xref: /freebsd/sys/dev/md/md.c (revision f1f890804985a1043da42a5def13c79dc005f5e9)
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/rwlock.h>
79 #include <sys/sbuf.h>
80 #include <sys/sched.h>
81 #include <sys/sf_buf.h>
82 #include <sys/sysctl.h>
83 #include <sys/vnode.h>
84 
85 #include <geom/geom.h>
86 
87 #include <vm/vm.h>
88 #include <vm/vm_param.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/swap_pager.h>
93 #include <vm/uma.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     "Enable md(4) debug messages");
110 static int md_malloc_wait;
111 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
112     "Allow malloc to wait for memory allocations");
113 
114 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
115 #define	MD_ROOT_FSTYPE	"ufs"
116 #endif
117 
118 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE)
119 /*
120  * Preloaded image gets put here.
121  * Applications that patch the object with the image can determine
122  * the size looking at the start and end markers (strings),
123  * so we want them contiguous.
124  */
125 static struct {
126 	u_char start[MD_ROOT_SIZE*1024];
127 	u_char end[128];
128 } mfs_root = {
129 	.start = "MFS Filesystem goes here",
130 	.end = "MFS Filesystem had better STOP here",
131 };
132 #endif
133 
134 static g_init_t g_md_init;
135 static g_fini_t g_md_fini;
136 static g_start_t g_md_start;
137 static g_access_t g_md_access;
138 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
139     struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
140 
141 static struct cdev *status_dev = 0;
142 static struct sx md_sx;
143 static struct unrhdr *md_uh;
144 
145 static d_ioctl_t mdctlioctl;
146 
147 static struct cdevsw mdctl_cdevsw = {
148 	.d_version =	D_VERSION,
149 	.d_ioctl =	mdctlioctl,
150 	.d_name =	MD_NAME,
151 };
152 
153 struct g_class g_md_class = {
154 	.name = "MD",
155 	.version = G_VERSION,
156 	.init = g_md_init,
157 	.fini = g_md_fini,
158 	.start = g_md_start,
159 	.access = g_md_access,
160 	.dumpconf = g_md_dumpconf,
161 };
162 
163 DECLARE_GEOM_CLASS(g_md_class, g_md);
164 
165 
166 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
167 
168 #define NINDIR	(PAGE_SIZE / sizeof(uintptr_t))
169 #define NMASK	(NINDIR-1)
170 static int nshift;
171 
172 struct indir {
173 	uintptr_t	*array;
174 	u_int		total;
175 	u_int		used;
176 	u_int		shift;
177 };
178 
179 struct md_s {
180 	int unit;
181 	LIST_ENTRY(md_s) list;
182 	struct bio_queue_head bio_queue;
183 	struct mtx queue_mtx;
184 	struct cdev *dev;
185 	enum md_types type;
186 	off_t mediasize;
187 	unsigned sectorsize;
188 	unsigned opencount;
189 	unsigned fwheads;
190 	unsigned fwsectors;
191 	unsigned flags;
192 	char name[20];
193 	struct proc *procp;
194 	struct g_geom *gp;
195 	struct g_provider *pp;
196 	int (*start)(struct md_s *sc, struct bio *bp);
197 	struct devstat *devstat;
198 
199 	/* MD_MALLOC related fields */
200 	struct indir *indir;
201 	uma_zone_t uma;
202 
203 	/* MD_PRELOAD related fields */
204 	u_char *pl_ptr;
205 	size_t pl_len;
206 
207 	/* MD_VNODE related fields */
208 	struct vnode *vnode;
209 	char file[PATH_MAX];
210 	struct ucred *cred;
211 
212 	/* MD_SWAP related fields */
213 	vm_object_t object;
214 };
215 
216 static struct indir *
217 new_indir(u_int shift)
218 {
219 	struct indir *ip;
220 
221 	ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
222 	    | M_ZERO);
223 	if (ip == NULL)
224 		return (NULL);
225 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
226 	    M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
227 	if (ip->array == NULL) {
228 		free(ip, M_MD);
229 		return (NULL);
230 	}
231 	ip->total = NINDIR;
232 	ip->shift = shift;
233 	return (ip);
234 }
235 
236 static void
237 del_indir(struct indir *ip)
238 {
239 
240 	free(ip->array, M_MDSECT);
241 	free(ip, M_MD);
242 }
243 
244 static void
245 destroy_indir(struct md_s *sc, struct indir *ip)
246 {
247 	int i;
248 
249 	for (i = 0; i < NINDIR; i++) {
250 		if (!ip->array[i])
251 			continue;
252 		if (ip->shift)
253 			destroy_indir(sc, (struct indir*)(ip->array[i]));
254 		else if (ip->array[i] > 255)
255 			uma_zfree(sc->uma, (void *)(ip->array[i]));
256 	}
257 	del_indir(ip);
258 }
259 
260 /*
261  * This function does the math and allocates the top level "indir" structure
262  * for a device of "size" sectors.
263  */
264 
265 static struct indir *
266 dimension(off_t size)
267 {
268 	off_t rcnt;
269 	struct indir *ip;
270 	int layer;
271 
272 	rcnt = size;
273 	layer = 0;
274 	while (rcnt > NINDIR) {
275 		rcnt /= NINDIR;
276 		layer++;
277 	}
278 
279 	/*
280 	 * XXX: the top layer is probably not fully populated, so we allocate
281 	 * too much space for ip->array in here.
282 	 */
283 	ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
284 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
285 	    M_MDSECT, M_WAITOK | M_ZERO);
286 	ip->total = NINDIR;
287 	ip->shift = layer * nshift;
288 	return (ip);
289 }
290 
291 /*
292  * Read a given sector
293  */
294 
295 static uintptr_t
296 s_read(struct indir *ip, off_t offset)
297 {
298 	struct indir *cip;
299 	int idx;
300 	uintptr_t up;
301 
302 	if (md_debug > 1)
303 		printf("s_read(%jd)\n", (intmax_t)offset);
304 	up = 0;
305 	for (cip = ip; cip != NULL;) {
306 		if (cip->shift) {
307 			idx = (offset >> cip->shift) & NMASK;
308 			up = cip->array[idx];
309 			cip = (struct indir *)up;
310 			continue;
311 		}
312 		idx = offset & NMASK;
313 		return (cip->array[idx]);
314 	}
315 	return (0);
316 }
317 
318 /*
319  * Write a given sector, prune the tree if the value is 0
320  */
321 
322 static int
323 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
324 {
325 	struct indir *cip, *lip[10];
326 	int idx, li;
327 	uintptr_t up;
328 
329 	if (md_debug > 1)
330 		printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
331 	up = 0;
332 	li = 0;
333 	cip = ip;
334 	for (;;) {
335 		lip[li++] = cip;
336 		if (cip->shift) {
337 			idx = (offset >> cip->shift) & NMASK;
338 			up = cip->array[idx];
339 			if (up != 0) {
340 				cip = (struct indir *)up;
341 				continue;
342 			}
343 			/* Allocate branch */
344 			cip->array[idx] =
345 			    (uintptr_t)new_indir(cip->shift - nshift);
346 			if (cip->array[idx] == 0)
347 				return (ENOSPC);
348 			cip->used++;
349 			up = cip->array[idx];
350 			cip = (struct indir *)up;
351 			continue;
352 		}
353 		/* leafnode */
354 		idx = offset & NMASK;
355 		up = cip->array[idx];
356 		if (up != 0)
357 			cip->used--;
358 		cip->array[idx] = ptr;
359 		if (ptr != 0)
360 			cip->used++;
361 		break;
362 	}
363 	if (cip->used != 0 || li == 1)
364 		return (0);
365 	li--;
366 	while (cip->used == 0 && cip != ip) {
367 		li--;
368 		idx = (offset >> lip[li]->shift) & NMASK;
369 		up = lip[li]->array[idx];
370 		KASSERT(up == (uintptr_t)cip, ("md screwed up"));
371 		del_indir(cip);
372 		lip[li]->array[idx] = 0;
373 		lip[li]->used--;
374 		cip = lip[li];
375 	}
376 	return (0);
377 }
378 
379 
380 static int
381 g_md_access(struct g_provider *pp, int r, int w, int e)
382 {
383 	struct md_s *sc;
384 
385 	sc = pp->geom->softc;
386 	if (sc == NULL) {
387 		if (r <= 0 && w <= 0 && e <= 0)
388 			return (0);
389 		return (ENXIO);
390 	}
391 	r += pp->acr;
392 	w += pp->acw;
393 	e += pp->ace;
394 	if ((sc->flags & MD_READONLY) != 0 && w > 0)
395 		return (EROFS);
396 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
397 		sc->opencount = 1;
398 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
399 		sc->opencount = 0;
400 	}
401 	return (0);
402 }
403 
404 static void
405 g_md_start(struct bio *bp)
406 {
407 	struct md_s *sc;
408 
409 	sc = bp->bio_to->geom->softc;
410 	if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
411 		devstat_start_transaction_bio(sc->devstat, bp);
412 	mtx_lock(&sc->queue_mtx);
413 	bioq_disksort(&sc->bio_queue, bp);
414 	mtx_unlock(&sc->queue_mtx);
415 	wakeup(sc);
416 }
417 
418 static int
419 mdstart_malloc(struct md_s *sc, struct bio *bp)
420 {
421 	int i, error;
422 	u_char *dst;
423 	off_t secno, nsec, uc;
424 	uintptr_t sp, osp;
425 
426 	switch (bp->bio_cmd) {
427 	case BIO_READ:
428 	case BIO_WRITE:
429 	case BIO_DELETE:
430 		break;
431 	default:
432 		return (EOPNOTSUPP);
433 	}
434 
435 	nsec = bp->bio_length / sc->sectorsize;
436 	secno = bp->bio_offset / sc->sectorsize;
437 	dst = bp->bio_data;
438 	error = 0;
439 	while (nsec--) {
440 		osp = s_read(sc->indir, secno);
441 		if (bp->bio_cmd == BIO_DELETE) {
442 			if (osp != 0)
443 				error = s_write(sc->indir, secno, 0);
444 		} else if (bp->bio_cmd == BIO_READ) {
445 			if (osp == 0)
446 				bzero(dst, sc->sectorsize);
447 			else if (osp <= 255)
448 				memset(dst, osp, sc->sectorsize);
449 			else {
450 				bcopy((void *)osp, dst, sc->sectorsize);
451 				cpu_flush_dcache(dst, sc->sectorsize);
452 			}
453 			osp = 0;
454 		} else if (bp->bio_cmd == BIO_WRITE) {
455 			if (sc->flags & MD_COMPRESS) {
456 				uc = dst[0];
457 				for (i = 1; i < sc->sectorsize; i++)
458 					if (dst[i] != uc)
459 						break;
460 			} else {
461 				i = 0;
462 				uc = 0;
463 			}
464 			if (i == sc->sectorsize) {
465 				if (osp != uc)
466 					error = s_write(sc->indir, secno, uc);
467 			} else {
468 				if (osp <= 255) {
469 					sp = (uintptr_t)uma_zalloc(sc->uma,
470 					    md_malloc_wait ? M_WAITOK :
471 					    M_NOWAIT);
472 					if (sp == 0) {
473 						error = ENOSPC;
474 						break;
475 					}
476 					bcopy(dst, (void *)sp, sc->sectorsize);
477 					error = s_write(sc->indir, secno, sp);
478 				} else {
479 					bcopy(dst, (void *)osp, sc->sectorsize);
480 					osp = 0;
481 				}
482 			}
483 		} else {
484 			error = EOPNOTSUPP;
485 		}
486 		if (osp > 255)
487 			uma_zfree(sc->uma, (void*)osp);
488 		if (error != 0)
489 			break;
490 		secno++;
491 		dst += sc->sectorsize;
492 	}
493 	bp->bio_resid = 0;
494 	return (error);
495 }
496 
497 static int
498 mdstart_preload(struct md_s *sc, struct bio *bp)
499 {
500 
501 	switch (bp->bio_cmd) {
502 	case BIO_READ:
503 		bcopy(sc->pl_ptr + bp->bio_offset, bp->bio_data,
504 		    bp->bio_length);
505 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
506 		break;
507 	case BIO_WRITE:
508 		bcopy(bp->bio_data, sc->pl_ptr + bp->bio_offset,
509 		    bp->bio_length);
510 		break;
511 	}
512 	bp->bio_resid = 0;
513 	return (0);
514 }
515 
516 static int
517 mdstart_vnode(struct md_s *sc, struct bio *bp)
518 {
519 	int error;
520 	struct uio auio;
521 	struct iovec aiov;
522 	struct mount *mp;
523 	struct vnode *vp;
524 	struct thread *td;
525 	off_t end, zerosize;
526 
527 	switch (bp->bio_cmd) {
528 	case BIO_READ:
529 	case BIO_WRITE:
530 	case BIO_DELETE:
531 	case BIO_FLUSH:
532 		break;
533 	default:
534 		return (EOPNOTSUPP);
535 	}
536 
537 	td = curthread;
538 	vp = sc->vnode;
539 
540 	/*
541 	 * VNODE I/O
542 	 *
543 	 * If an error occurs, we set BIO_ERROR but we do not set
544 	 * B_INVAL because (for a write anyway), the buffer is
545 	 * still valid.
546 	 */
547 
548 	if (bp->bio_cmd == BIO_FLUSH) {
549 		(void) vn_start_write(vp, &mp, V_WAIT);
550 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
551 		error = VOP_FSYNC(vp, MNT_WAIT, td);
552 		VOP_UNLOCK(vp, 0);
553 		vn_finished_write(mp);
554 		return (error);
555 	}
556 
557 	bzero(&auio, sizeof(auio));
558 
559 	/*
560 	 * Special case for BIO_DELETE.  On the surface, this is very
561 	 * similar to BIO_WRITE, except that we write from our own
562 	 * fixed-length buffer, so we have to loop.  The net result is
563 	 * that the two cases end up having very little in common.
564 	 */
565 	if (bp->bio_cmd == BIO_DELETE) {
566 		zerosize = ZERO_REGION_SIZE -
567 		    (ZERO_REGION_SIZE % sc->sectorsize);
568 		auio.uio_iov = &aiov;
569 		auio.uio_iovcnt = 1;
570 		auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
571 		auio.uio_segflg = UIO_SYSSPACE;
572 		auio.uio_rw = UIO_WRITE;
573 		auio.uio_td = td;
574 		end = bp->bio_offset + bp->bio_length;
575 		(void) vn_start_write(vp, &mp, V_WAIT);
576 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
577 		error = 0;
578 		while (auio.uio_offset < end) {
579 			aiov.iov_base = __DECONST(void *, zero_region);
580 			aiov.iov_len = end - auio.uio_offset;
581 			if (aiov.iov_len > zerosize)
582 				aiov.iov_len = zerosize;
583 			auio.uio_resid = aiov.iov_len;
584 			error = VOP_WRITE(vp, &auio,
585 			    sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred);
586 			if (error != 0)
587 				break;
588 		}
589 		VOP_UNLOCK(vp, 0);
590 		vn_finished_write(mp);
591 		bp->bio_resid = end - auio.uio_offset;
592 		return (error);
593 	}
594 
595 	aiov.iov_base = bp->bio_data;
596 	aiov.iov_len = bp->bio_length;
597 	auio.uio_iov = &aiov;
598 	auio.uio_iovcnt = 1;
599 	auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
600 	auio.uio_segflg = UIO_SYSSPACE;
601 	if (bp->bio_cmd == BIO_READ)
602 		auio.uio_rw = UIO_READ;
603 	else if (bp->bio_cmd == BIO_WRITE)
604 		auio.uio_rw = UIO_WRITE;
605 	else
606 		panic("wrong BIO_OP in mdstart_vnode");
607 	auio.uio_resid = bp->bio_length;
608 	auio.uio_td = td;
609 	/*
610 	 * When reading set IO_DIRECT to try to avoid double-caching
611 	 * the data.  When writing IO_DIRECT is not optimal.
612 	 */
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 	bp->bio_resid = auio.uio_resid;
626 	return (error);
627 }
628 
629 static int
630 mdstart_swap(struct md_s *sc, struct bio *bp)
631 {
632 	struct sf_buf *sf;
633 	int rv, offs, len, lastend;
634 	vm_pindex_t i, lastp;
635 	vm_page_t m;
636 	u_char *p;
637 
638 	switch (bp->bio_cmd) {
639 	case BIO_READ:
640 	case BIO_WRITE:
641 	case BIO_DELETE:
642 		break;
643 	default:
644 		return (EOPNOTSUPP);
645 	}
646 
647 	p = bp->bio_data;
648 
649 	/*
650 	 * offs is the offset at which to start operating on the
651 	 * next (ie, first) page.  lastp is the last page on
652 	 * which we're going to operate.  lastend is the ending
653 	 * position within that last page (ie, PAGE_SIZE if
654 	 * we're operating on complete aligned pages).
655 	 */
656 	offs = bp->bio_offset % PAGE_SIZE;
657 	lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
658 	lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
659 
660 	rv = VM_PAGER_OK;
661 	VM_OBJECT_WLOCK(sc->object);
662 	vm_object_pip_add(sc->object, 1);
663 	for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
664 		len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
665 
666 		m = vm_page_grab(sc->object, i,
667 		    VM_ALLOC_NORMAL|VM_ALLOC_RETRY);
668 		VM_OBJECT_WUNLOCK(sc->object);
669 		sched_pin();
670 		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
671 		VM_OBJECT_WLOCK(sc->object);
672 		if (bp->bio_cmd == BIO_READ) {
673 			if (m->valid != VM_PAGE_BITS_ALL)
674 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
675 			if (rv == VM_PAGER_ERROR) {
676 				sf_buf_free(sf);
677 				sched_unpin();
678 				vm_page_wakeup(m);
679 				break;
680 			} else if (rv == VM_PAGER_FAIL) {
681 				/*
682 				 * Pager does not have the page.  Zero
683 				 * the allocated page, and mark it as
684 				 * valid. Do not set dirty, the page
685 				 * can be recreated if thrown out.
686 				 */
687 				bzero((void *)sf_buf_kva(sf), PAGE_SIZE);
688 				m->valid = VM_PAGE_BITS_ALL;
689 			}
690 			bcopy((void *)(sf_buf_kva(sf) + offs), p, len);
691 			cpu_flush_dcache(p, len);
692 		} else if (bp->bio_cmd == BIO_WRITE) {
693 			if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
694 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
695 			if (rv == VM_PAGER_ERROR) {
696 				sf_buf_free(sf);
697 				sched_unpin();
698 				vm_page_wakeup(m);
699 				break;
700 			}
701 			bcopy(p, (void *)(sf_buf_kva(sf) + offs), len);
702 			m->valid = VM_PAGE_BITS_ALL;
703 		} else if (bp->bio_cmd == BIO_DELETE) {
704 			if (len != PAGE_SIZE && m->valid != VM_PAGE_BITS_ALL)
705 				rv = vm_pager_get_pages(sc->object, &m, 1, 0);
706 			if (rv == VM_PAGER_ERROR) {
707 				sf_buf_free(sf);
708 				sched_unpin();
709 				vm_page_wakeup(m);
710 				break;
711 			}
712 			if (len != PAGE_SIZE) {
713 				bzero((void *)(sf_buf_kva(sf) + offs), len);
714 				vm_page_clear_dirty(m, offs, len);
715 				m->valid = VM_PAGE_BITS_ALL;
716 			} else
717 				vm_pager_page_unswapped(m);
718 		}
719 		sf_buf_free(sf);
720 		sched_unpin();
721 		vm_page_wakeup(m);
722 		vm_page_lock(m);
723 		if (bp->bio_cmd == BIO_DELETE && len == PAGE_SIZE)
724 			vm_page_free(m);
725 		else
726 			vm_page_activate(m);
727 		vm_page_unlock(m);
728 		if (bp->bio_cmd == BIO_WRITE)
729 			vm_page_dirty(m);
730 
731 		/* Actions on further pages start at offset 0 */
732 		p += PAGE_SIZE - offs;
733 		offs = 0;
734 	}
735 	vm_object_pip_subtract(sc->object, 1);
736 	VM_OBJECT_WUNLOCK(sc->object);
737 	return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
738 }
739 
740 static void
741 md_kthread(void *arg)
742 {
743 	struct md_s *sc;
744 	struct bio *bp;
745 	int error;
746 
747 	sc = arg;
748 	thread_lock(curthread);
749 	sched_prio(curthread, PRIBIO);
750 	thread_unlock(curthread);
751 	if (sc->type == MD_VNODE)
752 		curthread->td_pflags |= TDP_NORUNNINGBUF;
753 
754 	for (;;) {
755 		mtx_lock(&sc->queue_mtx);
756 		if (sc->flags & MD_SHUTDOWN) {
757 			sc->flags |= MD_EXITING;
758 			mtx_unlock(&sc->queue_mtx);
759 			kproc_exit(0);
760 		}
761 		bp = bioq_takefirst(&sc->bio_queue);
762 		if (!bp) {
763 			msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
764 			continue;
765 		}
766 		mtx_unlock(&sc->queue_mtx);
767 		if (bp->bio_cmd == BIO_GETATTR) {
768 			if ((sc->fwsectors && sc->fwheads &&
769 			    (g_handleattr_int(bp, "GEOM::fwsectors",
770 			    sc->fwsectors) ||
771 			    g_handleattr_int(bp, "GEOM::fwheads",
772 			    sc->fwheads))) ||
773 			    g_handleattr_int(bp, "GEOM::candelete", 1))
774 				error = -1;
775 			else
776 				error = EOPNOTSUPP;
777 		} else {
778 			error = sc->start(sc, bp);
779 		}
780 
781 		if (error != -1) {
782 			bp->bio_completed = bp->bio_length;
783 			if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
784 				devstat_end_transaction_bio(sc->devstat, bp);
785 			g_io_deliver(bp, error);
786 		}
787 	}
788 }
789 
790 static struct md_s *
791 mdfind(int unit)
792 {
793 	struct md_s *sc;
794 
795 	LIST_FOREACH(sc, &md_softc_list, list) {
796 		if (sc->unit == unit)
797 			break;
798 	}
799 	return (sc);
800 }
801 
802 static struct md_s *
803 mdnew(int unit, int *errp, enum md_types type)
804 {
805 	struct md_s *sc;
806 	int error;
807 
808 	*errp = 0;
809 	if (unit == -1)
810 		unit = alloc_unr(md_uh);
811 	else
812 		unit = alloc_unr_specific(md_uh, unit);
813 
814 	if (unit == -1) {
815 		*errp = EBUSY;
816 		return (NULL);
817 	}
818 
819 	sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
820 	sc->type = type;
821 	bioq_init(&sc->bio_queue);
822 	mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
823 	sc->unit = unit;
824 	sprintf(sc->name, "md%d", unit);
825 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
826 	error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
827 	if (error == 0)
828 		return (sc);
829 	LIST_REMOVE(sc, list);
830 	mtx_destroy(&sc->queue_mtx);
831 	free_unr(md_uh, sc->unit);
832 	free(sc, M_MD);
833 	*errp = error;
834 	return (NULL);
835 }
836 
837 static void
838 mdinit(struct md_s *sc)
839 {
840 	struct g_geom *gp;
841 	struct g_provider *pp;
842 
843 	g_topology_lock();
844 	gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
845 	gp->softc = sc;
846 	pp = g_new_providerf(gp, "md%d", sc->unit);
847 	pp->mediasize = sc->mediasize;
848 	pp->sectorsize = sc->sectorsize;
849 	sc->gp = gp;
850 	sc->pp = pp;
851 	g_error_provider(pp, 0);
852 	g_topology_unlock();
853 	sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
854 	    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
855 }
856 
857 static int
858 mdcreate_malloc(struct md_s *sc, struct md_ioctl *mdio)
859 {
860 	uintptr_t sp;
861 	int error;
862 	off_t u;
863 
864 	error = 0;
865 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
866 		return (EINVAL);
867 	if (mdio->md_sectorsize != 0 && !powerof2(mdio->md_sectorsize))
868 		return (EINVAL);
869 	/* Compression doesn't make sense if we have reserved space */
870 	if (mdio->md_options & MD_RESERVE)
871 		mdio->md_options &= ~MD_COMPRESS;
872 	if (mdio->md_fwsectors != 0)
873 		sc->fwsectors = mdio->md_fwsectors;
874 	if (mdio->md_fwheads != 0)
875 		sc->fwheads = mdio->md_fwheads;
876 	sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE);
877 	sc->indir = dimension(sc->mediasize / sc->sectorsize);
878 	sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
879 	    0x1ff, 0);
880 	if (mdio->md_options & MD_RESERVE) {
881 		off_t nsectors;
882 
883 		nsectors = sc->mediasize / sc->sectorsize;
884 		for (u = 0; u < nsectors; u++) {
885 			sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
886 			    M_WAITOK : M_NOWAIT) | M_ZERO);
887 			if (sp != 0)
888 				error = s_write(sc->indir, u, sp);
889 			else
890 				error = ENOMEM;
891 			if (error != 0)
892 				break;
893 		}
894 	}
895 	return (error);
896 }
897 
898 
899 static int
900 mdsetcred(struct md_s *sc, struct ucred *cred)
901 {
902 	char *tmpbuf;
903 	int error = 0;
904 
905 	/*
906 	 * Set credits in our softc
907 	 */
908 
909 	if (sc->cred)
910 		crfree(sc->cred);
911 	sc->cred = crhold(cred);
912 
913 	/*
914 	 * Horrible kludge to establish credentials for NFS  XXX.
915 	 */
916 
917 	if (sc->vnode) {
918 		struct uio auio;
919 		struct iovec aiov;
920 
921 		tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
922 		bzero(&auio, sizeof(auio));
923 
924 		aiov.iov_base = tmpbuf;
925 		aiov.iov_len = sc->sectorsize;
926 		auio.uio_iov = &aiov;
927 		auio.uio_iovcnt = 1;
928 		auio.uio_offset = 0;
929 		auio.uio_rw = UIO_READ;
930 		auio.uio_segflg = UIO_SYSSPACE;
931 		auio.uio_resid = aiov.iov_len;
932 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
933 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
934 		VOP_UNLOCK(sc->vnode, 0);
935 		free(tmpbuf, M_TEMP);
936 	}
937 	return (error);
938 }
939 
940 static int
941 mdcreate_vnode(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
942 {
943 	struct vattr vattr;
944 	struct nameidata nd;
945 	char *fname;
946 	int error, flags;
947 
948 	/*
949 	 * Kernel-originated requests must have the filename appended
950 	 * to the mdio structure to protect against malicious software.
951 	 */
952 	fname = mdio->md_file;
953 	if ((void *)fname != (void *)(mdio + 1)) {
954 		error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
955 		if (error != 0)
956 			return (error);
957 	} else
958 		strlcpy(sc->file, fname, sizeof(sc->file));
959 
960 	/*
961 	 * If the user specified that this is a read only device, don't
962 	 * set the FWRITE mask before trying to open the backing store.
963 	 */
964 	flags = FREAD | ((mdio->md_options & MD_READONLY) ? 0 : FWRITE);
965 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
966 	error = vn_open(&nd, &flags, 0, NULL);
967 	if (error != 0)
968 		return (error);
969 	NDFREE(&nd, NDF_ONLY_PNBUF);
970 	if (nd.ni_vp->v_type != VREG) {
971 		error = EINVAL;
972 		goto bad;
973 	}
974 	error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
975 	if (error != 0)
976 		goto bad;
977 	if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
978 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
979 		if (nd.ni_vp->v_iflag & VI_DOOMED) {
980 			/* Forced unmount. */
981 			error = EBADF;
982 			goto bad;
983 		}
984 	}
985 	nd.ni_vp->v_vflag |= VV_MD;
986 	VOP_UNLOCK(nd.ni_vp, 0);
987 
988 	if (mdio->md_fwsectors != 0)
989 		sc->fwsectors = mdio->md_fwsectors;
990 	if (mdio->md_fwheads != 0)
991 		sc->fwheads = mdio->md_fwheads;
992 	sc->flags = mdio->md_options & (MD_FORCE | MD_ASYNC);
993 	if (!(flags & FWRITE))
994 		sc->flags |= MD_READONLY;
995 	sc->vnode = nd.ni_vp;
996 
997 	error = mdsetcred(sc, td->td_ucred);
998 	if (error != 0) {
999 		sc->vnode = NULL;
1000 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1001 		nd.ni_vp->v_vflag &= ~VV_MD;
1002 		goto bad;
1003 	}
1004 	return (0);
1005 bad:
1006 	VOP_UNLOCK(nd.ni_vp, 0);
1007 	(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1008 	return (error);
1009 }
1010 
1011 static int
1012 mddestroy(struct md_s *sc, struct thread *td)
1013 {
1014 
1015 	if (sc->gp) {
1016 		sc->gp->softc = NULL;
1017 		g_topology_lock();
1018 		g_wither_geom(sc->gp, ENXIO);
1019 		g_topology_unlock();
1020 		sc->gp = NULL;
1021 		sc->pp = NULL;
1022 	}
1023 	if (sc->devstat) {
1024 		devstat_remove_entry(sc->devstat);
1025 		sc->devstat = NULL;
1026 	}
1027 	mtx_lock(&sc->queue_mtx);
1028 	sc->flags |= MD_SHUTDOWN;
1029 	wakeup(sc);
1030 	while (!(sc->flags & MD_EXITING))
1031 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1032 	mtx_unlock(&sc->queue_mtx);
1033 	mtx_destroy(&sc->queue_mtx);
1034 	if (sc->vnode != NULL) {
1035 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1036 		sc->vnode->v_vflag &= ~VV_MD;
1037 		VOP_UNLOCK(sc->vnode, 0);
1038 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1039 		    FREAD : (FREAD|FWRITE), sc->cred, td);
1040 	}
1041 	if (sc->cred != NULL)
1042 		crfree(sc->cred);
1043 	if (sc->object != NULL)
1044 		vm_object_deallocate(sc->object);
1045 	if (sc->indir)
1046 		destroy_indir(sc, sc->indir);
1047 	if (sc->uma)
1048 		uma_zdestroy(sc->uma);
1049 
1050 	LIST_REMOVE(sc, list);
1051 	free_unr(md_uh, sc->unit);
1052 	free(sc, M_MD);
1053 	return (0);
1054 }
1055 
1056 static int
1057 mdresize(struct md_s *sc, struct md_ioctl *mdio)
1058 {
1059 	int error, res;
1060 	vm_pindex_t oldpages, newpages;
1061 
1062 	switch (sc->type) {
1063 	case MD_VNODE:
1064 		break;
1065 	case MD_SWAP:
1066 		if (mdio->md_mediasize <= 0 ||
1067 		    (mdio->md_mediasize % PAGE_SIZE) != 0)
1068 			return (EDOM);
1069 		oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1070 		newpages = OFF_TO_IDX(round_page(mdio->md_mediasize));
1071 		if (newpages < oldpages) {
1072 			VM_OBJECT_WLOCK(sc->object);
1073 			vm_object_page_remove(sc->object, newpages, 0, 0);
1074 			swap_pager_freespace(sc->object, newpages,
1075 			    oldpages - newpages);
1076 			swap_release_by_cred(IDX_TO_OFF(oldpages -
1077 			    newpages), sc->cred);
1078 			sc->object->charge = IDX_TO_OFF(newpages);
1079 			sc->object->size = newpages;
1080 			VM_OBJECT_WUNLOCK(sc->object);
1081 		} else if (newpages > oldpages) {
1082 			res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1083 			    oldpages), sc->cred);
1084 			if (!res)
1085 				return (ENOMEM);
1086 			if ((mdio->md_options & MD_RESERVE) ||
1087 			    (sc->flags & MD_RESERVE)) {
1088 				error = swap_pager_reserve(sc->object,
1089 				    oldpages, newpages - oldpages);
1090 				if (error < 0) {
1091 					swap_release_by_cred(
1092 					    IDX_TO_OFF(newpages - oldpages),
1093 					    sc->cred);
1094 					return (EDOM);
1095 				}
1096 			}
1097 			VM_OBJECT_WLOCK(sc->object);
1098 			sc->object->charge = IDX_TO_OFF(newpages);
1099 			sc->object->size = newpages;
1100 			VM_OBJECT_WUNLOCK(sc->object);
1101 		}
1102 		break;
1103 	default:
1104 		return (EOPNOTSUPP);
1105 	}
1106 
1107 	sc->mediasize = mdio->md_mediasize;
1108 	g_topology_lock();
1109 	g_resize_provider(sc->pp, sc->mediasize);
1110 	g_topology_unlock();
1111 	return (0);
1112 }
1113 
1114 static int
1115 mdcreate_swap(struct md_s *sc, struct md_ioctl *mdio, struct thread *td)
1116 {
1117 	vm_ooffset_t npage;
1118 	int error;
1119 
1120 	/*
1121 	 * Range check.  Disallow negative sizes or any size less then the
1122 	 * size of a page.  Then round to a page.
1123 	 */
1124 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1125 		return (EDOM);
1126 
1127 	/*
1128 	 * Allocate an OBJT_SWAP object.
1129 	 *
1130 	 * Note the truncation.
1131 	 */
1132 
1133 	npage = mdio->md_mediasize / PAGE_SIZE;
1134 	if (mdio->md_fwsectors != 0)
1135 		sc->fwsectors = mdio->md_fwsectors;
1136 	if (mdio->md_fwheads != 0)
1137 		sc->fwheads = mdio->md_fwheads;
1138 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1139 	    VM_PROT_DEFAULT, 0, td->td_ucred);
1140 	if (sc->object == NULL)
1141 		return (ENOMEM);
1142 	sc->flags = mdio->md_options & (MD_FORCE | MD_RESERVE);
1143 	if (mdio->md_options & MD_RESERVE) {
1144 		if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1145 			error = EDOM;
1146 			goto finish;
1147 		}
1148 	}
1149 	error = mdsetcred(sc, td->td_ucred);
1150  finish:
1151 	if (error != 0) {
1152 		vm_object_deallocate(sc->object);
1153 		sc->object = NULL;
1154 	}
1155 	return (error);
1156 }
1157 
1158 
1159 static int
1160 xmdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1161 {
1162 	struct md_ioctl *mdio;
1163 	struct md_s *sc;
1164 	int error, i;
1165 	unsigned sectsize;
1166 
1167 	if (md_debug)
1168 		printf("mdctlioctl(%s %lx %p %x %p)\n",
1169 			devtoname(dev), cmd, addr, flags, td);
1170 
1171 	mdio = (struct md_ioctl *)addr;
1172 	if (mdio->md_version != MDIOVERSION)
1173 		return (EINVAL);
1174 
1175 	/*
1176 	 * We assert the version number in the individual ioctl
1177 	 * handlers instead of out here because (a) it is possible we
1178 	 * may add another ioctl in the future which doesn't read an
1179 	 * mdio, and (b) the correct return value for an unknown ioctl
1180 	 * is ENOIOCTL, not EINVAL.
1181 	 */
1182 	error = 0;
1183 	switch (cmd) {
1184 	case MDIOCATTACH:
1185 		switch (mdio->md_type) {
1186 		case MD_MALLOC:
1187 		case MD_PRELOAD:
1188 		case MD_VNODE:
1189 		case MD_SWAP:
1190 			break;
1191 		default:
1192 			return (EINVAL);
1193 		}
1194 		if (mdio->md_sectorsize == 0)
1195 			sectsize = DEV_BSIZE;
1196 		else
1197 			sectsize = mdio->md_sectorsize;
1198 		if (sectsize > MAXPHYS || mdio->md_mediasize < sectsize)
1199 			return (EINVAL);
1200 		if (mdio->md_options & MD_AUTOUNIT)
1201 			sc = mdnew(-1, &error, mdio->md_type);
1202 		else {
1203 			if (mdio->md_unit > INT_MAX)
1204 				return (EINVAL);
1205 			sc = mdnew(mdio->md_unit, &error, mdio->md_type);
1206 		}
1207 		if (sc == NULL)
1208 			return (error);
1209 		if (mdio->md_options & MD_AUTOUNIT)
1210 			mdio->md_unit = sc->unit;
1211 		sc->mediasize = mdio->md_mediasize;
1212 		sc->sectorsize = sectsize;
1213 		error = EDOOFUS;
1214 		switch (sc->type) {
1215 		case MD_MALLOC:
1216 			sc->start = mdstart_malloc;
1217 			error = mdcreate_malloc(sc, mdio);
1218 			break;
1219 		case MD_PRELOAD:
1220 			/*
1221 			 * We disallow attaching preloaded memory disks via
1222 			 * ioctl. Preloaded memory disks are automatically
1223 			 * attached in g_md_init().
1224 			 */
1225 			error = EOPNOTSUPP;
1226 			break;
1227 		case MD_VNODE:
1228 			sc->start = mdstart_vnode;
1229 			error = mdcreate_vnode(sc, mdio, td);
1230 			break;
1231 		case MD_SWAP:
1232 			sc->start = mdstart_swap;
1233 			error = mdcreate_swap(sc, mdio, td);
1234 			break;
1235 		}
1236 		if (error != 0) {
1237 			mddestroy(sc, td);
1238 			return (error);
1239 		}
1240 
1241 		/* Prune off any residual fractional sector */
1242 		i = sc->mediasize % sc->sectorsize;
1243 		sc->mediasize -= i;
1244 
1245 		mdinit(sc);
1246 		return (0);
1247 	case MDIOCDETACH:
1248 		if (mdio->md_mediasize != 0 ||
1249 		    (mdio->md_options & ~MD_FORCE) != 0)
1250 			return (EINVAL);
1251 
1252 		sc = mdfind(mdio->md_unit);
1253 		if (sc == NULL)
1254 			return (ENOENT);
1255 		if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1256 		    !(mdio->md_options & MD_FORCE))
1257 			return (EBUSY);
1258 		return (mddestroy(sc, td));
1259 	case MDIOCRESIZE:
1260 		if ((mdio->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1261 			return (EINVAL);
1262 
1263 		sc = mdfind(mdio->md_unit);
1264 		if (sc == NULL)
1265 			return (ENOENT);
1266 		if (mdio->md_mediasize < sc->sectorsize)
1267 			return (EINVAL);
1268 		if (mdio->md_mediasize < sc->mediasize &&
1269 		    !(sc->flags & MD_FORCE) &&
1270 		    !(mdio->md_options & MD_FORCE))
1271 			return (EBUSY);
1272 		return (mdresize(sc, mdio));
1273 	case MDIOCQUERY:
1274 		sc = mdfind(mdio->md_unit);
1275 		if (sc == NULL)
1276 			return (ENOENT);
1277 		mdio->md_type = sc->type;
1278 		mdio->md_options = sc->flags;
1279 		mdio->md_mediasize = sc->mediasize;
1280 		mdio->md_sectorsize = sc->sectorsize;
1281 		if (sc->type == MD_VNODE)
1282 			error = copyout(sc->file, mdio->md_file,
1283 			    strlen(sc->file) + 1);
1284 		return (error);
1285 	case MDIOCLIST:
1286 		i = 1;
1287 		LIST_FOREACH(sc, &md_softc_list, list) {
1288 			if (i == MDNPAD - 1)
1289 				mdio->md_pad[i] = -1;
1290 			else
1291 				mdio->md_pad[i++] = sc->unit;
1292 		}
1293 		mdio->md_pad[0] = i - 1;
1294 		return (0);
1295 	default:
1296 		return (ENOIOCTL);
1297 	};
1298 }
1299 
1300 static int
1301 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
1302 {
1303 	int error;
1304 
1305 	sx_xlock(&md_sx);
1306 	error = xmdctlioctl(dev, cmd, addr, flags, td);
1307 	sx_xunlock(&md_sx);
1308 	return (error);
1309 }
1310 
1311 static void
1312 md_preloaded(u_char *image, size_t length, const char *name)
1313 {
1314 	struct md_s *sc;
1315 	int error;
1316 
1317 	sc = mdnew(-1, &error, MD_PRELOAD);
1318 	if (sc == NULL)
1319 		return;
1320 	sc->mediasize = length;
1321 	sc->sectorsize = DEV_BSIZE;
1322 	sc->pl_ptr = image;
1323 	sc->pl_len = length;
1324 	sc->start = mdstart_preload;
1325 #ifdef MD_ROOT
1326 	if (sc->unit == 0)
1327 		rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
1328 #endif
1329 	mdinit(sc);
1330 	if (name != NULL) {
1331 		printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
1332 		    MD_NAME, sc->unit, name, length, image);
1333 	}
1334 }
1335 
1336 static void
1337 g_md_init(struct g_class *mp __unused)
1338 {
1339 	caddr_t mod;
1340 	u_char *ptr, *name, *type;
1341 	unsigned len;
1342 	int i;
1343 
1344 	/* figure out log2(NINDIR) */
1345 	for (i = NINDIR, nshift = -1; i; nshift++)
1346 		i >>= 1;
1347 
1348 	mod = NULL;
1349 	sx_init(&md_sx, "MD config lock");
1350 	g_topology_unlock();
1351 	md_uh = new_unrhdr(0, INT_MAX, NULL);
1352 #ifdef MD_ROOT_SIZE
1353 	sx_xlock(&md_sx);
1354 	md_preloaded(mfs_root.start, sizeof(mfs_root.start), NULL);
1355 	sx_xunlock(&md_sx);
1356 #endif
1357 	/* XXX: are preload_* static or do they need Giant ? */
1358 	while ((mod = preload_search_next_name(mod)) != NULL) {
1359 		name = (char *)preload_search_info(mod, MODINFO_NAME);
1360 		if (name == NULL)
1361 			continue;
1362 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
1363 		if (type == NULL)
1364 			continue;
1365 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
1366 			continue;
1367 		ptr = preload_fetch_addr(mod);
1368 		len = preload_fetch_size(mod);
1369 		if (ptr != NULL && len != 0) {
1370 			sx_xlock(&md_sx);
1371 			md_preloaded(ptr, len, name);
1372 			sx_xunlock(&md_sx);
1373 		}
1374 	}
1375 	status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
1376 	    0600, MDCTL_NAME);
1377 	g_topology_lock();
1378 }
1379 
1380 static void
1381 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1382     struct g_consumer *cp __unused, struct g_provider *pp)
1383 {
1384 	struct md_s *mp;
1385 	char *type;
1386 
1387 	mp = gp->softc;
1388 	if (mp == NULL)
1389 		return;
1390 
1391 	switch (mp->type) {
1392 	case MD_MALLOC:
1393 		type = "malloc";
1394 		break;
1395 	case MD_PRELOAD:
1396 		type = "preload";
1397 		break;
1398 	case MD_VNODE:
1399 		type = "vnode";
1400 		break;
1401 	case MD_SWAP:
1402 		type = "swap";
1403 		break;
1404 	default:
1405 		type = "unknown";
1406 		break;
1407 	}
1408 
1409 	if (pp != NULL) {
1410 		if (indent == NULL) {
1411 			sbuf_printf(sb, " u %d", mp->unit);
1412 			sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
1413 			sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
1414 			sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
1415 			sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
1416 			sbuf_printf(sb, " t %s", type);
1417 			if (mp->type == MD_VNODE && mp->vnode != NULL)
1418 				sbuf_printf(sb, " file %s", mp->file);
1419 		} else {
1420 			sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
1421 			    mp->unit);
1422 			sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
1423 			    indent, (uintmax_t) mp->sectorsize);
1424 			sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
1425 			    indent, (uintmax_t) mp->fwheads);
1426 			sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
1427 			    indent, (uintmax_t) mp->fwsectors);
1428 			sbuf_printf(sb, "%s<length>%ju</length>\n",
1429 			    indent, (uintmax_t) mp->mediasize);
1430 			sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
1431 			    (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
1432 			sbuf_printf(sb, "%s<access>%s</access>\n", indent,
1433 			    (mp->flags & MD_READONLY) == 0 ? "read-write":
1434 			    "read-only");
1435 			sbuf_printf(sb, "%s<type>%s</type>\n", indent,
1436 			    type);
1437 			if (mp->type == MD_VNODE && mp->vnode != NULL)
1438 				sbuf_printf(sb, "%s<file>%s</file>\n",
1439 				    indent, mp->file);
1440 		}
1441 	}
1442 }
1443 
1444 static void
1445 g_md_fini(struct g_class *mp __unused)
1446 {
1447 
1448 	sx_destroy(&md_sx);
1449 	if (status_dev != NULL)
1450 		destroy_dev(status_dev);
1451 	delete_unrhdr(md_uh);
1452 }
1453