xref: /freebsd/sys/dev/md/md.c (revision 9207b4cff7b8d483f4dd3c62266c2b58819eb7f9)
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_SWAP related fields */
179 	vm_object_t object;
180 };
181 
182 static int
183 mdopen(dev_t dev, int flag, int fmt, struct thread *td)
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, td->td_proc);
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 thread *td)
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 thread *td)
217 {
218 
219 	if (md_debug)
220 		printf("mdioctl(%s %lx %p %x %p)\n",
221 			devtoname(dev), cmd, addr, flags, td);
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_td = curthread;
389 		/*
390 		 * When reading set IO_DIRECT to try to avoid double-caching
391 		 * the data.  When writing IO_DIRECT is not optimal, but we
392 		 * must set IO_NOWDRAIN to avoid a wdrain deadlock.
393 		 */
394 		if (bp->bio_cmd == BIO_READ) {
395 			vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread);
396 			error = VOP_READ(sc->vnode, &auio, IO_DIRECT, sc->cred);
397 		} else {
398 			(void) vn_start_write(sc->vnode, &mp, V_WAIT);
399 			vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread);
400 			error = VOP_WRITE(sc->vnode, &auio, IO_NOWDRAIN, sc->cred);
401 			vn_finished_write(mp);
402 		}
403 		VOP_UNLOCK(sc->vnode, 0, curthread);
404 		bp->bio_resid = auio.uio_resid;
405 		biofinish(bp, &sc->stats, error);
406 	}
407 	return;
408 }
409 
410 static void
411 mdstart_swap(struct md_s *sc)
412 {
413 	struct bio *bp;
414 
415 	for (;;) {
416 		/* XXX: LOCK(unique unit numbers) */
417 		bp = bioq_first(&sc->bio_queue);
418 		if (bp)
419 			bioq_remove(&sc->bio_queue, bp);
420 		/* XXX: UNLOCK(unique unit numbers) */
421 		if (!bp)
422 			break;
423 
424 #if 0
425 		devstat_start_transaction(&sc->stats);
426 #endif
427 
428 		if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE))
429 			biodone(bp);
430 		else
431 			vm_pager_strategy(sc->object, bp);
432 
433 #if 0
434 		devstat_end_transaction_bio(&sc->stats, bp);
435 #endif
436 	}
437 	return;
438 }
439 
440 static void
441 mdstrategy(struct bio *bp)
442 {
443 	struct md_s *sc;
444 
445 	if (md_debug > 1)
446 		printf("mdstrategy(%p) %s %x, %d, %ld, %p)\n",
447 		    bp, devtoname(bp->bio_dev), bp->bio_flags, bp->bio_blkno,
448 		    bp->bio_bcount / DEV_BSIZE, bp->bio_data);
449 
450 	sc = bp->bio_dev->si_drv1;
451 
452 	/* XXX: LOCK(sc->lock) */
453 	bioqdisksort(&sc->bio_queue, bp);
454 	/* XXX: UNLOCK(sc->lock) */
455 
456 	if (atomic_cmpset_int(&sc->busy, 0, 1) == 0)
457 		return;
458 
459 	switch (sc->type) {
460 	case MD_MALLOC:
461 		mdstart_malloc(sc);
462 		break;
463 	case MD_PRELOAD:
464 		mdstart_preload(sc);
465 		break;
466 	case MD_VNODE:
467 		mdstart_vnode(sc);
468 		break;
469 	case MD_SWAP:
470 		mdstart_swap(sc);
471 		break;
472 	default:
473 		panic("Impossible md(type)");
474 		break;
475 	}
476 	sc->busy = 0;
477 }
478 
479 static struct md_s *
480 mdfind(int unit)
481 {
482 	struct md_s *sc;
483 
484 	/* XXX: LOCK(unique unit numbers) */
485 	LIST_FOREACH(sc, &md_softc_list, list) {
486 		if (sc->unit == unit)
487 			break;
488 	}
489 	/* XXX: UNLOCK(unique unit numbers) */
490 	return (sc);
491 }
492 
493 static struct md_s *
494 mdnew(int unit)
495 {
496 	struct md_s *sc;
497 	int max = -1;
498 
499 	/* XXX: LOCK(unique unit numbers) */
500 	LIST_FOREACH(sc, &md_softc_list, list) {
501 		if (sc->unit == unit) {
502 			/* XXX: UNLOCK(unique unit numbers) */
503 			return (NULL);
504 		}
505 		if (sc->unit > max)
506 			max = sc->unit;
507 	}
508 	if (unit == -1)
509 		unit = max + 1;
510 	if (unit > DKMAXUNIT)
511 		return (NULL);
512 	MALLOC(sc, struct md_s *, sizeof(*sc), M_MD, M_WAITOK | M_ZERO);
513 	sc->unit = unit;
514 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
515 	/* XXX: UNLOCK(unique unit numbers) */
516 	return (sc);
517 }
518 
519 static void
520 mdinit(struct md_s *sc)
521 {
522 
523 	bioq_init(&sc->bio_queue);
524 	devstat_add_entry(&sc->stats, MD_NAME, sc->unit, sc->secsize,
525 		DEVSTAT_NO_ORDERED_TAGS,
526 		DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER,
527 		DEVSTAT_PRIORITY_OTHER);
528 	sc->dev = disk_create(sc->unit, &sc->disk, 0, &md_cdevsw, &mddisk_cdevsw);
529 	sc->dev->si_drv1 = sc;
530 }
531 
532 /*
533  * XXX: we should check that the range they feed us is mapped.
534  * XXX: we should implement read-only.
535  */
536 
537 static int
538 mdcreate_preload(struct md_ioctl *mdio)
539 {
540 	struct md_s *sc;
541 
542 	if (mdio->md_size == 0)
543 		return (EINVAL);
544 	if (mdio->md_options & ~(MD_AUTOUNIT))
545 		return (EINVAL);
546 	if (mdio->md_options & MD_AUTOUNIT) {
547 		sc = mdnew(-1);
548 		if (sc == NULL)
549 			return (ENOMEM);
550 		mdio->md_unit = sc->unit;
551 	} else {
552 		sc = mdnew(mdio->md_unit);
553 		if (sc == NULL)
554 			return (EBUSY);
555 	}
556 	sc->type = MD_PRELOAD;
557 	sc->secsize = DEV_BSIZE;
558 	sc->nsect = mdio->md_size;
559 	sc->flags = mdio->md_options & MD_FORCE;
560 	/* Cast to pointer size, then to pointer to avoid warning */
561 	sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base;
562 	sc->pl_len = (mdio->md_size << DEV_BSHIFT);
563 	mdinit(sc);
564 	return (0);
565 }
566 
567 
568 static int
569 mdcreate_malloc(struct md_ioctl *mdio)
570 {
571 	struct md_s *sc;
572 	unsigned u;
573 
574 	if (mdio->md_size == 0)
575 		return (EINVAL);
576 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
577 		return (EINVAL);
578 	/* Compression doesn't make sense if we have reserved space */
579 	if (mdio->md_options & MD_RESERVE)
580 		mdio->md_options &= ~MD_COMPRESS;
581 	if (mdio->md_options & MD_AUTOUNIT) {
582 		sc = mdnew(-1);
583 		if (sc == NULL)
584 			return (ENOMEM);
585 		mdio->md_unit = sc->unit;
586 	} else {
587 		sc = mdnew(mdio->md_unit);
588 		if (sc == NULL)
589 			return (EBUSY);
590 	}
591 	sc->type = MD_MALLOC;
592 	sc->secsize = DEV_BSIZE;
593 	sc->nsect = mdio->md_size;
594 	sc->flags = mdio->md_options & (MD_COMPRESS | MD_FORCE);
595 	MALLOC(sc->secp, u_char **, sc->nsect * sizeof(u_char *), M_MD, M_WAITOK | M_ZERO);
596 	if (mdio->md_options & MD_RESERVE) {
597 		for (u = 0; u < sc->nsect; u++)
598 			MALLOC(sc->secp[u], u_char *, DEV_BSIZE, M_MDSECT, M_WAITOK | M_ZERO);
599 	}
600 	printf("%s%d: Malloc disk\n", MD_NAME, sc->unit);
601 	mdinit(sc);
602 	return (0);
603 }
604 
605 
606 static int
607 mdsetcred(struct md_s *sc, struct ucred *cred)
608 {
609 	char *tmpbuf;
610 	int error = 0;
611 
612 	/*
613 	 * Set credits in our softc
614 	 */
615 
616 	if (sc->cred)
617 		crfree(sc->cred);
618 	sc->cred = crhold(cred);
619 
620 	/*
621 	 * Horrible kludge to establish credentials for NFS  XXX.
622 	 */
623 
624 	if (sc->vnode) {
625 		struct uio auio;
626 		struct iovec aiov;
627 
628 		tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK);
629 		bzero(&auio, sizeof(auio));
630 
631 		aiov.iov_base = tmpbuf;
632 		aiov.iov_len = sc->secsize;
633 		auio.uio_iov = &aiov;
634 		auio.uio_iovcnt = 1;
635 		auio.uio_offset = 0;
636 		auio.uio_rw = UIO_READ;
637 		auio.uio_segflg = UIO_SYSSPACE;
638 		auio.uio_resid = aiov.iov_len;
639 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curthread);
640 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
641 		VOP_UNLOCK(sc->vnode, 0, curthread);
642 		free(tmpbuf, M_TEMP);
643 	}
644 	return (error);
645 }
646 
647 static int
648 mdcreate_vnode(struct md_ioctl *mdio, struct thread *td)
649 {
650 	struct proc *p = td->td_proc;
651 	struct md_s *sc;
652 	struct vattr vattr;
653 	struct nameidata nd;
654 	int error, flags;
655 
656 	if (mdio->md_options & MD_AUTOUNIT) {
657 		sc = mdnew(-1);
658 		mdio->md_unit = sc->unit;
659 	} else {
660 		sc = mdnew(mdio->md_unit);
661 	}
662 	if (sc == NULL)
663 		return (EBUSY);
664 
665 	sc->type = MD_VNODE;
666 	sc->flags = mdio->md_options & MD_FORCE;
667 
668 	flags = FREAD|FWRITE;
669 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td);
670 	error = vn_open(&nd, &flags, 0);
671 	if (error) {
672 		if (error != EACCES && error != EPERM && error != EROFS)
673 			return (error);
674 		flags &= ~FWRITE;
675 		sc->flags |= MD_READONLY;
676 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, td);
677 		error = vn_open(&nd, &flags, 0);
678 		if (error)
679 			return (error);
680 	}
681 	NDFREE(&nd, NDF_ONLY_PNBUF);
682 	if (nd.ni_vp->v_type != VREG ||
683 	    (error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, td))) {
684 		VOP_UNLOCK(nd.ni_vp, 0, td);
685 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, td);
686 		return (error ? error : EINVAL);
687 	}
688 	VOP_UNLOCK(nd.ni_vp, 0, td);
689 	sc->secsize = DEV_BSIZE;
690 	sc->vnode = nd.ni_vp;
691 
692 	/*
693 	 * If the size is specified, override the file attributes.
694 	 */
695 	if (mdio->md_size)
696 		sc->nsect = mdio->md_size;
697 	else
698 		sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */
699 	if (sc->nsect == 0) {
700 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, td);
701 		return (EINVAL);
702 	}
703 	error = mdsetcred(sc, p->p_ucred);
704 	if (error) {
705 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, td);
706 		return (error);
707 	}
708 	mdinit(sc);
709 	return (0);
710 }
711 
712 static int
713 mddestroy(struct md_s *sc, struct thread *td)
714 {
715 	unsigned u;
716 
717 	GIANT_REQUIRED;
718 
719 	if (sc->dev != NULL) {
720 		devstat_remove_entry(&sc->stats);
721 		disk_destroy(sc->dev);
722 	}
723 	if (sc->vnode != NULL)
724 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
725 		    FREAD : (FREAD|FWRITE), sc->cred, td);
726 	if (sc->cred != NULL)
727 		crfree(sc->cred);
728 	if (sc->object != NULL) {
729 		vm_pager_deallocate(sc->object);
730 	}
731 	if (sc->secp != NULL) {
732 		for (u = 0; u < sc->nsect; u++)
733 			if ((uintptr_t)sc->secp[u] > 255)
734 				FREE(sc->secp[u], M_MDSECT);
735 		FREE(sc->secp, M_MD);
736 	}
737 
738 	/* XXX: LOCK(unique unit numbers) */
739 	LIST_REMOVE(sc, list);
740 	/* XXX: UNLOCK(unique unit numbers) */
741 	FREE(sc, M_MD);
742 	return (0);
743 }
744 
745 static int
746 mdcreate_swap(struct md_ioctl *mdio, struct thread *td)
747 {
748 	int error;
749 	struct md_s *sc;
750 
751 	GIANT_REQUIRED;
752 
753 	if (mdio->md_options & MD_AUTOUNIT) {
754 		sc = mdnew(-1);
755 		mdio->md_unit = sc->unit;
756 	} else {
757 		sc = mdnew(mdio->md_unit);
758 	}
759 	if (sc == NULL)
760 		return (EBUSY);
761 
762 	sc->type = MD_SWAP;
763 
764 	/*
765 	 * Range check.  Disallow negative sizes or any size less then the
766 	 * size of a page.  Then round to a page.
767 	 */
768 
769 	if (mdio->md_size == 0) {
770 		mddestroy(sc, td);
771 		return (EDOM);
772 	}
773 
774 	/*
775 	 * Allocate an OBJT_SWAP object.
776 	 *
777 	 * sc_secsize is PAGE_SIZE'd
778 	 *
779 	 * mdio->size is in DEV_BSIZE'd chunks.
780 	 * Note the truncation.
781 	 */
782 
783 	sc->secsize = PAGE_SIZE;
784 	sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE);
785 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0);
786 	sc->flags = mdio->md_options & MD_FORCE;
787 	if (mdio->md_options & MD_RESERVE) {
788 		if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) {
789 			vm_pager_deallocate(sc->object);
790 			sc->object = NULL;
791 			mddestroy(sc, td);
792 			return (EDOM);
793 		}
794 	}
795 	error = mdsetcred(sc, td->td_proc->p_ucred);
796 	if (error)
797 		mddestroy(sc, td);
798 	else
799 		mdinit(sc);
800 	return (error);
801 }
802 
803 static int
804 mddetach(int unit, struct thread *td)
805 {
806 	struct md_s *sc;
807 
808 	sc = mdfind(unit);
809 	if (sc == NULL)
810 		return (ENOENT);
811 	if (sc->opencount != 0 && !(sc->flags & MD_FORCE))
812 		return (EBUSY);
813 	switch(sc->type) {
814 	case MD_VNODE:
815 	case MD_SWAP:
816 	case MD_MALLOC:
817 	case MD_PRELOAD:
818 		return (mddestroy(sc, td));
819 	default:
820 		return (EOPNOTSUPP);
821 	}
822 }
823 
824 static int
825 mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
826 {
827 	struct md_ioctl *mdio;
828 	struct md_s *sc;
829 
830 	if (md_debug)
831 		printf("mdctlioctl(%s %lx %p %x %p)\n",
832 			devtoname(dev), cmd, addr, flags, td);
833 
834 	mdio = (struct md_ioctl *)addr;
835 	switch (cmd) {
836 	case MDIOCATTACH:
837 		switch (mdio->md_type) {
838 		case MD_MALLOC:
839 			return (mdcreate_malloc(mdio));
840 		case MD_PRELOAD:
841 			return (mdcreate_preload(mdio));
842 		case MD_VNODE:
843 			return (mdcreate_vnode(mdio, td));
844 		case MD_SWAP:
845 			return (mdcreate_swap(mdio, td));
846 		default:
847 			return (EINVAL);
848 		}
849 	case MDIOCDETACH:
850 		if (mdio->md_file != NULL || mdio->md_size != 0 ||
851 		    mdio->md_options != 0)
852 			return (EINVAL);
853 		return (mddetach(mdio->md_unit, td));
854 	case MDIOCQUERY:
855 		sc = mdfind(mdio->md_unit);
856 		if (sc == NULL)
857 			return (ENOENT);
858 		mdio->md_type = sc->type;
859 		mdio->md_options = sc->flags;
860 		switch (sc->type) {
861 		case MD_MALLOC:
862 			mdio->md_size = sc->nsect;
863 			break;
864 		case MD_PRELOAD:
865 			mdio->md_size = sc->nsect;
866 			(u_char *)(uintptr_t)mdio->md_base = sc->pl_ptr;
867 			break;
868 		case MD_SWAP:
869 			mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE);
870 			break;
871 		case MD_VNODE:
872 			mdio->md_size = sc->nsect;
873 			/* XXX fill this in */
874 			mdio->md_file = NULL;
875 			break;
876 		}
877 		return (0);
878 	default:
879 		return (ENOIOCTL);
880 	};
881 	return (ENOIOCTL);
882 }
883 
884 static void
885 md_preloaded(u_char *image, unsigned length)
886 {
887 	struct md_s *sc;
888 
889 	sc = mdnew(-1);
890 	if (sc == NULL)
891 		return;
892 	sc->type = MD_PRELOAD;
893 	sc->secsize = DEV_BSIZE;
894 	sc->nsect = length / DEV_BSIZE;
895 	sc->pl_ptr = image;
896 	sc->pl_len = length;
897 	if (sc->unit == 0)
898 		mdrootready = 1;
899 	mdinit(sc);
900 }
901 
902 static void
903 md_drvinit(void *unused)
904 {
905 
906 	caddr_t mod;
907 	caddr_t c;
908 	u_char *ptr, *name, *type;
909 	unsigned len;
910 
911 #ifdef MD_ROOT_SIZE
912 	md_preloaded(mfs_root, MD_ROOT_SIZE*1024);
913 #endif
914 	mod = NULL;
915 	while ((mod = preload_search_next_name(mod)) != NULL) {
916 		name = (char *)preload_search_info(mod, MODINFO_NAME);
917 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
918 		if (name == NULL)
919 			continue;
920 		if (type == NULL)
921 			continue;
922 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
923 			continue;
924 		c = preload_search_info(mod, MODINFO_ADDR);
925 		ptr = *(u_char **)c;
926 		c = preload_search_info(mod, MODINFO_SIZE);
927 		len = *(unsigned *)c;
928 		printf("%s%d: Preloaded image <%s> %d bytes at %p\n",
929 		    MD_NAME, mdunits, name, len, ptr);
930 		md_preloaded(ptr, len);
931 	}
932 	status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL,
933 	    0600, MDCTL_NAME);
934 }
935 
936 static int
937 md_modevent(module_t mod, int type, void *data)
938 {
939 	int error;
940 	struct md_s *sc;
941 
942 	switch (type) {
943 	case MOD_LOAD:
944 		md_drvinit(NULL);
945 		break;
946 	case MOD_UNLOAD:
947 		LIST_FOREACH(sc, &md_softc_list, list) {
948 			error = mddetach(sc->unit, curthread);
949 			if (error != 0)
950 				return (error);
951 		}
952 		if (status_dev)
953 			destroy_dev(status_dev);
954 		status_dev = 0;
955 		break;
956 	default:
957 		break;
958 	}
959 	return (0);
960 }
961 
962 static moduledata_t md_mod = {
963 	MD_NAME,
964 	md_modevent,
965 	NULL
966 };
967 DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR);
968 MODULE_VERSION(md, MD_MODVER);
969 
970 
971 #ifdef MD_ROOT
972 static void
973 md_takeroot(void *junk)
974 {
975 	if (mdrootready)
976 		rootdevnames[0] = "ufs:/dev/md0c";
977 }
978 
979 SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL);
980 #endif
981