xref: /freebsd/sys/dev/md/md.c (revision 0bc7cf6fde4083c1a0cdf29b26075699028909e0)
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/sbuf.h>
79 #include <sys/sched.h>
80 #include <sys/sf_buf.h>
81 #include <sys/sysctl.h>
82 #include <sys/vnode.h>
83 
84 #include <geom/geom.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/swap_pager.h>
91 #include <vm/uma.h>
92 
93 #include <machine/vmparam.h>
94 
95 #define MD_MODVER 1
96 
97 #define MD_SHUTDOWN	0x10000		/* Tell worker thread to terminate. */
98 #define	MD_EXITING	0x20000		/* Worker thread is exiting. */
99 
100 #ifndef MD_NSECT
101 #define MD_NSECT (10000 * 2)
102 #endif
103 
104 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
105 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
106 
107 static int md_debug;
108 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, "");
109 static int md_malloc_wait;
110 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0, "");
111 
112 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE)
113 /*
114  * Preloaded image gets put here.
115  * Applications that patch the object with the image can determine
116  * the size looking at the start and end markers (strings),
117  * so we want them contiguous.
118  */
119 static struct {
120 	u_char start[MD_ROOT_SIZE*1024];
121 	u_char end[128];
122 } mfs_root = {
123 	.start = "MFS Filesystem goes here",
124 	.end = "MFS Filesystem had better STOP here",
125 };
126 #endif
127 
128 static g_init_t g_md_init;
129 static g_fini_t g_md_fini;
130 static g_start_t g_md_start;
131 static g_access_t g_md_access;
132 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
133     struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
134 
135 static int mdunits;
136 static struct cdev *status_dev = 0;
137 static struct sx md_sx;
138 static struct unrhdr *md_uh;
139 
140 static d_ioctl_t mdctlioctl;
141 
142 static struct cdevsw mdctl_cdevsw = {
143 	.d_version =	D_VERSION,
144 	.d_ioctl =	mdctlioctl,
145 	.d_name =	MD_NAME,
146 };
147 
148 struct g_class g_md_class = {
149 	.name = "MD",
150 	.version = G_VERSION,
151 	.init = g_md_init,
152 	.fini = g_md_fini,
153 	.start = g_md_start,
154 	.access = g_md_access,
155 	.dumpconf = g_md_dumpconf,
156 };
157 
158 DECLARE_GEOM_CLASS(g_md_class, g_md);
159 
160 
161 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
162 
163 #define NINDIR	(PAGE_SIZE / sizeof(uintptr_t))
164 #define NMASK	(NINDIR-1)
165 static int nshift;
166 
167 struct indir {
168 	uintptr_t	*array;
169 	u_int		total;
170 	u_int		used;
171 	u_int		shift;
172 };
173 
174 struct md_s {
175 	int unit;
176 	LIST_ENTRY(md_s) list;
177 	struct bio_queue_head bio_queue;
178 	struct mtx queue_mtx;
179 	struct cdev *dev;
180 	enum md_types type;
181 	off_t mediasize;
182 	unsigned sectorsize;
183 	unsigned opencount;
184 	unsigned fwheads;
185 	unsigned fwsectors;
186 	unsigned flags;
187 	char name[20];
188 	struct proc *procp;
189 	struct g_geom *gp;
190 	struct g_provider *pp;
191 	int (*start)(struct md_s *sc, struct bio *bp);
192 	struct devstat *devstat;
193 
194 	/* MD_MALLOC related fields */
195 	struct indir *indir;
196 	uma_zone_t uma;
197 
198 	/* MD_PRELOAD related fields */
199 	u_char *pl_ptr;
200 	size_t pl_len;
201 
202 	/* MD_VNODE related fields */
203 	struct vnode *vnode;
204 	char file[PATH_MAX];
205 	struct ucred *cred;
206 
207 	/* MD_SWAP related fields */
208 	vm_object_t object;
209 };
210 
211 static struct indir *
212 new_indir(u_int shift)
213 {
214 	struct indir *ip;
215 
216 	ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
217 	    | M_ZERO);
218 	if (ip == NULL)
219 		return (NULL);
220 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
221 	    M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
222 	if (ip->array == NULL) {
223 		free(ip, M_MD);
224 		return (NULL);
225 	}
226 	ip->total = NINDIR;
227 	ip->shift = shift;
228 	return (ip);
229 }
230 
231 static void
232 del_indir(struct indir *ip)
233 {
234 
235 	free(ip->array, M_MDSECT);
236 	free(ip, M_MD);
237 }
238 
239 static void
240 destroy_indir(struct md_s *sc, struct indir *ip)
241 {
242 	int i;
243 
244 	for (i = 0; i < NINDIR; i++) {
245 		if (!ip->array[i])
246 			continue;
247 		if (ip->shift)
248 			destroy_indir(sc, (struct indir*)(ip->array[i]));
249 		else if (ip->array[i] > 255)
250 			uma_zfree(sc->uma, (void *)(ip->array[i]));
251 	}
252 	del_indir(ip);
253 }
254 
255 /*
256  * This function does the math and allocates the top level "indir" structure
257  * for a device of "size" sectors.
258  */
259 
260 static struct indir *
261 dimension(off_t size)
262 {
263 	off_t rcnt;
264 	struct indir *ip;
265 	int layer;
266 
267 	rcnt = size;
268 	layer = 0;
269 	while (rcnt > NINDIR) {
270 		rcnt /= NINDIR;
271 		layer++;
272 	}
273 
274 	/*
275 	 * XXX: the top layer is probably not fully populated, so we allocate
276 	 * too much space for ip->array in here.
277 	 */
278 	ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
279 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
280 	    M_MDSECT, M_WAITOK | M_ZERO);
281 	ip->total = NINDIR;
282 	ip->shift = layer * nshift;
283 	return (ip);
284 }
285 
286 /*
287  * Read a given sector
288  */
289 
290 static uintptr_t
291 s_read(struct indir *ip, off_t offset)
292 {
293 	struct indir *cip;
294 	int idx;
295 	uintptr_t up;
296 
297 	if (md_debug > 1)
298 		printf("s_read(%jd)\n", (intmax_t)offset);
299 	up = 0;
300 	for (cip = ip; cip != NULL;) {
301 		if (cip->shift) {
302 			idx = (offset >> cip->shift) & NMASK;
303 			up = cip->array[idx];
304 			cip = (struct indir *)up;
305 			continue;
306 		}
307 		idx = offset & NMASK;
308 		return (cip->array[idx]);
309 	}
310 	return (0);
311 }
312 
313 /*
314  * Write a given sector, prune the tree if the value is 0
315  */
316 
317 static int
318 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
319 {
320 	struct indir *cip, *lip[10];
321 	int idx, li;
322 	uintptr_t up;
323 
324 	if (md_debug > 1)
325 		printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
326 	up = 0;
327 	li = 0;
328 	cip = ip;
329 	for (;;) {
330 		lip[li++] = cip;
331 		if (cip->shift) {
332 			idx = (offset >> cip->shift) & NMASK;
333 			up = cip->array[idx];
334 			if (up != 0) {
335 				cip = (struct indir *)up;
336 				continue;
337 			}
338 			/* Allocate branch */
339 			cip->array[idx] =
340 			    (uintptr_t)new_indir(cip->shift - nshift);
341 			if (cip->array[idx] == 0)
342 				return (ENOSPC);
343 			cip->used++;
344 			up = cip->array[idx];
345 			cip = (struct indir *)up;
346 			continue;
347 		}
348 		/* leafnode */
349 		idx = offset & NMASK;
350 		up = cip->array[idx];
351 		if (up != 0)
352 			cip->used--;
353 		cip->array[idx] = ptr;
354 		if (ptr != 0)
355 			cip->used++;
356 		break;
357 	}
358 	if (cip->used != 0 || li == 1)
359 		return (0);
360 	li--;
361 	while (cip->used == 0 && cip != ip) {
362 		li--;
363 		idx = (offset >> lip[li]->shift) & NMASK;
364 		up = lip[li]->array[idx];
365 		KASSERT(up == (uintptr_t)cip, ("md screwed up"));
366 		del_indir(cip);
367 		lip[li]->array[idx] = 0;
368 		lip[li]->used--;
369 		cip = lip[li];
370 	}
371 	return (0);
372 }
373 
374 
375 static int
376 g_md_access(struct g_provider *pp, int r, int w, int e)
377 {
378 	struct md_s *sc;
379 
380 	sc = pp->geom->softc;
381 	if (sc == NULL) {
382 		if (r <= 0 && w <= 0 && e <= 0)
383 			return (0);
384 		return (ENXIO);
385 	}
386 	r += pp->acr;
387 	w += pp->acw;
388 	e += pp->ace;
389 	if ((sc->flags & MD_READONLY) != 0 && w > 0)
390 		return (EROFS);
391 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
392 		sc->opencount = 1;
393 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
394 		sc->opencount = 0;
395 	}
396 	return (0);
397 }
398 
399 static void
400 g_md_start(struct bio *bp)
401 {
402 	struct md_s *sc;
403 
404 	sc = bp->bio_to->geom->softc;
405 	if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
406 		devstat_start_transaction_bio(sc->devstat, bp);
407 	mtx_lock(&sc->queue_mtx);
408 	bioq_disksort(&sc->bio_queue, bp);
409 	mtx_unlock(&sc->queue_mtx);
410 	wakeup(sc);
411 }
412 
413 static int
414 mdstart_malloc(struct md_s *sc, struct bio *bp)
415 {
416 	int i, error;
417 	u_char *dst;
418 	off_t secno, nsec, uc;
419 	uintptr_t sp, osp;
420 
421 	switch (bp->bio_cmd) {
422 	case BIO_READ:
423 	case BIO_WRITE:
424 	case BIO_DELETE:
425 		break;
426 	default:
427 		return (EOPNOTSUPP);
428 	}
429 
430 	nsec = bp->bio_length / sc->sectorsize;
431 	secno = bp->bio_offset / sc->sectorsize;
432 	dst = bp->bio_data;
433 	error = 0;
434 	while (nsec--) {
435 		osp = s_read(sc->indir, secno);
436 		if (bp->bio_cmd == BIO_DELETE) {
437 			if (osp != 0)
438 				error = s_write(sc->indir, secno, 0);
439 		} else if (bp->bio_cmd == BIO_READ) {
440 			if (osp == 0)
441 				bzero(dst, sc->sectorsize);
442 			else if (osp <= 255)
443 				memset(dst, osp, sc->sectorsize);
444 			else {
445 				bcopy((void *)osp, dst, sc->sectorsize);
446 				cpu_flush_dcache(dst, sc->sectorsize);
447 			}
448 			osp = 0;
449 		} else if (bp->bio_cmd == BIO_WRITE) {
450 			if (sc->flags & MD_COMPRESS) {
451 				uc = dst[0];
452 				for (i = 1; i < sc->sectorsize; i++)
453 					if (dst[i] != uc)
454 						break;
455 			} else {
456 				i = 0;
457 				uc = 0;
458 			}
459 			if (i == sc->sectorsize) {
460 				if (osp != uc)
461 					error = s_write(sc->indir, secno, uc);
462 			} else {
463 				if (osp <= 255) {
464 					sp = (uintptr_t)uma_zalloc(sc->uma,
465 					    md_malloc_wait ? M_WAITOK :
466 					    M_NOWAIT);
467 					if (sp == 0) {
468 						error = ENOSPC;
469 						break;
470 					}
471 					bcopy(dst, (void *)sp, sc->sectorsize);
472 					error = s_write(sc->indir, secno, sp);
473 				} else {
474 					bcopy(dst, (void *)osp, sc->sectorsize);
475 					osp = 0;
476 				}
477 			}
478 		} else {
479 			error = EOPNOTSUPP;
480 		}
481 		if (osp > 255)
482 			uma_zfree(sc->uma, (void*)osp);
483 		if (error != 0)
484 			break;
485 		secno++;
486 		dst += sc->sectorsize;
487 	}
488 	bp->bio_resid = 0;
489 	return (error);
490 }
491 
492 static int
493 mdstart_preload(struct md_s *sc, struct bio *bp)
494 {
495 
496 	switch (bp->bio_cmd) {
497 	case BIO_READ:
498 		bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data,
499 		    bp->bio_length);
500 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
501 		break;
502 	case BIO_WRITE:
503 		bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset,
504 		    bp->bio_length);
505 		break;
506 	}
507 	bp->bio_resid = 0;
508 	return (0);
509 }
510 
511 static int
512 mdstart_vnode(struct md_s *sc, struct bio *bp)
513 {
514 	int error, vfslocked;
515 	struct uio auio;
516 	struct iovec aiov;
517 	struct mount *mp;
518 	struct vnode *vp;
519 	struct thread *td;
520 	off_t end, zerosize;
521 
522 	switch (bp->bio_cmd) {
523 	case BIO_READ:
524 	case BIO_WRITE:
525 	case BIO_DELETE:
526 	case BIO_FLUSH:
527 		break;
528 	default:
529 		return (EOPNOTSUPP);
530 	}
531 
532 	td = curthread;
533 	vp = sc->vnode;
534 
535 	/*
536 	 * VNODE I/O
537 	 *
538 	 * If an error occurs, we set BIO_ERROR but we do not set
539 	 * B_INVAL because (for a write anyway), the buffer is
540 	 * still valid.
541 	 */
542 
543 	if (bp->bio_cmd == BIO_FLUSH) {
544 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
545 		(void) vn_start_write(vp, &mp, V_WAIT);
546 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
547 		error = VOP_FSYNC(vp, MNT_WAIT, td);
548 		VOP_UNLOCK(vp, 0);
549 		vn_finished_write(mp);
550 		VFS_UNLOCK_GIANT(vfslocked);
551 		return (error);
552 	}
553 
554 	bzero(&auio, sizeof(auio));
555 
556 	/*
557 	 * Special case for BIO_DELETE.  On the surface, this is very
558 	 * similar to BIO_WRITE, except that we write from our own
559 	 * fixed-length buffer, so we have to loop.  The net result is
560 	 * that the two cases end up having very little in common.
561 	 */
562 	if (bp->bio_cmd == BIO_DELETE) {
563 		zerosize = ZERO_REGION_SIZE -
564 		    (ZERO_REGION_SIZE % sc->sectorsize);
565 		auio.uio_iov = &aiov;
566 		auio.uio_iovcnt = 1;
567 		auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
568 		auio.uio_segflg = UIO_SYSSPACE;
569 		auio.uio_rw = UIO_WRITE;
570 		auio.uio_td = td;
571 		end = bp->bio_offset + bp->bio_length;
572 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
573 		(void) vn_start_write(vp, &mp, V_WAIT);
574 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
575 		error = 0;
576 		while (auio.uio_offset < end) {
577 			aiov.iov_base = __DECONST(void *, zero_region);
578 			aiov.iov_len = end - auio.uio_offset;
579 			if (aiov.iov_len > zerosize)
580 				aiov.iov_len = zerosize;
581 			auio.uio_resid = aiov.iov_len;
582 			error = VOP_WRITE(vp, &auio,
583 			    sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred);
584 			if (error != 0)
585 				break;
586 		}
587 		VOP_UNLOCK(vp, 0);
588 		vn_finished_write(mp);
589 		bp->bio_resid = end - auio.uio_offset;
590 		VFS_UNLOCK_GIANT(vfslocked);
591 		return (error);
592 	}
593 
594 	aiov.iov_base = bp->bio_data;
595 	aiov.iov_len = bp->bio_length;
596 	auio.uio_iov = &aiov;
597 	auio.uio_iovcnt = 1;
598 	auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
599 	auio.uio_segflg = UIO_SYSSPACE;
600 	if (bp->bio_cmd == BIO_READ)
601 		auio.uio_rw = UIO_READ;
602 	else if (bp->bio_cmd == BIO_WRITE)
603 		auio.uio_rw = UIO_WRITE;
604 	else
605 		panic("wrong BIO_OP in mdstart_vnode");
606 	auio.uio_resid = bp->bio_length;
607 	auio.uio_td = td;
608 	/*
609 	 * When reading set IO_DIRECT to try to avoid double-caching
610 	 * the data.  When writing IO_DIRECT is not optimal.
611 	 */
612 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
613 	if (bp->bio_cmd == BIO_READ) {
614 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
615 		error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred);
616 		VOP_UNLOCK(vp, 0);
617 	} else {
618 		(void) vn_start_write(vp, &mp, V_WAIT);
619 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
620 		error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
621 		    sc->cred);
622 		VOP_UNLOCK(vp, 0);
623 		vn_finished_write(mp);
624 	}
625 	VFS_UNLOCK_GIANT(vfslocked);
626 	bp->bio_resid = auio.uio_resid;
627 	return (error);
628 }
629 
630 static int
631 mdstart_swap(struct md_s *sc, struct bio *bp)
632 {
633 	struct sf_buf *sf;
634 	int rv, offs, len, lastend;
635 	vm_pindex_t i, lastp;
636 	vm_page_t m;
637 	u_char *p;
638 
639 	switch (bp->bio_cmd) {
640 	case BIO_READ:
641 	case BIO_WRITE:
642 	case BIO_DELETE:
643 		break;
644 	default:
645 		return (EOPNOTSUPP);
646 	}
647 
648 	p = bp->bio_data;
649 
650 	/*
651 	 * offs is the offset at which to start operating on the
652 	 * next (ie, first) page.  lastp is the last page on
653 	 * which we're going to operate.  lastend is the ending
654 	 * position within that last page (ie, PAGE_SIZE if
655 	 * we're operating on complete aligned pages).
656 	 */
657 	offs = bp->bio_offset % PAGE_SIZE;
658 	lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
659 	lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
660 
661 	rv = VM_PAGER_OK;
662 	VM_OBJECT_LOCK(sc->object);
663 	vm_object_pip_add(sc->object, 1);
664 	for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
665 		len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
666 
667 		m = vm_page_grab(sc->object, i,
668 		    VM_ALLOC_NORMAL|VM_ALLOC_RETRY);
669 		VM_OBJECT_UNLOCK(sc->object);
670 		sched_pin();
671 		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
672 		VM_OBJECT_LOCK(sc->object);
673 		if (bp->bio_cmd == BIO_READ) {
674 			if (m->valid != VM_PAGE_BITS_ALL)
675 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
676 			if (rv == VM_PAGER_ERROR) {
677 				sf_buf_free(sf);
678 				sched_unpin();
679 				vm_page_wakeup(m);
680 				break;
681 			}
682 			bcopy((void *)(sf_buf_kva(sf) + offs), p, len);
683 			cpu_flush_dcache(p, len);
684 		} else if (bp->bio_cmd == BIO_WRITE) {
685 			if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
686 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
687 			if (rv == VM_PAGER_ERROR) {
688 				sf_buf_free(sf);
689 				sched_unpin();
690 				vm_page_wakeup(m);
691 				break;
692 			}
693 			bcopy(p, (void *)(sf_buf_kva(sf) + offs), len);
694 			m->valid = VM_PAGE_BITS_ALL;
695 		} else if (bp->bio_cmd == BIO_DELETE) {
696 			if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
697 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
698 			if (rv == VM_PAGER_ERROR) {
699 				sf_buf_free(sf);
700 				sched_unpin();
701 				vm_page_wakeup(m);
702 				break;
703 			}
704 			if (len != PAGE_SIZE) {
705 				bzero((void *)(sf_buf_kva(sf) + offs), len);
706 				vm_page_clear_dirty(m, offs, len);
707 				m->valid = VM_PAGE_BITS_ALL;
708 			} else
709 				vm_pager_page_unswapped(m);
710 		}
711 		sf_buf_free(sf);
712 		sched_unpin();
713 		vm_page_wakeup(m);
714 		vm_page_lock(m);
715 		if (bp->bio_cmd == BIO_DELETE && len == PAGE_SIZE)
716 			vm_page_free(m);
717 		else
718 			vm_page_activate(m);
719 		vm_page_unlock(m);
720 		if (bp->bio_cmd == BIO_WRITE)
721 			vm_page_dirty(m);
722 
723 		/* Actions on further pages start at offset 0 */
724 		p += PAGE_SIZE - offs;
725 		offs = 0;
726 #if 0
727 if (bootverbose || bp->bio_offset / PAGE_SIZE < 17)
728 printf("wire_count %d busy %d flags %x hold_count %d act_count %d queue %d valid %d dirty %d @ %d\n",
729     m->wire_count, m->busy,
730     m->flags, m->hold_count, m->act_count, m->queue, m->valid, m->dirty, i);
731 #endif
732 	}
733 	vm_object_pip_subtract(sc->object, 1);
734 	VM_OBJECT_UNLOCK(sc->object);
735 	return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
736 }
737 
738 static void
739 md_kthread(void *arg)
740 {
741 	struct md_s *sc;
742 	struct bio *bp;
743 	int error;
744 
745 	sc = arg;
746 	thread_lock(curthread);
747 	sched_prio(curthread, PRIBIO);
748 	thread_unlock(curthread);
749 	if (sc->type == MD_VNODE)
750 		curthread->td_pflags |= TDP_NORUNNINGBUF;
751 
752 	for (;;) {
753 		mtx_lock(&sc->queue_mtx);
754 		if (sc->flags & MD_SHUTDOWN) {
755 			sc->flags |= MD_EXITING;
756 			mtx_unlock(&sc->queue_mtx);
757 			kproc_exit(0);
758 		}
759 		bp = bioq_takefirst(&sc->bio_queue);
760 		if (!bp) {
761 			msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
762 			continue;
763 		}
764 		mtx_unlock(&sc->queue_mtx);
765 		if (bp->bio_cmd == BIO_GETATTR) {
766 			if ((sc->fwsectors && sc->fwheads &&
767 			    (g_handleattr_int(bp, "GEOM::fwsectors",
768 			    sc->fwsectors) ||
769 			    g_handleattr_int(bp, "GEOM::fwheads",
770 			    sc->fwheads))) ||
771 			    g_handleattr_int(bp, "GEOM::candelete", 1))
772 				error = -1;
773 			else
774 				error = EOPNOTSUPP;
775 		} else {
776 			error = sc->start(sc, bp);
777 		}
778 
779 		if (error != -1) {
780 			bp->bio_completed = bp->bio_length;
781 			if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
782 				devstat_end_transaction_bio(sc->devstat, bp);
783 			g_io_deliver(bp, error);
784 		}
785 	}
786 }
787 
788 static struct md_s *
789 mdfind(int unit)
790 {
791 	struct md_s *sc;
792 
793 	LIST_FOREACH(sc, &md_softc_list, list) {
794 		if (sc->unit == unit)
795 			break;
796 	}
797 	return (sc);
798 }
799 
800 static struct md_s *
801 mdnew(int unit, int *errp, enum md_types type)
802 {
803 	struct md_s *sc;
804 	int error;
805 
806 	*errp = 0;
807 	if (unit == -1)
808 		unit = alloc_unr(md_uh);
809 	else
810 		unit = alloc_unr_specific(md_uh, unit);
811 
812 	if (unit == -1) {
813 		*errp = EBUSY;
814 		return (NULL);
815 	}
816 
817 	sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
818 	sc->type = type;
819 	bioq_init(&sc->bio_queue);
820 	mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
821 	sc->unit = unit;
822 	sprintf(sc->name, "md%d", unit);
823 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
824 	error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
825 	if (error == 0)
826 		return (sc);
827 	LIST_REMOVE(sc, list);
828 	mtx_destroy(&sc->queue_mtx);
829 	free_unr(md_uh, sc->unit);
830 	free(sc, M_MD);
831 	*errp = error;
832 	return (NULL);
833 }
834 
835 static void
836 mdinit(struct md_s *sc)
837 {
838 	struct g_geom *gp;
839 	struct g_provider *pp;
840 
841 	g_topology_lock();
842 	gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
843 	gp->softc = sc;
844 	pp = g_new_providerf(gp, "md%d", sc->unit);
845 	pp->mediasize = sc->mediasize;
846 	pp->sectorsize = sc->sectorsize;
847 	sc->gp = gp;
848 	sc->pp = pp;
849 	g_error_provider(pp, 0);
850 	g_topology_unlock();
851 	sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
852 	    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
853 }
854 
855 /*
856  * XXX: we should check that the range they feed us is mapped.
857  * XXX: we should implement read-only.
858  */
859 
860 static int
861 mdcreate_preload(struct md_s *sc, struct md_ioctl *mdio)
862 {
863 
864 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_FORCE))
865 		return (EINVAL);
866 	if (mdio->md_base == 0)
867 		return (EINVAL);
868 	sc->flags = mdio->md_options & MD_FORCE;
869 	/* Cast to pointer size, then to pointer to avoid warning */
870 	sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base;
871 	sc->pl_len = (size_t)sc->mediasize;
872 	return (0);
873 }
874 
875 
876 static int
877 mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio)
878 {
879 	uintptr_t sp;
880 	int error;
881 	off_t u;
882 
883 	error = 0;
884 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
885 		return (EINVAL);
886 	if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize))
887 		return (EINVAL);
888 	/* Compression doesn't make sense if we have reserved space */
889 	if (mdio->md_options & MD_RESERVE)
890 		mdio->md_options &= ~MD_COMPRESS;
891 	if (mdio->md_fwsectors != 0)
892 		sc->fwsectors = mdio->md_fwsectors;
893 	if (mdio->md_fwheads != 0)
894 		sc->fwheads = mdio->md_fwheads;
895 	sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE);
896 	sc->indir = dimension(sc->mediasize / sc->sectorsize);
897 	sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
898 	    0x1ff, 0);
899 	if (mdio->md_options & MD_RESERVE) {
900 		off_t nsectors;
901 
902 		nsectors = sc->mediasize / sc->sectorsize;
903 		for (u = 0; u < nsectors; u++) {
904 			sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
905 			    M_WAITOK : M_NOWAIT) | M_ZERO);
906 			if (sp != 0)
907 				error = s_write(sc->indir, u, sp);
908 			else
909 				error = ENOMEM;
910 			if (error != 0)
911 				break;
912 		}
913 	}
914 	return (error);
915 }
916 
917 
918 static int
919 mdsetcred(struct md_s *sc, struct ucred *cred)
920 {
921 	char *tmpbuf;
922 	int error = 0;
923 
924 	/*
925 	 * Set credits in our softc
926 	 */
927 
928 	if (sc->cred)
929 		crfree(sc->cred);
930 	sc->cred = crhold(cred);
931 
932 	/*
933 	 * Horrible kludge to establish credentials for NFS  XXX.
934 	 */
935 
936 	if (sc->vnode) {
937 		struct uio auio;
938 		struct iovec aiov;
939 
940 		tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
941 		bzero(&auio, sizeof(auio));
942 
943 		aiov.iov_base = tmpbuf;
944 		aiov.iov_len = sc->sectorsize;
945 		auio.uio_iov = &aiov;
946 		auio.uio_iovcnt = 1;
947 		auio.uio_offset = 0;
948 		auio.uio_rw = UIO_READ;
949 		auio.uio_segflg = UIO_SYSSPACE;
950 		auio.uio_resid = aiov.iov_len;
951 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
952 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
953 		VOP_UNLOCK(sc->vnode, 0);
954 		free(tmpbuf, M_TEMP);
955 	}
956 	return (error);
957 }
958 
959 static int
960 mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
961 {
962 	struct vattr vattr;
963 	struct nameidata nd;
964 	char *fname;
965 	int error, flags, vfslocked;
966 
967 	/*
968 	 * Kernel-originated requests must have the filename appended
969 	 * to the mdio structure to protect against malicious software.
970 	 */
971 	fname = mdio->md_file;
972 	if ((void *)fname != (void *)(mdio + 1)) {
973 		error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
974 		if (error != 0)
975 			return (error);
976 	} else
977 		strlcpy(sc->file, fname, sizeof(sc->file));
978 
979 	/*
980 	 * If the user specified that this is a read only device, don't
981 	 * set the FWRITE mask before trying to open the backing store.
982 	 */
983 	flags = FREAD | ((mdio->md_options & MD_READONLY) ? 0 : FWRITE);
984 	NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, sc->file, td);
985 	error = vn_open(&nd, &flags, 0, NULL);
986 	if (error != 0)
987 		return (error);
988 	vfslocked = NDHASGIANT(&nd);
989 	NDFREE(&nd, NDF_ONLY_PNBUF);
990 	if (nd.ni_vp->v_type != VREG) {
991 		error = EINVAL;
992 		goto bad;
993 	}
994 	error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
995 	if (error != 0)
996 		goto bad;
997 	if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
998 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
999 		if (nd.ni_vp->v_iflag & VI_DOOMED) {
1000 			/* Forced unmount. */
1001 			error = EBADF;
1002 			goto bad;
1003 		}
1004 	}
1005 	nd.ni_vp->v_vflag |= VV_MD;
1006 	VOP_UNLOCK(nd.ni_vp, 0);
1007 
1008 	if (mdio->md_fwsectors != 0)
1009 		sc->fwsectors = mdio->md_fwsectors;
1010 	if (mdio->md_fwheads != 0)
1011 		sc->fwheads = mdio->md_fwheads;
1012 	sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC);
1013 	if (!(flags & FWRITE))
1014 		sc->flags |= MD_READONLY;
1015 	sc->vnode = nd.ni_vp;
1016 
1017 	error = mdsetcred(sc, td->td_ucred);
1018 	if (error != 0) {
1019 		sc->vnode = NULL;
1020 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1021 		nd.ni_vp->v_vflag &= ~VV_MD;
1022 		goto bad;
1023 	}
1024 	VFS_UNLOCK_GIANT(vfslocked);
1025 	return (0);
1026 bad:
1027 	VOP_UNLOCK(nd.ni_vp, 0);
1028 	(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1029 	VFS_UNLOCK_GIANT(vfslocked);
1030 	return (error);
1031 }
1032 
1033 static int
1034 mddestroy(struct md_s *sc, struct thread *td)
1035 {
1036 	int vfslocked;
1037 
1038 	if (sc->gp) {
1039 		sc->gp->softc = NULL;
1040 		g_topology_lock();
1041 		g_wither_geom(sc->gp, ENXIO);
1042 		g_topology_unlock();
1043 		sc->gp = NULL;
1044 		sc->pp = NULL;
1045 	}
1046 	if (sc->devstat) {
1047 		devstat_remove_entry(sc->devstat);
1048 		sc->devstat = NULL;
1049 	}
1050 	mtx_lock(&sc->queue_mtx);
1051 	sc->flags |= MD_SHUTDOWN;
1052 	wakeup(sc);
1053 	while (!(sc->flags & MD_EXITING))
1054 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1055 	mtx_unlock(&sc->queue_mtx);
1056 	mtx_destroy(&sc->queue_mtx);
1057 	if (sc->vnode != NULL) {
1058 		vfslocked = VFS_LOCK_GIANT(sc->vnode->v_mount);
1059 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1060 		sc->vnode->v_vflag &= ~VV_MD;
1061 		VOP_UNLOCK(sc->vnode, 0);
1062 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1063 		    FREAD : (FREAD|FWRITE), sc->cred, td);
1064 		VFS_UNLOCK_GIANT(vfslocked);
1065 	}
1066 	if (sc->cred != NULL)
1067 		crfree(sc->cred);
1068 	if (sc->object != NULL)
1069 		vm_object_deallocate(sc->object);
1070 	if (sc->indir)
1071 		destroy_indir(sc, sc->indir);
1072 	if (sc->uma)
1073 		uma_zdestroy(sc->uma);
1074 
1075 	LIST_REMOVE(sc, list);
1076 	free_unr(md_uh, sc->unit);
1077 	free(sc, M_MD);
1078 	return (0);
1079 }
1080 
1081 static int
1082 mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
1083 {
1084 	vm_ooffset_t npage;
1085 	int error;
1086 
1087 	/*
1088 	 * Range check.  Disallow negative sizes or any size less then the
1089 	 * size of a page.  Then round to a page.
1090 	 */
1091 	if (sc->mediasize == 0 || (sc->mediasize % PAGE_SIZE) != 0)
1092 		return (EDOM);
1093 
1094 	/*
1095 	 * Allocate an OBJT_SWAP object.
1096 	 *
1097 	 * Note the truncation.
1098 	 */
1099 
1100 	npage = mdio->md_mediasize / PAGE_SIZE;
1101 	if (mdio->md_fwsectors != 0)
1102 		sc->fwsectors = mdio->md_fwsectors;
1103 	if (mdio->md_fwheads != 0)
1104 		sc->fwheads = mdio->md_fwheads;
1105 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1106 	    VM_PROT_DEFAULT, 0, td->td_ucred);
1107 	if (sc->object == NULL)
1108 		return (ENOMEM);
1109 	sc->flags = mdio->md_options & MD_FORCE;
1110 	if (mdio->md_options & MD_RESERVE) {
1111 		if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1112 			error = EDOM;
1113 			goto finish;
1114 		}
1115 	}
1116 	error = mdsetcred(sc, td->td_ucred);
1117  finish:
1118 	if (error != 0) {
1119 		vm_object_deallocate(sc->object);
1120 		sc->object = NULL;
1121 	}
1122 	return (error);
1123 }
1124 
1125 
1126 static int
1127 xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1128 {
1129 	struct md_ioctl *mdio;
1130 	struct md_s *sc;
1131 	int error, i;
1132 
1133 	if (md_debug)
1134 		printf("mdctlioctl(%s %lx %p %x %p)\n",
1135 			devtoname(dev), cmd, addr, flags, td);
1136 
1137 	mdio = (struct md_ioctl *)addr;
1138 	if (mdio->md_version != MDIOVERSION)
1139 		return (EINVAL);
1140 
1141 	/*
1142 	 * We assert the version number in the individual ioctl
1143 	 * handlers instead of out here because (a) it is possible we
1144 	 * may add another ioctl in the future which doesn't read an
1145 	 * mdio, and (b) the correct return value for an unknown ioctl
1146 	 * is ENOIOCTL, not EINVAL.
1147 	 */
1148 	error = 0;
1149 	switch (cmd) {
1150 	case MDIOCATTACH:
1151 		switch (mdio->md_type) {
1152 		case MD_MALLOC:
1153 		case MD_PRELOAD:
1154 		case MD_VNODE:
1155 		case MD_SWAP:
1156 			break;
1157 		default:
1158 			return (EINVAL);
1159 		}
1160 		if (mdio->md_options & MD_AUTOUNIT)
1161 			sc = mdnew(-1, &error, mdio->md_type);
1162 		else {
1163 			if (mdio->md_unit > INT_MAX)
1164 				return (EINVAL);
1165 			sc = mdnew(mdio->md_unit, &error, mdio->md_type);
1166 		}
1167 		if (sc == NULL)
1168 			return (error);
1169 		if (mdio->md_options & MD_AUTOUNIT)
1170 			mdio->md_unit = sc->unit;
1171 		sc->mediasize = mdio->md_mediasize;
1172 		if (mdio->md_sectorsize == 0)
1173 			sc->sectorsize = DEV_BSIZE;
1174 		else
1175 			sc->sectorsize = mdio->md_sectorsize;
1176 		error = EDOOFUS;
1177 		switch (sc->type) {
1178 		case MD_MALLOC:
1179 			sc->start = mdstart_malloc;
1180 			error = mdcreate_malloc(sc, mdio);
1181 			break;
1182 		case MD_PRELOAD:
1183 			sc->start = mdstart_preload;
1184 			error = mdcreate_preload(sc, mdio);
1185 			break;
1186 		case MD_VNODE:
1187 			sc->start = mdstart_vnode;
1188 			error = mdcreate_vnode(sc, mdio, td);
1189 			break;
1190 		case MD_SWAP:
1191 			sc->start = mdstart_swap;
1192 			error = mdcreate_swap(sc, mdio, td);
1193 			break;
1194 		}
1195 		if (error != 0) {
1196 			mddestroy(sc, td);
1197 			return (error);
1198 		}
1199 
1200 		/* Prune off any residual fractional sector */
1201 		i = sc->mediasize % sc->sectorsize;
1202 		sc->mediasize -= i;
1203 
1204 		mdinit(sc);
1205 		return (0);
1206 	case MDIOCDETACH:
1207 		if (mdio->md_mediasize != 0 ||
1208 		    (mdio->md_options & ~MD_FORCE) != 0)
1209 			return (EINVAL);
1210 
1211 		sc = mdfind(mdio->md_unit);
1212 		if (sc == NULL)
1213 			return (ENOENT);
1214 		if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1215 		    !(mdio->md_options & MD_FORCE))
1216 			return (EBUSY);
1217 		return (mddestroy(sc, td));
1218 	case MDIOCQUERY:
1219 		sc = mdfind(mdio->md_unit);
1220 		if (sc == NULL)
1221 			return (ENOENT);
1222 		mdio->md_type = sc->type;
1223 		mdio->md_options = sc->flags;
1224 		mdio->md_mediasize = sc->mediasize;
1225 		mdio->md_sectorsize = sc->sectorsize;
1226 		if (sc->type == MD_VNODE)
1227 			error = copyout(sc->file, mdio->md_file,
1228 			    strlen(sc->file) + 1);
1229 		return (error);
1230 	case MDIOCLIST:
1231 		i = 1;
1232 		LIST_FOREACH(sc, &md_softc_list, list) {
1233 			if (i == MDNPAD - 1)
1234 				mdio->md_pad[i] = -1;
1235 			else
1236 				mdio->md_pad[i++] = sc->unit;
1237 		}
1238 		mdio->md_pad[0] = i - 1;
1239 		return (0);
1240 	default:
1241 		return (ENOIOCTL);
1242 	};
1243 }
1244 
1245 static int
1246 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1247 {
1248 	int error;
1249 
1250 	sx_xlock(&md_sx);
1251 	error = xmdctlioctl(dev, cmd, addr, flags, td);
1252 	sx_xunlock(&md_sx);
1253 	return (error);
1254 }
1255 
1256 static void
1257 md_preloaded(u_char *image, size_t length)
1258 {
1259 	struct md_s *sc;
1260 	int error;
1261 
1262 	sc = mdnew(-1, &error, MD_PRELOAD);
1263 	if (sc == NULL)
1264 		return;
1265 	sc->mediasize = length;
1266 	sc->sectorsize = DEV_BSIZE;
1267 	sc->pl_ptr = image;
1268 	sc->pl_len = length;
1269 	sc->start = mdstart_preload;
1270 #ifdef MD_ROOT
1271 	if (sc->unit == 0)
1272 		rootdevnames[0] = "ufs:/dev/md0";
1273 #endif
1274 	mdinit(sc);
1275 }
1276 
1277 static void
1278 g_md_init(struct g_class *mp __unused)
1279 {
1280 	caddr_t mod;
1281 	u_char *ptr, *name, *type;
1282 	unsigned len;
1283 	int i;
1284 
1285 	/* figure out log2(NINDIR) */
1286 	for (i = NINDIR, nshift = -1; i; nshift++)
1287 		i >>= 1;
1288 
1289 	mod = NULL;
1290 	sx_init(&md_sx, "MD config lock");
1291 	g_topology_unlock();
1292 	md_uh = new_unrhdr(0, INT_MAX, NULL);
1293 #ifdef MD_ROOT_SIZE
1294 	sx_xlock(&md_sx);
1295 	md_preloaded(mfs_root.start, sizeof(mfs_root.start));
1296 	sx_xunlock(&md_sx);
1297 #endif
1298 	/* XXX: are preload_* static or do they need Giant ? */
1299 	while ((mod = preload_search_next_name(mod)) != NULL) {
1300 		name = (char *)preload_search_info(mod, MODINFO_NAME);
1301 		if (name == NULL)
1302 			continue;
1303 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
1304 		if (type == NULL)
1305 			continue;
1306 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
1307 			continue;
1308 		ptr = preload_fetch_addr(mod);
1309 		len = preload_fetch_size(mod);
1310 		if (ptr != NULL && len != 0) {
1311 			printf("%s%d: Preloaded image <%s> %d bytes at %p\n",
1312 			    MD_NAME, mdunits, name, len, ptr);
1313 			sx_xlock(&md_sx);
1314 			md_preloaded(ptr, len);
1315 			sx_xunlock(&md_sx);
1316 		}
1317 	}
1318 	status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
1319 	    0600, MDCTL_NAME);
1320 	g_topology_lock();
1321 }
1322 
1323 static void
1324 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1325     struct g_consumer *cp __unused, struct g_provider *pp)
1326 {
1327 	struct md_s *mp;
1328 	char *type;
1329 
1330 	mp = gp->softc;
1331 	if (mp == NULL)
1332 		return;
1333 
1334 	switch (mp->type) {
1335 	case MD_MALLOC:
1336 		type = "malloc";
1337 		break;
1338 	case MD_PRELOAD:
1339 		type = "preload";
1340 		break;
1341 	case MD_VNODE:
1342 		type = "vnode";
1343 		break;
1344 	case MD_SWAP:
1345 		type = "swap";
1346 		break;
1347 	default:
1348 		type = "unknown";
1349 		break;
1350 	}
1351 
1352 	if (pp != NULL) {
1353 		if (indent == NULL) {
1354 			sbuf_printf(sb, " u %d", mp->unit);
1355 			sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
1356 			sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
1357 			sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
1358 			sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
1359 			sbuf_printf(sb, " t %s", type);
1360 			if (mp->type == MD_VNODE && mp->vnode != NULL)
1361 				sbuf_printf(sb, " file %s", mp->file);
1362 		} else {
1363 			sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
1364 			    mp->unit);
1365 			sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
1366 			    indent, (uintmax_t) mp->sectorsize);
1367 			sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
1368 			    indent, (uintmax_t) mp->fwheads);
1369 			sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
1370 			    indent, (uintmax_t) mp->fwsectors);
1371 			sbuf_printf(sb, "%s<length>%ju</length>\n",
1372 			    indent, (uintmax_t) mp->mediasize);
1373 			sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
1374 			    (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
1375 			sbuf_printf(sb, "%s<access>%s</access>\n", indent,
1376 			    (mp->flags & MD_READONLY) == 0 ? "read-write":
1377 			    "read-only");
1378 			sbuf_printf(sb, "%s<type>%s</type>\n", indent,
1379 			    type);
1380 			if (mp->type == MD_VNODE && mp->vnode != NULL)
1381 				sbuf_printf(sb, "%s<file>%s</file>\n",
1382 				    indent, mp->file);
1383 		}
1384 	}
1385 }
1386 
1387 static void
1388 g_md_fini(struct g_class *mp __unused)
1389 {
1390 
1391 	sx_destroy(&md_sx);
1392 	if (status_dev != NULL)
1393 		destroy_dev(status_dev);
1394 	delete_unrhdr(md_uh);
1395 }
1396