xref: /freebsd/sys/dev/md/md.c (revision 6adf353a56a161443406b44a45d00c688ca7b857)
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  * 3. All advertising materials mentioning features or use of this software
35  *    must display the following acknowledgement:
36  *	This product includes software developed by the University of
37  *	California, Berkeley and its contributors.
38  * 4. Neither the name of the University nor the names of its contributors
39  *    may be used to endorse or promote products derived from this software
40  *    without specific prior written permission.
41  *
42  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
43  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
46  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
52  * SUCH DAMAGE.
53  *
54  * from: Utah Hdr: vn.c 1.13 94/04/02
55  *
56  *	from: @(#)vn.c	8.6 (Berkeley) 4/1/94
57  * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
58  */
59 
60 #include "opt_md.h"
61 
62 #include <sys/param.h>
63 #include <sys/systm.h>
64 #include <sys/bio.h>
65 #include <sys/conf.h>
66 #include <sys/devicestat.h>
67 #include <sys/disk.h>
68 #include <sys/fcntl.h>
69 #include <sys/kernel.h>
70 #include <sys/linker.h>
71 #include <sys/lock.h>
72 #include <sys/malloc.h>
73 #include <sys/mdioctl.h>
74 #include <sys/mutex.h>
75 #include <sys/namei.h>
76 #include <sys/proc.h>
77 #include <sys/queue.h>
78 #include <sys/sysctl.h>
79 #include <sys/vnode.h>
80 
81 #include <machine/atomic.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_zone.h>
88 #include <vm/swap_pager.h>
89 
90 #define MD_MODVER 1
91 
92 #ifndef MD_NSECT
93 #define MD_NSECT (10000 * 2)
94 #endif
95 
96 MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk");
97 MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors");
98 
99 static int md_debug;
100 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, "");
101 
102 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE)
103 /* Image gets put here: */
104 static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here";
105 static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here";
106 #endif
107 
108 static int	mdrootready;
109 static int	mdunits;
110 static dev_t	status_dev = 0;
111 
112 
113 #define CDEV_MAJOR	95
114 
115 static d_strategy_t mdstrategy;
116 static d_open_t mdopen;
117 static d_close_t mdclose;
118 static d_ioctl_t mdioctl, mdctlioctl;
119 
120 static struct cdevsw md_cdevsw = {
121         /* open */      mdopen,
122         /* close */     mdclose,
123         /* read */      physread,
124         /* write */     physwrite,
125         /* ioctl */     mdioctl,
126         /* poll */      nopoll,
127         /* mmap */      nommap,
128         /* strategy */  mdstrategy,
129         /* name */      MD_NAME,
130         /* maj */       CDEV_MAJOR,
131         /* dump */      nodump,
132         /* psize */     nopsize,
133         /* flags */     D_DISK | D_CANFREE | D_MEMDISK,
134 };
135 
136 static struct cdevsw mdctl_cdevsw = {
137         /* open */      nullopen,
138         /* close */     nullclose,
139         /* read */      noread,
140         /* write */     nowrite,
141         /* ioctl */     mdctlioctl,
142         /* poll */      nopoll,
143         /* mmap */      nommap,
144         /* strategy */  nostrategy,
145         /* name */      MD_NAME,
146         /* maj */       CDEV_MAJOR
147 };
148 
149 static struct cdevsw mddisk_cdevsw;
150 
151 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list);
152 
153 struct md_s {
154 	int unit;
155 	LIST_ENTRY(md_s) list;
156 	struct devstat stats;
157 	struct bio_queue_head bio_queue;
158 	struct disk disk;
159 	dev_t dev;
160 	int busy;
161 	enum md_types type;
162 	unsigned nsect;
163 	unsigned opencount;
164 	unsigned secsize;
165 	unsigned flags;
166 
167 	/* MD_MALLOC related fields */
168 	u_char **secp;
169 
170 	/* MD_PRELOAD related fields */
171 	u_char *pl_ptr;
172 	unsigned pl_len;
173 
174 	/* MD_VNODE related fields */
175 	struct vnode *vnode;
176 	struct ucred *cred;
177 
178 	/* MD_OBJET related fields */
179 	vm_object_t object;
180 };
181 
182 static int
183 mdopen(dev_t dev, int flag, int fmt, struct proc *p)
184 {
185 	struct md_s *sc;
186 	struct disklabel *dl;
187 
188 	if (md_debug)
189 		printf("mdopen(%s %x %x %p)\n",
190 			devtoname(dev), flag, fmt, p);
191 
192 	sc = dev->si_drv1;
193 
194 	dl = &sc->disk.d_label;
195 	bzero(dl, sizeof(*dl));
196 	dl->d_secsize = sc->secsize;
197 	dl->d_nsectors = sc->nsect > 63 ? 63 : sc->nsect;
198 	dl->d_ntracks = 1;
199 	dl->d_secpercyl = dl->d_nsectors * dl->d_ntracks;
200 	dl->d_secperunit = sc->nsect;
201 	dl->d_ncylinders = dl->d_secperunit / dl->d_secpercyl;
202 	sc->opencount++;
203 	return (0);
204 }
205 
206 static int
207 mdclose(dev_t dev, int flags, int fmt, struct proc *p)
208 {
209 	struct md_s *sc = dev->si_drv1;
210 
211 	sc->opencount--;
212 	return (0);
213 }
214 
215 static int
216 mdioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct proc *p)
217 {
218 
219 	if (md_debug)
220 		printf("mdioctl(%s %lx %p %x %p)\n",
221 			devtoname(dev), cmd, addr, flags, p);
222 
223 	return (ENOIOCTL);
224 }
225 
226 static void
227 mdstart_malloc(struct md_s *sc)
228 {
229 	int i;
230 	struct bio *bp;
231 	devstat_trans_flags dop;
232 	u_char *secp, **secpp, *dst;
233 	unsigned secno, nsec, secval, uc;
234 
235 	for (;;) {
236 		/* XXX: LOCK(unique unit numbers) */
237 		bp = bioq_first(&sc->bio_queue);
238 		if (bp)
239 			bioq_remove(&sc->bio_queue, bp);
240 		/* XXX: UNLOCK(unique unit numbers) */
241 		if (!bp)
242 			break;
243 
244 		devstat_start_transaction(&sc->stats);
245 
246 		if (bp->bio_cmd == BIO_DELETE)
247 			dop = DEVSTAT_NO_DATA;
248 		else if (bp->bio_cmd == BIO_READ)
249 			dop = DEVSTAT_READ;
250 		else
251 			dop = DEVSTAT_WRITE;
252 
253 		nsec = bp->bio_bcount / sc->secsize;
254 		secno = bp->bio_pblkno;
255 		dst = bp->bio_data;
256 		while (nsec--) {
257 			secpp = &sc->secp[secno];
258 			if ((uintptr_t)*secpp > 255) {
259 				secp = *secpp;
260 				secval = 0;
261 			} else {
262 				secp = NULL;
263 				secval = (uintptr_t) *secpp;
264 			}
265 
266 			if (md_debug > 2)
267 				printf("%x %p %p %d\n",
268 				    bp->bio_flags, secpp, secp, secval);
269 
270 			if (bp->bio_cmd == BIO_DELETE) {
271 				if (!(sc->flags & MD_RESERVE) && secp != NULL) {
272 					FREE(secp, M_MDSECT);
273 					*secpp = 0;
274 				}
275 			} else if (bp->bio_cmd == BIO_READ) {
276 				if (secp != NULL) {
277 					bcopy(secp, dst, sc->secsize);
278 				} else if (secval) {
279 					for (i = 0; i < sc->secsize; i++)
280 						dst[i] = secval;
281 				} else {
282 					bzero(dst, sc->secsize);
283 				}
284 			} else {
285 				if (sc->flags & MD_COMPRESS) {
286 					uc = dst[0];
287 					for (i = 1; i < sc->secsize; i++)
288 						if (dst[i] != uc)
289 							break;
290 				} else {
291 					i = 0;
292 					uc = 0;
293 				}
294 				if (i == sc->secsize) {
295 					if (secp)
296 						FREE(secp, M_MDSECT);
297 					*secpp = (u_char *)(uintptr_t)uc;
298 				} else {
299 					if (secp == NULL)
300 						MALLOC(secp, u_char *, sc->secsize, M_MDSECT, M_WAITOK);
301 					bcopy(dst, secp, sc->secsize);
302 					*secpp = secp;
303 				}
304 			}
305 			secno++;
306 			dst += sc->secsize;
307 		}
308 		bp->bio_resid = 0;
309 		biofinish(bp, &sc->stats, 0);
310 	}
311 	return;
312 }
313 
314 
315 static void
316 mdstart_preload(struct md_s *sc)
317 {
318 	struct bio *bp;
319 	devstat_trans_flags dop;
320 
321 	for (;;) {
322 		/* XXX: LOCK(unique unit numbers) */
323 		bp = bioq_first(&sc->bio_queue);
324 		if (bp)
325 			bioq_remove(&sc->bio_queue, bp);
326 		/* XXX: UNLOCK(unique unit numbers) */
327 		if (!bp)
328 			break;
329 
330 		devstat_start_transaction(&sc->stats);
331 
332 		if (bp->bio_cmd == BIO_DELETE) {
333 			dop = DEVSTAT_NO_DATA;
334 		} else if (bp->bio_cmd == BIO_READ) {
335 			dop = DEVSTAT_READ;
336 			bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount);
337 		} else {
338 			dop = DEVSTAT_WRITE;
339 			bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount);
340 		}
341 		bp->bio_resid = 0;
342 		biofinish(bp, &sc->stats, 0);
343 	}
344 	return;
345 }
346 
347 static void
348 mdstart_vnode(struct md_s *sc)
349 {
350 	int error;
351 	struct bio *bp;
352 	struct uio auio;
353 	struct iovec aiov;
354 	struct mount *mp;
355 
356 	/*
357 	 * VNODE I/O
358 	 *
359 	 * If an error occurs, we set BIO_ERROR but we do not set
360 	 * B_INVAL because (for a write anyway), the buffer is
361 	 * still valid.
362 	 */
363 
364 	for (;;) {
365 		/* XXX: LOCK(unique unit numbers) */
366 		bp = bioq_first(&sc->bio_queue);
367 		if (bp)
368 			bioq_remove(&sc->bio_queue, bp);
369 		/* XXX: UNLOCK(unique unit numbers) */
370 		if (!bp)
371 			break;
372 
373 		devstat_start_transaction(&sc->stats);
374 
375 		bzero(&auio, sizeof(auio));
376 
377 		aiov.iov_base = bp->bio_data;
378 		aiov.iov_len = bp->bio_bcount;
379 		auio.uio_iov = &aiov;
380 		auio.uio_iovcnt = 1;
381 		auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize;
382 		auio.uio_segflg = UIO_SYSSPACE;
383 		if(bp->bio_cmd == BIO_READ)
384 			auio.uio_rw = UIO_READ;
385 		else
386 			auio.uio_rw = UIO_WRITE;
387 		auio.uio_resid = bp->bio_bcount;
388 		auio.uio_procp = curproc;
389 		if (VOP_ISLOCKED(sc->vnode, NULL))
390 			vprint("unexpected md driver lock", sc->vnode);
391 		if (bp->bio_cmd == BIO_READ) {
392 			vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curproc);
393 			error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
394 		} else {
395 			(void) vn_start_write(sc->vnode, &mp, V_WAIT);
396 			vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curproc);
397 			error = VOP_WRITE(sc->vnode, &auio, 0, sc->cred);
398 			vn_finished_write(mp);
399 		}
400 		VOP_UNLOCK(sc->vnode, 0, curproc);
401 		bp->bio_resid = auio.uio_resid;
402 		biofinish(bp, &sc->stats, error);
403 	}
404 	return;
405 }
406 
407 static void
408 mdstart_swap(struct md_s *sc)
409 {
410 	struct bio *bp;
411 
412 	for (;;) {
413 		/* XXX: LOCK(unique unit numbers) */
414 		bp = bioq_first(&sc->bio_queue);
415 		if (bp)
416 			bioq_remove(&sc->bio_queue, bp);
417 		/* XXX: UNLOCK(unique unit numbers) */
418 		if (!bp)
419 			break;
420 
421 #if 0
422 		devstat_start_transaction(&sc->stats);
423 #endif
424 
425 		if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE))
426 			biodone(bp);
427 		else
428 			vm_pager_strategy(sc->object, bp);
429 
430 #if 0
431 		devstat_end_transaction_bio(&sc->stats, bp);
432 #endif
433 	}
434 	return;
435 }
436 
437 static void
438 mdstrategy(struct bio *bp)
439 {
440 	struct md_s *sc;
441 
442 	if (md_debug > 1)
443 		printf("mdstrategy(%p) %s %x, %d, %ld, %p)\n",
444 		    bp, devtoname(bp->bio_dev), bp->bio_flags, bp->bio_blkno,
445 		    bp->bio_bcount / DEV_BSIZE, bp->bio_data);
446 
447 	sc = bp->bio_dev->si_drv1;
448 
449 	/* XXX: LOCK(sc->lock) */
450 	bioqdisksort(&sc->bio_queue, bp);
451 	/* XXX: UNLOCK(sc->lock) */
452 
453 	if (atomic_cmpset_int(&sc->busy, 0, 1) == 0)
454 		return;
455 
456 	switch (sc->type) {
457 	case MD_MALLOC:
458 		mdstart_malloc(sc);
459 		break;
460 	case MD_PRELOAD:
461 		mdstart_preload(sc);
462 		break;
463 	case MD_VNODE:
464 		mdstart_vnode(sc);
465 		break;
466 	case MD_SWAP:
467 		mdstart_swap(sc);
468 		break;
469 	default:
470 		panic("Impossible md(type)");
471 		break;
472 	}
473 	sc->busy = 0;
474 }
475 
476 static struct md_s *
477 mdfind(int unit)
478 {
479 	struct md_s *sc;
480 
481 	/* XXX: LOCK(unique unit numbers) */
482 	LIST_FOREACH(sc, &md_softc_list, list) {
483 		if (sc->unit == unit)
484 			break;
485 	}
486 	/* XXX: UNLOCK(unique unit numbers) */
487 	return (sc);
488 }
489 
490 static struct md_s *
491 mdnew(int unit)
492 {
493 	struct md_s *sc;
494 	int max = -1;
495 
496 	/* XXX: LOCK(unique unit numbers) */
497 	LIST_FOREACH(sc, &md_softc_list, list) {
498 		if (sc->unit == unit) {
499 			/* XXX: UNLOCK(unique unit numbers) */
500 			return (NULL);
501 		}
502 		if (sc->unit > max)
503 			max = sc->unit;
504 	}
505 	if (unit == -1)
506 		unit = max + 1;
507 	if (unit > DKMAXUNIT)
508 		return (NULL);
509 	MALLOC(sc, struct md_s *,sizeof(*sc), M_MD, M_WAITOK | M_ZERO);
510 	sc->unit = unit;
511 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
512 	/* XXX: UNLOCK(unique unit numbers) */
513 	return (sc);
514 }
515 
516 static void
517 mdinit(struct md_s *sc)
518 {
519 
520 	bioq_init(&sc->bio_queue);
521 	devstat_add_entry(&sc->stats, MD_NAME, sc->unit, sc->secsize,
522 		DEVSTAT_NO_ORDERED_TAGS,
523 		DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER,
524 		DEVSTAT_PRIORITY_OTHER);
525 	sc->dev = disk_create(sc->unit, &sc->disk, 0, &md_cdevsw, &mddisk_cdevsw);
526 	sc->dev->si_drv1 = sc;
527 }
528 
529 /*
530  * XXX: we should check that the range they feed us is mapped.
531  * XXX: we should implement read-only.
532  */
533 
534 static int
535 mdcreate_preload(struct md_ioctl *mdio)
536 {
537 	struct md_s *sc;
538 
539 	if (mdio->md_size == 0)
540 		return(EINVAL);
541 	if (mdio->md_options & ~(MD_AUTOUNIT))
542 		return(EINVAL);
543 	if (mdio->md_options & MD_AUTOUNIT) {
544 		sc = mdnew(-1);
545 		if (sc == NULL)
546 			return (ENOMEM);
547 		mdio->md_unit = sc->unit;
548 	} else {
549 		sc = mdnew(mdio->md_unit);
550 		if (sc == NULL)
551 			return (EBUSY);
552 	}
553 	sc->type = MD_PRELOAD;
554 	sc->secsize = DEV_BSIZE;
555 	sc->nsect = mdio->md_size;
556 	/* Cast to pointer size, then to pointer to avoid warning */
557 	sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base;
558 	sc->pl_len = (mdio->md_size << DEV_BSHIFT);
559 	mdinit(sc);
560 	return (0);
561 }
562 
563 
564 static int
565 mdcreate_malloc(struct md_ioctl *mdio)
566 {
567 	struct md_s *sc;
568 	unsigned u;
569 
570 	if (mdio->md_size == 0)
571 		return(EINVAL);
572 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
573 		return(EINVAL);
574 	/* Compression doesn't make sense if we have reserved space */
575 	if (mdio->md_options & MD_RESERVE)
576 		mdio->md_options &= ~MD_COMPRESS;
577 	if (mdio->md_options & MD_AUTOUNIT) {
578 		sc = mdnew(-1);
579 		if (sc == NULL)
580 			return (ENOMEM);
581 		mdio->md_unit = sc->unit;
582 	} else {
583 		sc = mdnew(mdio->md_unit);
584 		if (sc == NULL)
585 			return (EBUSY);
586 	}
587 	sc->type = MD_MALLOC;
588 	sc->secsize = DEV_BSIZE;
589 	sc->nsect = mdio->md_size;
590 	sc->flags = mdio->md_options & MD_COMPRESS;
591 	MALLOC(sc->secp, u_char **, sc->nsect * sizeof(u_char *), M_MD, M_WAITOK | M_ZERO);
592 	if (mdio->md_options & MD_RESERVE) {
593 		for (u = 0; u < sc->nsect; u++)
594 			MALLOC(sc->secp[u], u_char *, DEV_BSIZE, M_MDSECT, M_WAITOK | M_ZERO);
595 	}
596 	printf("%s%d: Malloc disk\n", MD_NAME, sc->unit);
597 	mdinit(sc);
598 	return (0);
599 }
600 
601 
602 static int
603 mdsetcred(struct md_s *sc, struct ucred *cred)
604 {
605 	char *tmpbuf;
606 	int error = 0;
607 
608 	/*
609 	 * Set credits in our softc
610 	 */
611 
612 	if (sc->cred)
613 		crfree(sc->cred);
614 	sc->cred = crdup(cred);
615 
616 	/*
617 	 * Horrible kludge to establish credentials for NFS  XXX.
618 	 */
619 
620 	if (sc->vnode) {
621 		struct uio auio;
622 		struct iovec aiov;
623 
624 		tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK);
625 		bzero(&auio, sizeof(auio));
626 
627 		aiov.iov_base = tmpbuf;
628 		aiov.iov_len = sc->secsize;
629 		auio.uio_iov = &aiov;
630 		auio.uio_iovcnt = 1;
631 		auio.uio_offset = 0;
632 		auio.uio_rw = UIO_READ;
633 		auio.uio_segflg = UIO_SYSSPACE;
634 		auio.uio_resid = aiov.iov_len;
635 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curproc);
636 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
637 		VOP_UNLOCK(sc->vnode, 0, curproc);
638 		free(tmpbuf, M_TEMP);
639 	}
640 	return (error);
641 }
642 
643 static int
644 mdcreate_vnode(struct md_ioctl *mdio, struct proc *p)
645 {
646 	struct md_s *sc;
647 	struct vattr vattr;
648 	struct nameidata nd;
649 	int error, flags;
650 
651 	if (mdio->md_options & MD_AUTOUNIT) {
652 		sc = mdnew(-1);
653 		mdio->md_unit = sc->unit;
654 	} else {
655 		sc = mdnew(mdio->md_unit);
656 	}
657 	if (sc == NULL)
658 		return (EBUSY);
659 
660 	sc->type = MD_VNODE;
661 
662 	flags = FREAD|FWRITE;
663 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, p);
664 	error = vn_open(&nd, &flags, 0);
665 	if (error) {
666 		if (error != EACCES && error != EPERM && error != EROFS)
667 			return (error);
668 		flags &= ~FWRITE;
669 		sc->flags |= MD_READONLY;
670 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, p);
671 		error = vn_open(&nd, &flags, 0);
672 		if (error)
673 			return (error);
674 	}
675 	NDFREE(&nd, NDF_ONLY_PNBUF);
676 	if (nd.ni_vp->v_type != VREG ||
677 	    (error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p))) {
678 		VOP_UNLOCK(nd.ni_vp, 0, p);
679 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, p);
680 		return (error ? error : EINVAL);
681 	}
682 	VOP_UNLOCK(nd.ni_vp, 0, p);
683 	sc->secsize = DEV_BSIZE;
684 	sc->vnode = nd.ni_vp;
685 
686 	/*
687 	 * If the size is specified, override the file attributes.
688 	 */
689 	if (mdio->md_size)
690 		sc->nsect = mdio->md_size;
691 	else
692 		sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */
693 	if (sc->nsect == 0) {
694 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, p);
695 		return (EINVAL);
696 	}
697 	error = mdsetcred(sc, p->p_ucred);
698 	if (error) {
699 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, p);
700 		return(error);
701 	}
702 	mdinit(sc);
703 	return (0);
704 }
705 
706 static int
707 mddestroy(struct md_s *sc, struct md_ioctl *mdio, struct proc *p)
708 {
709 	unsigned u;
710 
711 	GIANT_REQUIRED;
712 
713 	if (sc->dev != NULL) {
714 		devstat_remove_entry(&sc->stats);
715 		disk_destroy(sc->dev);
716 	}
717 	if (sc->vnode != NULL)
718 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?  FREAD : (FREAD|FWRITE), sc->cred, p);
719 	if (sc->cred != NULL)
720 		crfree(sc->cred);
721 	if (sc->object != NULL) {
722 		vm_pager_deallocate(sc->object);
723 	}
724 	if (sc->secp != NULL) {
725 		for (u = 0; u < sc->nsect; u++)
726 			if ((uintptr_t)sc->secp[u] > 255)
727 				FREE(sc->secp[u], M_MDSECT);
728 		FREE(sc->secp, M_MD);
729 	}
730 
731 	/* XXX: LOCK(unique unit numbers) */
732 	LIST_REMOVE(sc, list);
733 	/* XXX: UNLOCK(unique unit numbers) */
734 	FREE(sc, M_MD);
735 	return (0);
736 }
737 
738 static int
739 mdcreate_swap(struct md_ioctl *mdio, struct proc *p)
740 {
741 	int error;
742 	struct md_s *sc;
743 
744 	GIANT_REQUIRED;
745 
746 	if (mdio->md_options & MD_AUTOUNIT) {
747 		sc = mdnew(-1);
748 		mdio->md_unit = sc->unit;
749 	} else {
750 		sc = mdnew(mdio->md_unit);
751 	}
752 	if (sc == NULL)
753 		return (EBUSY);
754 
755 	sc->type = MD_SWAP;
756 
757 	/*
758 	 * Range check.  Disallow negative sizes or any size less then the
759 	 * size of a page.  Then round to a page.
760 	 */
761 
762 	if (mdio->md_size == 0) {
763 		mddestroy(sc, mdio, p);
764 		return(EDOM);
765 	}
766 
767 	/*
768 	 * Allocate an OBJT_SWAP object.
769 	 *
770 	 * sc_secsize is PAGE_SIZE'd
771 	 *
772 	 * mdio->size is in DEV_BSIZE'd chunks.
773 	 * Note the truncation.
774 	 */
775 
776 	sc->secsize = PAGE_SIZE;
777 	sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE);
778 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0);
779 	if (mdio->md_options & MD_RESERVE) {
780 		if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) {
781 			vm_pager_deallocate(sc->object);
782 			sc->object = NULL;
783 			mddestroy(sc, mdio, p);
784 			return(EDOM);
785 		}
786 	}
787 	error = mdsetcred(sc, p->p_ucred);
788 	if (error)
789 		mddestroy(sc, mdio, p);
790 	else
791 		mdinit(sc);
792 	return(error);
793 }
794 
795 static int
796 mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct proc *p)
797 {
798 	struct md_ioctl *mdio;
799 	struct md_s *sc;
800 
801 	if (md_debug)
802 		printf("mdctlioctl(%s %lx %p %x %p)\n",
803 			devtoname(dev), cmd, addr, flags, p);
804 
805 	mdio = (struct md_ioctl *)addr;
806 	switch (cmd) {
807 	case MDIOCATTACH:
808 		switch (mdio->md_type) {
809 		case MD_MALLOC:
810 			return(mdcreate_malloc(mdio));
811 		case MD_PRELOAD:
812 			return(mdcreate_preload(mdio));
813 		case MD_VNODE:
814 			return(mdcreate_vnode(mdio, p));
815 		case MD_SWAP:
816 			return(mdcreate_swap(mdio, p));
817 		default:
818 			return (EINVAL);
819 		}
820 	case MDIOCDETACH:
821 		if (mdio->md_file != NULL)
822 			return(EINVAL);
823 		if (mdio->md_size != 0)
824 			return(EINVAL);
825 		if (mdio->md_options != 0)
826 			return(EINVAL);
827 		sc = mdfind(mdio->md_unit);
828 		if (sc == NULL)
829 			return (ENOENT);
830 		if (sc->opencount != 0)
831 			return (EBUSY);
832 		switch(sc->type) {
833 		case MD_VNODE:
834 		case MD_SWAP:
835 		case MD_MALLOC:
836 		case MD_PRELOAD:
837 			return(mddestroy(sc, mdio, p));
838 		default:
839 			return (EOPNOTSUPP);
840 		}
841 	case MDIOCQUERY:
842 		sc = mdfind(mdio->md_unit);
843 		if (sc == NULL)
844 			return (ENOENT);
845 		mdio->md_type = sc->type;
846 		mdio->md_options = sc->flags;
847 		switch (sc->type) {
848 		case MD_MALLOC:
849 			mdio->md_size = sc->nsect;
850 			break;
851 		case MD_PRELOAD:
852 			mdio->md_size = sc->nsect;
853 			(u_char *)(uintptr_t)mdio->md_base = sc->pl_ptr;
854 			break;
855 		case MD_SWAP:
856 			mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE);
857 			break;
858 		case MD_VNODE:
859 			mdio->md_size = sc->nsect;
860 			/* XXX fill this in */
861 			mdio->md_file = NULL;
862 			break;
863 		}
864 		return (0);
865 	default:
866 		return (ENOIOCTL);
867 	};
868 	return (ENOIOCTL);
869 }
870 
871 static void
872 md_preloaded(u_char *image, unsigned length)
873 {
874 	struct md_s *sc;
875 
876 	sc = mdnew(-1);
877 	if (sc == NULL)
878 		return;
879 	sc->type = MD_PRELOAD;
880 	sc->secsize = DEV_BSIZE;
881 	sc->nsect = length / DEV_BSIZE;
882 	sc->pl_ptr = image;
883 	sc->pl_len = length;
884 	if (sc->unit == 0)
885 		mdrootready = 1;
886 	mdinit(sc);
887 }
888 
889 static void
890 md_drvinit(void *unused)
891 {
892 
893 	caddr_t mod;
894 	caddr_t c;
895 	u_char *ptr, *name, *type;
896 	unsigned len;
897 
898 #ifdef MD_ROOT_SIZE
899 	md_preloaded(mfs_root, MD_ROOT_SIZE*1024);
900 #endif
901 	mod = NULL;
902 	while ((mod = preload_search_next_name(mod)) != NULL) {
903 		name = (char *)preload_search_info(mod, MODINFO_NAME);
904 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
905 		if (name == NULL)
906 			continue;
907 		if (type == NULL)
908 			continue;
909 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
910 			continue;
911 		c = preload_search_info(mod, MODINFO_ADDR);
912 		ptr = *(u_char **)c;
913 		c = preload_search_info(mod, MODINFO_SIZE);
914 		len = *(unsigned *)c;
915 		printf("%s%d: Preloaded image <%s> %d bytes at %p\n",
916 		    MD_NAME, mdunits, name, len, ptr);
917 		md_preloaded(ptr, len);
918 	}
919 	status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL,
920 	    0600, MDCTL_NAME);
921 }
922 
923 static int
924 md_modevent(module_t mod, int type, void *data)
925 {
926         switch (type) {
927         case MOD_LOAD:
928 		md_drvinit(NULL);
929                 break;
930         case MOD_UNLOAD:
931 		if (!LIST_EMPTY(&md_softc_list))
932 			return EBUSY;
933                 if (status_dev)
934                         destroy_dev(status_dev);
935                 status_dev = 0;
936                 break;
937         default:
938                 break;
939         }
940         return 0;
941 }
942 
943 static moduledata_t md_mod = {
944         MD_NAME,
945         md_modevent,
946         NULL
947 };
948 DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR);
949 MODULE_VERSION(md, MD_MODVER);
950 
951 
952 #ifdef MD_ROOT
953 static void
954 md_takeroot(void *junk)
955 {
956 	if (mdrootready)
957 		rootdevnames[0] = "ufs:/dev/md0c";
958 }
959 
960 SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL);
961 #endif
962 
963