio.uio_iovcnt = 1;
551 	auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
552 	auio.uio_segflg = UIO_SYSSPACE;
553 	if (bp->bio_cmd == BIO_READ)
554 		auio.uio_rw = UIO_READ;
555 	else if (bp->bio_cmd == BIO_WRITE)
556 		auio.uio_rw = UIO_WRITE;
557 	else
558 		panic("wrong BIO_OP in mdstart_vnode");
559 	auio.uio_resid = bp->bio_length;
560 	auio.uio_td = td;
561 	/*
562 	 * When reading set IO_DIRECT to try to avoid double-caching
563 	 * the data.  When writing IO_DIRECT is not optimal.
564 	 */
565 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
566 	if (bp->bio_cmd == BIO_READ) {
567 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
568 		error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred);
569 		VOP_UNLOCK(vp, 0);
570 	} else {
571 		(void) vn_start_write(vp, &mp, V_WAIT);
572 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
573 		error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
574 		    sc->cred);
575 		VOP_UNLOCK(vp, 0);
576 		vn_finished_write(mp);
577 	}
578 	VFS_UNLOCK_GIANT(vfslocked);
579 	bp->bio_resid = auio.uio_resid;
580 	return (error);
581 }
582 
583 static int
584 mdstart_swap(struct md_s *sc, struct bio *bp)
585 {
586 	struct sf_buf *sf;
587 	int rv, offs, len, lastend;
588 	vm_pindex_t i, lastp;
589 	vm_page_t m;
590 	u_char *p;
591 
592 	switch (bp->bio_cmd) {
593 	case BIO_READ:
594 	case BIO_WRITE:
595 	case BIO_DELETE:
596 		break;
597 	default:
598 		return (EOPNOTSUPP);
599 	}
600 
601 	p = bp->bio_data;
602 
603 	/*
604 	 * offs is the offset at which to start operating on the
605 	 * next (ie, first) page.  lastp is the last page on
606 	 * which we're going to operate.  lastend is the ending
607 	 * position within that last page (ie, PAGE_SIZE if
608 	 * we're operating on complete aligned pages).
609 	 */
610 	offs = bp->bio_offset % PAGE_SIZE;
611 	lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
612 	lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
613 
614 	rv = VM_PAGER_OK;
615 	VM_OBJECT_LOCK(sc->object);
616 	vm_object_pip_add(sc->object, 1);
617 	for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
618 		len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
619 
620 		m = vm_page_grab(sc->object, i,
621 		    VM_ALLOC_NORMAL|VM_ALLOC_RETRY);
622 		VM_OBJECT_UNLOCK(sc->object);
623 		sched_pin();
624 		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
625 		VM_OBJECT_LOCK(sc->object);
626 		if (bp->bio_cmd == BIO_READ) {
627 			if (m->valid != VM_PAGE_BITS_ALL)
628 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
629 			if (rv == VM_PAGER_ERROR) {
630 				sf_buf_free(sf);
631 				sched_unpin();
632 				vm_page_lock_queues();
633 				vm_page_wakeup(m);
634 				vm_page_unlock_queues();
635 				break;
636 			}
637 			bcopy((void *)(sf_buf_kva(sf) + offs), p, len);
638 		} else if (bp->bio_cmd == BIO_WRITE) {
639 			if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
640 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
641 			if (rv == VM_PAGER_ERROR) {
642 				sf_buf_free(sf);
643 				sched_unpin();
644 				vm_page_lock_queues();
645 				vm_page_wakeup(m);
646 				vm_page_unlock_queues();
647 				break;
648 			}
649 			bcopy(p, (void *)(sf_buf_kva(sf) + offs), len);
650 			m->valid = VM_PAGE_BITS_ALL;
651 #if 0
652 		} else if (bp->bio_cmd == BIO_DELETE) {
653 			if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
654 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
655 			if (rv == VM_PAGER_ERROR) {
656 				sf_buf_free(sf);
657 				sched_unpin();
658 				vm_page_lock_queues();
659 				vm_page_wakeup(m);
660 				vm_page_unlock_queues();
661 				break;
662 			}
663 			bzero((void *)(sf_buf_kva(sf) + offs), len);
664 			vm_page_dirty(m);
665 			m->valid = VM_PAGE_BITS_ALL;
666 #endif
667 		}
668 		sf_buf_free(sf);
669 		sched_unpin();
670 		vm_page_lock_queues();
671 		vm_page_wakeup(m);
672 		vm_page_activate(m);
673 		if (bp->bio_cmd == BIO_WRITE)
674 			vm_page_dirty(m);
675 		vm_page_unlock_queues();
676 
677 		/* Actions on further pages start at offset 0 */
678 		p += PAGE_SIZE - offs;
679 		offs = 0;
680 #if 0
681 if (bootverbose || bp->bio_offset / PAGE_SIZE < 17)
682 printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n",
683     m->wire_count, m->busy,
684     m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i);
685 #endif
686 	}
687 	vm_object_pip_subtract(sc->object, 1);
688 	vm_object_set_writeable_dirty(sc->object);
689 	VM_OBJECT_UNLOCK(sc->object);
690 	return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
691 }
692 
693 static void
694 md_kthread(void *arg)
695 {
696 	struct md_s *sc;
697 	struct bio *bp;
698 	int error;
699 
700 	sc = arg;
701 	thread_lock(curthread);
702 	sched_prio(curthread, PRIBIO);
703 	thread_unlock(curthread);
704 	if (sc->type == MD_VNODE)
705 		curthread->td_pflags |= TDP_NORUNNINGBUF;
706 
707 	for (;;) {
708 		mtx_lock(&sc->queue_mtx);
709 		if (sc->flags & MD_SHUTDOWN) {
710 			sc->flags |= MD_EXITING;
711 			mtx_unlock(&sc->queue_mtx);
712 			kproc_exit(0);
713 		}
714 		bp = bioq_takefirst(&sc->bio_queue);
715 		if (!bp) {
716 			msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
717 			continue;
718 		}
719 		mtx_unlock(&sc->queue_mtx);
720 		if (bp->bio_cmd == BIO_GETATTR) {
721 			if (sc->fwsectors && sc->fwheads &&
722 			    (g_handleattr_int(bp, "GEOM::fwsectors",
723 			    sc->fwsectors) ||
724 			    g_handleattr_int(bp, "GEOM::fwheads",
725 			    sc->fwheads)))
726 				error = -1;
727 			else
728 				error = EOPNOTSUPP;
729 		} else {
730 			error = sc->start(sc, bp);
731 		}
732 
733 		if (error != -1) {
734 			bp->bio_completed = bp->bio_length;
735 			g_io_deliver(bp, error);
736 			if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
737 				devstat_end_transaction_bio(sc->devstat, bp);
738 		}
739 	}
740 }
741 
742 static struct md_s *
743 mdfind(int unit)
744 {
745 	struct md_s *sc;
746 
747 	LIST_FOREACH(sc, &md_softc_list, list) {
748 		if (sc->unit == unit)
749 			break;
750 	}
751 	return (sc);
752 }
753 
754 static struct md_s *
755 mdnew(int unit, int *errp, enum md_types type)
756 {
757 	struct md_s *sc, *sc2;
758 	int error, max = -1;
759 
760 	*errp = 0;
761 	LIST_FOREACH(sc2, &md_softc_list, list) {
762 		if (unit == sc2->unit) {
763 			*errp = EBUSY;
764 			return (NULL);
765 		}
766 		if (unit == -1 && sc2->unit > max)
767 			max = sc2->unit;
768 	}
769 	if (unit == -1)
770 		unit = max + 1;
771 	sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
772 	sc->type = type;
773 	bioq_init(&sc->bio_queue);
774 	mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
775 	sc->unit = unit;
776 	sprintf(sc->name, "md%d", unit);
777 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
778 	error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
779 	if (error == 0)
780 		return (sc);
781 	LIST_REMOVE(sc, list);
782 	mtx_destroy(&sc->queue_mtx);
783 	free(sc, M_MD);
784 	*errp = error;
785 	return (NULL);
786 }
787 
788 static void
789 mdinit(struct md_s *sc)
790 {
791 
792 	struct g_geom *gp;
793 	struct g_provider *pp;
794 
795 	g_topology_lock();
796 	gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
797 	gp->softc = sc;
798 	pp = g_new_providerf(gp, "md%d", sc->unit);
799 	pp->mediasize = sc->mediasize;
800 	pp->sectorsize = sc->sectorsize;
801 	sc->gp = gp;
802 	sc->pp = pp;
803 	g_error_provider(pp, 0);
804 	g_topology_unlock();
805 	sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
806 	    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
807 }
808 
809 /*
810  * XXX: we should check that the range they feed us is mapped.
811  * XXX: we should implement read-only.
812  */
813 
814 static int
815 mdcreate_preload(struct md_s *sc, struct md_ioctl *mdio)
816 {
817 
818 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_FORCE))
819 		return (EINVAL);
820 	sc->flags = mdio->md_options & MD_FORCE;
821 	/* Cast to pointer size, then to pointer to avoid warning */
822 	sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base;
823 	sc->pl_len = (size_t)sc->mediasize;
824 	return (0);
825 }
826 
827 
828 static int
829 mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio)
830 {
831 	uintptr_t sp;
832 	int error;
833 	off_t u;
834 
835 	error = 0;
836 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
837 		return (EINVAL);
838 	if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize))
839 		return (EINVAL);
840 	/* Compression doesn't make sense if we have reserved space */
841 	if (mdio->md_options & MD_RESERVE)
842 		mdio->md_options &= ~MD_COMPRESS;
843 	if (mdio->md_fwsectors != 0)
844 		sc->fwsectors = mdio->md_fwsectors;
845 	if (mdio->md_fwheads != 0)
846 		sc->fwheads = mdio->md_fwheads;
847 	sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE);
848 	sc->indir = dimension(sc->mediasize / sc->sectorsize);
849 	sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
850 	    0x1ff, 0);
851 	if (mdio->md_options & MD_RESERVE) {
852 		off_t nsectors;
853 
854 		nsectors = sc->mediasize / sc->sectorsize;
855 		for (u = 0; u < nsectors; u++) {
856 			sp = (uintptr_t)uma_zalloc(sc->uma, M_NOWAIT | M_ZERO);
857 			if (sp != 0)
858 				error = s_write(sc->indir, u, sp);
859 			else
860 				error = ENOMEM;
861 			if (error != 0)
862 				break;
863 		}
864 	}
865 	return (error);
866 }
867 
868 
869 static int
870 mdsetcred(struct md_s *sc, struct ucred *cred)
871 {
872 	char *tmpbuf;
873 	int error = 0;
874 
875 	/*
876 	 * Set credits in our softc
877 	 */
878 
879 	if (sc->cred)
880 		crfree(sc->cred);
881 	sc->cred = crhold(cred);
882 
883 	/*
884 	 * Horrible kludge to establish credentials for NFS  XXX.
885 	 */
886 
887 	if (sc->vnode) {
888 		struct uio auio;
889 		struct iovec aiov;
890 
891 		tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
892 		bzero(&auio, sizeof(auio));
893 
894 		aiov.iov_base = tmpbuf;
895 		aiov.iov_len = sc->sectorsize;
896 		auio.uio_iov = &aiov;
897 		auio.uio_iovcnt = 1;
898 		auio.uio_offset = 0;
899 		auio.uio_rw = UIO_READ;
900 		auio.uio_segflg = UIO_SYSSPACE;
901 		auio.uio_resid = aiov.iov_len;
902 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
903 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
904 		VOP_UNLOCK(sc->vnode, 0);
905 		free(tmpbuf, M_TEMP);
906 	}
907 	return (error);
908 }
909 
910 static int
911 mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
912 {
913 	struct vattr vattr;
914 	struct nameidata nd;
915 	int error, flags, vfslocked;
916 
917 	error = copyinstr(mdio->md_file, sc->file, sizeof(sc->file), NULL);
918 	if (error != 0)
919 		return (error);
920 	flags = FREAD|FWRITE;
921 	/*
922 	 * If the user specified that this is a read only device, unset the
923 	 * FWRITE mask before trying to open the backing store.
924 	 */
925 	if ((mdio->md_options & MD_READONLY) != 0)
926 		flags &= ~FWRITE;
927 	NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, sc->file, td);
928 	error = vn_open(&nd, &flags, 0, NULL);
929 	if (error != 0)
930 		return (error);
931 	vfslocked = NDHASGIANT(&nd);
932 	NDFREE(&nd, NDF_ONLY_PNBUF);
933 	if (nd.ni_vp->v_type != VREG ||
934 	    (error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred))) {
935 		VOP_UNLOCK(nd.ni_vp, 0);
936 		(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
937 		VFS_UNLOCK_GIANT(vfslocked);
938 		return (error ? error : EINVAL);
939 	}
940 	nd.ni_vp->v_vflag |= VV_MD;
941 	VOP_UNLOCK(nd.ni_vp, 0);
942 
943 	if (mdio->md_fwsectors != 0)
944 		sc->fwsectors = mdio->md_fwsectors;
945 	if (mdio->md_fwheads != 0)
946 		sc->fwheads = mdio->md_fwheads;
947 	sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC);
948 	if (!(flags & FWRITE))
949 		sc->flags |= MD_READONLY;
950 	sc->vnode = nd.ni_vp;
951 
952 	error = mdsetcred(sc, td->td_ucred);
953 	if (error != 0) {
954 		sc->vnode = NULL;
955 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
956 		nd.ni_vp->v_vflag &= ~VV_MD;
957 		VOP_UNLOCK(nd.ni_vp, 0);
958 		(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
959 		VFS_UNLOCK_GIANT(vfslocked);
960 		return (error);
961 	}
962 	VFS_UNLOCK_GIANT(vfslocked);
963 	return (0);
964 }
965 
966 static int
967 mddestroy(struct md_s *sc, struct thread *td)
968 {
969 	int vfslocked;
970 
971 	if (sc->gp) {
972 		sc->gp->softc = NULL;
973 		g_topology_lock();
974 		g_wither_geom(sc->gp, ENXIO);
975 		g_topology_unlock();
976 		sc->gp = NULL;
977 		sc->pp = NULL;
978 	}
979 	if (sc->devstat) {
980 		devstat_remove_entry(sc->devstat);
981 		sc->devstat = NULL;
982 	}
983 	mtx_lock(&sc->queue_mtx);
984 	sc->flags |= MD_SHUTDOWN;
985 	wakeup(sc);
986 	while (!(sc->flags & MD_EXITING))
987 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
988 	mtx_unlock(&sc->queue_mtx);
989 	mtx_destroy(&sc->queue_mtx);
990 	if (sc->vnode != NULL) {
991 		vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount);
992 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
993 		sc->vnode->v_vflag &= ~VV_MD;
994 		VOP_UNLOCK(sc->vnode, 0);
995 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
996 		    FREAD : (FREAD|FWRITE), sc->cred, td);
997 		VFS_UNLOCK_GIANT(vfslocked);
998 	}
999 	if (sc->cred != NULL)
1000 		crfree(sc->cred);
1001 	if (sc->object != NULL)
1002 		vm_object_deallocate(sc->object);
1003 	if (sc->indir)
1004 		destroy_indir(sc, sc->indir);
1005 	if (sc->uma)
1006 		uma_zdestroy(sc->uma);
1007 
1008 	LIST_REMOVE(sc, list);
1009 	free(sc, M_MD);
1010 	return (0);
1011 }
1012 
1013 static int
1014 mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
1015 {
1016 	vm_ooffset_t npage;
1017 	int error;
1018 
1019 	/*
1020 	 * Range check.  Disallow negative sizes or any size less then the
1021 	 * size of a page.  Then round to a page.
1022 	 */
1023 	if (sc->mediasize == 0 || (sc->mediasize % PAGE_SIZE) != 0)
1024 		return (EDOM);
1025 
1026 	/*
1027 	 * Allocate an OBJT_SWAP object.
1028 	 *
1029 	 * Note the truncation.
1030 	 */
1031 
1032 	npage = mdio->md_mediasize / PAGE_SIZE;
1033 	if (mdio->md_fwsectors != 0)
1034 		sc->fwsectors = mdio->md_fwsectors;
1035 	if (mdio->md_fwheads != 0)
1036 		sc->fwheads = mdio->md_fwheads;
1037 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1038 	    VM_PROT_DEFAULT, 0);
1039 	if (sc->object == NULL)
1040 		return (ENOMEM);
1041 	sc->flags = mdio->md_options & MD_FORCE;
1042 	if (mdio->md_options & MD_RESERVE) {
1043 		if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1044 			vm_object_deallocate(sc->object);
1045 			sc->object = NULL;
1046 			return (EDOM);
1047 		}
1048 	}
1049 	error = mdsetcred(sc, td->td_ucred);
1050 	if (error != 0) {
1051 		vm_object_deallocate(sc->object);
1052 		sc->object = NULL;
1053 	}
1054 	return (error);
1055 }
1056 
1057 
1058 static int
1059 xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1060 {
1061 	struct md_ioctl *mdio;
1062 	struct md_s *sc;
1063 	int error, i;
1064 
1065 	if (md_debug)
1066 		printf("mdctlioctl(%s %lx %p %x %p)\n",
1067 			devtoname(dev), cmd, addr, flags, td);
1068 
1069 	mdio = (struct md_ioctl *)addr;
1070 	if (mdio->md_version != MDIOVERSION)
1071 		return (EINVAL);
1072 
1073 	/*
1074 	 * We assert the version number in the individual ioctl
1075 	 * handlers instead of out here because (a) it is possible we
1076 	 * may add another ioctl in the future which doesn't read an
1077 	 * mdio, and (b) the correct return value for an unknown ioctl
1078 	 * is ENOIOCTL, not EINVAL.
1079 	 */
1080 	error = 0;
1081 	switch (cmd) {
1082 	case MDIOCATTACH:
1083 		switch (mdio->md_type) {
1084 		case MD_MALLOC:
1085 		case MD_PRELOAD:
1086 		case MD_VNODE:
1087 		case MD_SWAP:
1088 			break;
1089 		default:
1090 			return (EINVAL);
1091 		}
1092 		if (mdio->md_options & MD_AUTOUNIT)
1093 			sc = mdnew(-1, &error, mdio->md_type);
1094 		else
1095 			sc = mdnew(mdio->md_unit, &error, mdio->md_type);
1096 		if (sc == NULL)
1097 			return (error);
1098 		if (mdio->md_options & MD_AUTOUNIT)
1099 			mdio->md_unit = sc->unit;
1100 		sc->mediasize = mdio->md_mediasize;
1101 		if (mdio->md_sectorsize == 0)
1102 			sc->sectorsize = DEV_BSIZE;
1103 		else
1104 			sc->sectorsize = mdio->md_sectorsize;
1105 		error = EDOOFUS;
1106 		switch (sc->type) {
1107 		case MD_MALLOC:
1108 			sc->start = mdstart_malloc;
1109 			error = mdcreate_malloc(sc, mdio);
1110 			break;
1111 		case MD_PRELOAD:
1112 			sc->start = mdstart_preload;
1113 			error = mdcreate_preload(sc, mdio);
1114 			break;
1115 		case MD_VNODE:
1116 			sc->start = mdstart_vnode;
1117 			error = mdcreate_vnode(sc, mdio, td);
1118 			break;
1119 		case MD_SWAP:
1120 			sc->start = mdstart_swap;
1121 			error = mdcreate_swap(sc, mdio, td);
1122 			break;
1123 		}
1124 		if (error != 0) {
1125 			mddestroy(sc, td);
1126 			return (error);
1127 		}
1128 
1129 		/* Prune off any residual fractional sector */
1130 		i = sc->mediasize % sc->sectorsize;
1131 		sc->mediasize -= i;
1132 
1133 		mdinit(sc);
1134 		return (0);
1135 	case MDIOCDETACH:
1136 		if (mdio->md_mediasize != 0 || mdio->md_options != 0)
1137 			return (EINVAL);
1138 
1139 		sc = mdfind(mdio->md_unit);
1140 		if (sc == NULL)
1141 			return (ENOENT);
1142 		if (sc->opencount != 0 && !(sc->flags & MD_FORCE))
1143 			return (EBUSY);
1144 		return (mddestroy(sc, td));
1145 	case MDIOCQUERY:
1146 		sc = mdfind(mdio->md_unit);
1147 		if (sc == NULL)
1148 			return (ENOENT);
1149 		mdio->md_type = sc->type;
1150 		mdio->md_options = sc->flags;
1151 		mdio->md_mediasize = sc->mediasize;
1152 		mdio->md_sectorsize = sc->sectorsize;
1153 		if (sc->type == MD_VNODE)
1154 			error = copyout(sc->file, mdio->md_file,
1155 			    strlen(sc->file) + 1);
1156 		return (error);
1157 	case MDIOCLIST:
1158 		i = 1;
1159 		LIST_FOREACH(sc, &md_softc_list, list) {
1160 			if (i == MDNPAD - 1)
1161 				mdio->md_pad[i] = -1;
1162 			else
1163 				mdio->md_pad[i++] = sc->unit;
1164 		}
1165 		mdio->md_pad[0] = i - 1;
1166 		return (0);
1167 	default:
1168 		return (ENOIOCTL);
1169 	};
1170 }
1171 
1172 static int
1173 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1174 {
1175 	int error;
1176 
1177 	sx_xlock(&md_sx);
1178 	error = xmdctlioctl(dev, cmd, addr, flags, td);
1179 	sx_xunlock(&md_sx);
1180 	return (error);
1181 }
1182 
1183 static void
1184 md_preloaded(u_char *image, size_t length)
1185 {
1186 	struct md_s *sc;
1187 	int error;
1188 
1189 	sc = mdnew(-1, &error, MD_PRELOAD);
1190 	if (sc == NULL)
1191 		return;
1192 	sc->mediasize = length;
1193 	sc->sectorsize = DEV_BSIZE;
1194 	sc->pl_ptr = image;
1195 	sc->pl_len = length;
1196 	sc->start = mdstart_preload;
1197 #ifdef MD_ROOT
1198 	if (sc->unit == 0)
1199 		rootdevnames[0] = "ufs:/dev/md0";
1200 #endif
1201 	mdinit(sc);
1202 }
1203 
1204 static void
1205 g_md_init(struct g_class *mp __unused)
1206 {
1207 
1208 	caddr_t mod;
1209 	caddr_t c;
1210 	u_char *ptr, *name, *type;
1211 	unsigned len;
1212 
1213 	mod = NULL;
1214 	sx_init(&md_sx, "MD config lock");
1215 	g_topology_unlock();
1216 #ifdef MD_ROOT_SIZE
1217 	sx_xlock(&md_sx);
1218 	md_preloaded(mfs_root.start, sizeof(mfs_root.start));
1219 	sx_xunlock(&md_sx);
1220 #endif
1221 	/* XXX: are preload_* static or do they need Giant ? */
1222 	while ((mod = preload_search_next_name(mod)) != NULL) {
1223 		name = (char *)preload_search_info(mod, MODINFO_NAME);
1224 		if (name == NULL)
1225 			continue;
1226 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
1227 		if (type == NULL)
1228 			continue;
1229 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
1230 			continue;
1231 		c = preload_search_info(mod, MODINFO_ADDR);
1232 		ptr = *(u_char **)c;
1233 		c = preload_search_info(mod, MODINFO_SIZE);
1234 		len = *(size_t *)c;
1235 		printf("%s%d: Preloaded image <%s> %d bytes at %p\n",
1236 		    MD_NAME, mdunits, name, len, ptr);
1237 		sx_xlock(&md_sx);
1238 		md_preloaded(ptr, len);
1239 		sx_xunlock(&md_sx);
1240 	}
1241 	status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
1242 	    0600, MDCTL_NAME);
1243 	g_topology_lock();
1244 }
1245 
1246 static void
1247 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1248     struct g_consumer *cp __unused, struct g_provider *pp)
1249 {
1250 	struct md_s *mp;
1251 	char *type;
1252 
1253 	mp = gp->softc;
1254 	if (mp == NULL)
1255 		return;
1256 
1257 	switch (mp->type) {
1258 	case MD_MALLOC:
1259 		type = "malloc";
1260 		break;
1261 	case MD_PRELOAD:
1262 		type = "preload";
1263 		break;
1264 	case MD_VNODE:
1265 		type = "vnode";
1266 		break;
1267 	case MD_SWAP:
1268 		type = "swap";
1269 		break;
1270 	default:
1271 		type = "unknown";
1272 		break;
1273 	}
1274 
1275 	if (pp != NULL) {
1276 		if (indent == NULL) {
1277 			sbuf_printf(sb, " u %d", mp->unit);
1278 			sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
1279 			sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
1280 			sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
1281 			sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
1282 			sbuf_printf(sb, " t %s", type);
1283 			if (mp->type == MD_VNODE && mp->vnode != NULL)
1284 				sbuf_printf(sb, " file %s", mp->file);
1285 		} else {
1286 			sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
1287 			    mp->unit);
1288 			sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
1289 			    indent, (uintmax_t) mp->sectorsize);
1290 			sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
1291 			    indent, (uintmax_t) mp->fwheads);
1292 			sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
1293 			    indent, (uintmax_t) mp->fwsectors);
1294 			sbuf_printf(sb, "%s<length>%ju</length>\n",
1295 			    indent, (uintmax_t) mp->mediasize);
1296 			sbuf_printf(sb, "%s<type>%s</type>\n", indent,
1297 			    type);
1298 			if (mp->type == MD_VNODE && mp->vnode != NULL)
1299 				sbuf_printf(sb, "%s<file>%s</file>\n",
1300 				    indent, mp->file);
1301 		}
1302 	}
1303 }
1304 
1305 static void
1306 g_md_fini(struct g_class *mp __unused)
1307 {
1308 
1309 	sx_destroy(&md_sx);
1310 	if (status_dev != NULL)
1311 		destroy_dev(status_dev);
1312 }
1313