xref: /freebsd/sys/dev/md/md.c (revision 99e8005137088aafb1350e23b113d69b01b0820f)
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_mfs.h"		/* We have adopted some tasks from MFS */
61 #include "opt_md.h"
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/bio.h>
66 #include <sys/conf.h>
67 #include <sys/devicestat.h>
68 #include <sys/disk.h>
69 #include <sys/fcntl.h>
70 #include <sys/kernel.h>
71 #include <sys/linker.h>
72 #include <sys/lock.h>
73 #include <sys/malloc.h>
74 #include <sys/mdioctl.h>
75 #include <sys/mutex.h>
76 #include <sys/namei.h>
77 #include <sys/proc.h>
78 #include <sys/queue.h>
79 #include <sys/sysctl.h>
80 #include <sys/vnode.h>
81 
82 #include <machine/atomic.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/vm_zone.h>
89 #include <vm/swap_pager.h>
90 
91 #define MD_MODVER 1
92 
93 #ifndef MD_NSECT
94 #define MD_NSECT (10000 * 2)
95 #endif
96 
97 MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk");
98 MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors");
99 
100 static int md_debug;
101 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, "");
102 
103 #if defined(MFS_ROOT) && !defined(MD_ROOT)
104 #define MD_ROOT MFS_ROOT
105 #warning "option MFS_ROOT has been superceeded by MD_ROOT"
106 #endif
107 
108 #if defined(MFS_ROOT_SIZE) && !defined(MD_ROOT_SIZE)
109 #define MD_ROOT_SIZE MFS_ROOT_SIZE
110 #warning "option MFS_ROOT_SIZE has been superceeded by MD_ROOT_SIZE"
111 #endif
112 
113 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE)
114 /* Image gets put here: */
115 static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here";
116 static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here";
117 #endif
118 
119 static int	mdrootready;
120 static int	mdunits;
121 static dev_t	status_dev = 0;
122 
123 
124 #define CDEV_MAJOR	95
125 
126 static d_strategy_t mdstrategy;
127 static d_open_t mdopen;
128 static d_ioctl_t mdioctl, mdctlioctl;
129 
130 static struct cdevsw md_cdevsw = {
131         /* open */      mdopen,
132         /* close */     nullclose,
133         /* read */      physread,
134         /* write */     physwrite,
135         /* ioctl */     mdioctl,
136         /* poll */      nopoll,
137         /* mmap */      nommap,
138         /* strategy */  mdstrategy,
139         /* name */      MD_NAME,
140         /* maj */       CDEV_MAJOR,
141         /* dump */      nodump,
142         /* psize */     nopsize,
143         /* flags */     D_DISK | D_CANFREE | D_MEMDISK,
144 };
145 
146 static struct cdevsw mdctl_cdevsw = {
147         /* open */      nullopen,
148         /* close */     nullclose,
149         /* read */      noread,
150         /* write */     nowrite,
151         /* ioctl */     mdctlioctl,
152         /* poll */      nopoll,
153         /* mmap */      nommap,
154         /* strategy */  nostrategy,
155         /* name */      MD_NAME,
156         /* maj */       CDEV_MAJOR
157 };
158 
159 static struct cdevsw mddisk_cdevsw;
160 
161 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(&md_softc_list);
162 
163 struct md_s {
164 	int unit;
165 	LIST_ENTRY(md_s) list;
166 	struct devstat stats;
167 	struct bio_queue_head bio_queue;
168 	struct disk disk;
169 	dev_t dev;
170 	int busy;
171 	enum md_types type;
172 	unsigned nsect;
173 	unsigned secsize;
174 	unsigned flags;
175 
176 	/* MD_MALLOC related fields */
177 	u_char **secp;
178 
179 	/* MD_PRELOAD related fields */
180 	u_char *pl_ptr;
181 	unsigned pl_len;
182 
183 	/* MD_VNODE related fields */
184 	struct vnode *vnode;
185 	struct ucred *cred;
186 
187 	/* MD_OBJET related fields */
188 	vm_object_t object;
189 };
190 
191 static int
192 mdopen(dev_t dev, int flag, int fmt, struct proc *p)
193 {
194 	struct md_s *sc;
195 	struct disklabel *dl;
196 
197 	if (md_debug)
198 		printf("mdopen(%s %x %x %p)\n",
199 			devtoname(dev), flag, fmt, p);
200 
201 	sc = dev->si_drv1;
202 
203 	dl = &sc->disk.d_label;
204 	bzero(dl, sizeof(*dl));
205 	dl->d_secsize = sc->secsize;
206 	dl->d_nsectors = sc->nsect > 63 ? 63 : sc->nsect;
207 	dl->d_ntracks = 1;
208 	dl->d_secpercyl = dl->d_nsectors * dl->d_ntracks;
209 	dl->d_secperunit = sc->nsect;
210 	dl->d_ncylinders = dl->d_secperunit / dl->d_secpercyl;
211 	return (0);
212 }
213 
214 static int
215 mdioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct proc *p)
216 {
217 
218 	if (md_debug)
219 		printf("mdioctl(%s %lx %p %x %p)\n",
220 			devtoname(dev), cmd, addr, flags, p);
221 
222 	return (ENOIOCTL);
223 }
224 
225 static void
226 mdstart_malloc(struct md_s *sc)
227 {
228 	int i;
229 	struct bio *bp;
230 	devstat_trans_flags dop;
231 	u_char *secp, **secpp, *dst;
232 	unsigned secno, nsec, secval, uc;
233 
234 	for (;;) {
235 		/* XXX: LOCK(unique unit numbers) */
236 		bp = bioq_first(&sc->bio_queue);
237 		if (bp)
238 			bioq_remove(&sc->bio_queue, bp);
239 		/* XXX: UNLOCK(unique unit numbers) */
240 		if (!bp)
241 			break;
242 
243 		devstat_start_transaction(&sc->stats);
244 
245 		if (bp->bio_cmd == BIO_DELETE)
246 			dop = DEVSTAT_NO_DATA;
247 		else if (bp->bio_cmd == BIO_READ)
248 			dop = DEVSTAT_READ;
249 		else
250 			dop = DEVSTAT_WRITE;
251 
252 		nsec = bp->bio_bcount / sc->secsize;
253 		secno = bp->bio_pblkno;
254 		dst = bp->bio_data;
255 		while (nsec--) {
256 			secpp = &sc->secp[secno];
257 			if ((uintptr_t)*secpp > 255) {
258 				secp = *secpp;
259 				secval = 0;
260 			} else {
261 				secp = NULL;
262 				secval = (uintptr_t) *secpp;
263 			}
264 
265 			if (md_debug > 2)
266 				printf("%x %p %p %d\n",
267 				    bp->bio_flags, secpp, secp, secval);
268 
269 			if (bp->bio_cmd == BIO_DELETE) {
270 				if (!(sc->flags & MD_RESERVE) && secp != NULL) {
271 					FREE(secp, M_MDSECT);
272 					*secpp = 0;
273 				}
274 			} else if (bp->bio_cmd == BIO_READ) {
275 				if (secp != NULL) {
276 					bcopy(secp, dst, sc->secsize);
277 				} else if (secval) {
278 					for (i = 0; i < sc->secsize; i++)
279 						dst[i] = secval;
280 				} else {
281 					bzero(dst, sc->secsize);
282 				}
283 			} else {
284 				if (sc->flags & MD_COMPRESS) {
285 					uc = dst[0];
286 					for (i = 1; i < sc->secsize; i++)
287 						if (dst[i] != uc)
288 							break;
289 				} else {
290 					i = 0;
291 					uc = 0;
292 				}
293 				if (i == sc->secsize) {
294 					if (secp)
295 						FREE(secp, M_MDSECT);
296 					*secpp = (u_char *)(uintptr_t)uc;
297 				} else {
298 					if (secp == NULL)
299 						MALLOC(secp, u_char *, sc->secsize, M_MDSECT, M_WAITOK);
300 					bcopy(dst, secp, sc->secsize);
301 					*secpp = secp;
302 				}
303 			}
304 			secno++;
305 			dst += sc->secsize;
306 		}
307 		bp->bio_resid = 0;
308 		biofinish(bp, &sc->stats, 0);
309 	}
310 	return;
311 }
312 
313 
314 static void
315 mdstart_preload(struct md_s *sc)
316 {
317 	struct bio *bp;
318 	devstat_trans_flags dop;
319 
320 	for (;;) {
321 		/* XXX: LOCK(unique unit numbers) */
322 		bp = bioq_first(&sc->bio_queue);
323 		if (bp)
324 			bioq_remove(&sc->bio_queue, bp);
325 		/* XXX: UNLOCK(unique unit numbers) */
326 		if (!bp)
327 			break;
328 
329 		devstat_start_transaction(&sc->stats);
330 
331 		if (bp->bio_cmd == BIO_DELETE) {
332 			dop = DEVSTAT_NO_DATA;
333 		} else if (bp->bio_cmd == BIO_READ) {
334 			dop = DEVSTAT_READ;
335 			bcopy(sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_data, bp->bio_bcount);
336 		} else {
337 			dop = DEVSTAT_WRITE;
338 			bcopy(bp->bio_data, sc->pl_ptr + (bp->bio_pblkno << DEV_BSHIFT), bp->bio_bcount);
339 		}
340 		bp->bio_resid = 0;
341 		biofinish(bp, &sc->stats, 0);
342 	}
343 	return;
344 }
345 
346 static void
347 mdstart_vnode(struct md_s *sc)
348 {
349 	int error;
350 	struct bio *bp;
351 	struct uio auio;
352 	struct iovec aiov;
353 	struct mount *mp;
354 
355 	/*
356 	 * VNODE I/O
357 	 *
358 	 * If an error occurs, we set BIO_ERROR but we do not set
359 	 * B_INVAL because (for a write anyway), the buffer is
360 	 * still valid.
361 	 */
362 
363 	for (;;) {
364 		/* XXX: LOCK(unique unit numbers) */
365 		bp = bioq_first(&sc->bio_queue);
366 		if (bp)
367 			bioq_remove(&sc->bio_queue, bp);
368 		/* XXX: UNLOCK(unique unit numbers) */
369 		if (!bp)
370 			break;
371 
372 		devstat_start_transaction(&sc->stats);
373 
374 		bzero(&auio, sizeof(auio));
375 
376 		aiov.iov_base = bp->bio_data;
377 		aiov.iov_len = bp->bio_bcount;
378 		auio.uio_iov = &aiov;
379 		auio.uio_iovcnt = 1;
380 		auio.uio_offset = (vm_ooffset_t)bp->bio_pblkno * sc->secsize;
381 		auio.uio_segflg = UIO_SYSSPACE;
382 		if(bp->bio_cmd == BIO_READ)
383 			auio.uio_rw = UIO_READ;
384 		else
385 			auio.uio_rw = UIO_WRITE;
386 		auio.uio_resid = bp->bio_bcount;
387 		auio.uio_procp = curproc;
388 		if (VOP_ISLOCKED(sc->vnode, NULL))
389 			vprint("unexpected md driver lock", sc->vnode);
390 		if (bp->bio_cmd == BIO_READ) {
391 			vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curproc);
392 			error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
393 		} else {
394 			(void) vn_start_write(sc->vnode, &mp, V_WAIT);
395 			vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curproc);
396 			error = VOP_WRITE(sc->vnode, &auio, 0, sc->cred);
397 			vn_finished_write(mp);
398 		}
399 		VOP_UNLOCK(sc->vnode, 0, curproc);
400 		bp->bio_resid = auio.uio_resid;
401 		biofinish(bp, &sc->stats, error);
402 	}
403 	return;
404 }
405 
406 static void
407 mdstart_swap(struct md_s *sc)
408 {
409 	struct bio *bp;
410 
411 	for (;;) {
412 		/* XXX: LOCK(unique unit numbers) */
413 		bp = bioq_first(&sc->bio_queue);
414 		if (bp)
415 			bioq_remove(&sc->bio_queue, bp);
416 		/* XXX: UNLOCK(unique unit numbers) */
417 		if (!bp)
418 			break;
419 
420 #if 0
421 		devstat_start_transaction(&sc->stats);
422 #endif
423 
424 		if ((bp->bio_cmd == BIO_DELETE) && (sc->flags & MD_RESERVE))
425 			biodone(bp);
426 		else
427 			vm_pager_strategy(sc->object, bp);
428 
429 #if 0
430 		devstat_end_transaction_bio(&sc->stats, bp);
431 #endif
432 	}
433 	return;
434 }
435 
436 static void
437 mdstrategy(struct bio *bp)
438 {
439 	struct md_s *sc;
440 
441 	if (md_debug > 1)
442 		printf("mdstrategy(%p) %s %x, %d, %ld, %p)\n",
443 		    bp, devtoname(bp->bio_dev), bp->bio_flags, bp->bio_blkno,
444 		    bp->bio_bcount / DEV_BSIZE, bp->bio_data);
445 
446 	sc = bp->bio_dev->si_drv1;
447 
448 	/* XXX: LOCK(sc->lock) */
449 	bioqdisksort(&sc->bio_queue, bp);
450 	/* XXX: UNLOCK(sc->lock) */
451 
452 	if (atomic_cmpset_int(&sc->busy, 0, 1) == 0)
453 		return;
454 
455 	switch (sc->type) {
456 	case MD_MALLOC:
457 		mdstart_malloc(sc);
458 		break;
459 	case MD_PRELOAD:
460 		mdstart_preload(sc);
461 		break;
462 	case MD_VNODE:
463 		mdstart_vnode(sc);
464 		break;
465 	case MD_SWAP:
466 		mdstart_swap(sc);
467 		break;
468 	default:
469 		panic("Impossible md(type)");
470 		break;
471 	}
472 	sc->busy = 0;
473 }
474 
475 static struct md_s *
476 mdfind(int unit)
477 {
478 	struct md_s *sc;
479 
480 	/* XXX: LOCK(unique unit numbers) */
481 	LIST_FOREACH(sc, &md_softc_list, list) {
482 		if (sc->unit == unit)
483 			break;
484 	}
485 	/* XXX: UNLOCK(unique unit numbers) */
486 	return (sc);
487 }
488 
489 static struct md_s *
490 mdnew(int unit)
491 {
492 	struct md_s *sc;
493 	int max = -1;
494 
495 	/* XXX: LOCK(unique unit numbers) */
496 	LIST_FOREACH(sc, &md_softc_list, list) {
497 		if (sc->unit == unit) {
498 			/* XXX: UNLOCK(unique unit numbers) */
499 			return (NULL);
500 		}
501 		if (sc->unit > max)
502 			max = sc->unit;
503 	}
504 	if (unit == -1)
505 		unit = max + 1;
506 	if (unit > DKMAXUNIT)
507 		return (NULL);
508 	MALLOC(sc, struct md_s *,sizeof(*sc), M_MD, M_WAITOK | M_ZERO);
509 	sc->unit = unit;
510 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
511 	/* XXX: UNLOCK(unique unit numbers) */
512 	return (sc);
513 }
514 
515 static void
516 mdinit(struct md_s *sc)
517 {
518 
519 	bioq_init(&sc->bio_queue);
520 	devstat_add_entry(&sc->stats, MD_NAME, sc->unit, sc->secsize,
521 		DEVSTAT_NO_ORDERED_TAGS,
522 		DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER,
523 		DEVSTAT_PRIORITY_OTHER);
524 	sc->dev = disk_create(sc->unit, &sc->disk, 0, &md_cdevsw, &mddisk_cdevsw);
525 	sc->dev->si_drv1 = sc;
526 }
527 
528 /*
529  * XXX: we should check that the range they feed us is mapped.
530  * XXX: we should implement read-only.
531  */
532 
533 static int
534 mdcreate_preload(struct md_ioctl *mdio)
535 {
536 	struct md_s *sc;
537 
538 	if (mdio->md_size == 0)
539 		return(EINVAL);
540 	if (mdio->md_options & ~(MD_AUTOUNIT))
541 		return(EINVAL);
542 	if (mdio->md_options & MD_AUTOUNIT) {
543 		sc = mdnew(-1);
544 		if (sc == NULL)
545 			return (ENOMEM);
546 		mdio->md_unit = sc->unit;
547 	} else {
548 		sc = mdnew(mdio->md_unit);
549 		if (sc == NULL)
550 			return (EBUSY);
551 	}
552 	sc->type = MD_PRELOAD;
553 	sc->secsize = DEV_BSIZE;
554 	sc->nsect = mdio->md_size;
555 	/* Cast to pointer size, then to pointer to avoid warning */
556 	sc->pl_ptr = (u_char *)(uintptr_t)mdio->md_base;
557 	sc->pl_len = (mdio->md_size << DEV_BSHIFT);
558 	mdinit(sc);
559 	return (0);
560 }
561 
562 
563 static int
564 mdcreate_malloc(struct md_ioctl *mdio)
565 {
566 	struct md_s *sc;
567 	unsigned u;
568 
569 	if (mdio->md_size == 0)
570 		return(EINVAL);
571 	if (mdio->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
572 		return(EINVAL);
573 	/* Compression doesn't make sense if we have reserved space */
574 	if (mdio->md_options & MD_RESERVE)
575 		mdio->md_options &= ~MD_COMPRESS;
576 	if (mdio->md_options & MD_AUTOUNIT) {
577 		sc = mdnew(-1);
578 		if (sc == NULL)
579 			return (ENOMEM);
580 		mdio->md_unit = sc->unit;
581 	} else {
582 		sc = mdnew(mdio->md_unit);
583 		if (sc == NULL)
584 			return (EBUSY);
585 	}
586 	sc->type = MD_MALLOC;
587 	sc->secsize = DEV_BSIZE;
588 	sc->nsect = mdio->md_size;
589 	sc->flags = mdio->md_options & MD_COMPRESS;
590 	MALLOC(sc->secp, u_char **, sc->nsect * sizeof(u_char *), M_MD, M_WAITOK | M_ZERO);
591 	if (mdio->md_options & MD_RESERVE) {
592 		for (u = 0; u < sc->nsect; u++)
593 			MALLOC(sc->secp[u], u_char *, DEV_BSIZE, M_MDSECT, M_WAITOK | M_ZERO);
594 	}
595 	printf("%s%d: Malloc disk\n", MD_NAME, sc->unit);
596 	mdinit(sc);
597 	return (0);
598 }
599 
600 
601 static int
602 mdsetcred(struct md_s *sc, struct ucred *cred)
603 {
604 	char *tmpbuf;
605 	int error = 0;
606 
607 	/*
608 	 * Set credits in our softc
609 	 */
610 
611 	if (sc->cred)
612 		crfree(sc->cred);
613 	sc->cred = crdup(cred);
614 
615 	/*
616 	 * Horrible kludge to establish credentials for NFS  XXX.
617 	 */
618 
619 	if (sc->vnode) {
620 		struct uio auio;
621 		struct iovec aiov;
622 
623 		tmpbuf = malloc(sc->secsize, M_TEMP, M_WAITOK);
624 		bzero(&auio, sizeof(auio));
625 
626 		aiov.iov_base = tmpbuf;
627 		aiov.iov_len = sc->secsize;
628 		auio.uio_iov = &aiov;
629 		auio.uio_iovcnt = 1;
630 		auio.uio_offset = 0;
631 		auio.uio_rw = UIO_READ;
632 		auio.uio_segflg = UIO_SYSSPACE;
633 		auio.uio_resid = aiov.iov_len;
634 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY, curproc);
635 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
636 		VOP_UNLOCK(sc->vnode, 0, curproc);
637 		free(tmpbuf, M_TEMP);
638 	}
639 	return (error);
640 }
641 
642 static int
643 mdcreate_vnode(struct md_ioctl *mdio, struct proc *p)
644 {
645 	struct md_s *sc;
646 	struct vattr vattr;
647 	struct nameidata nd;
648 	int error, flags;
649 
650 	if (mdio->md_options & MD_AUTOUNIT) {
651 		sc = mdnew(-1);
652 		mdio->md_unit = sc->unit;
653 	} else {
654 		sc = mdnew(mdio->md_unit);
655 	}
656 	if (sc == NULL)
657 		return (EBUSY);
658 
659 	sc->type = MD_VNODE;
660 
661 	flags = FREAD|FWRITE;
662 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, p);
663 	error = vn_open(&nd, &flags, 0);
664 	if (error) {
665 		if (error != EACCES && error != EPERM && error != EROFS)
666 			return (error);
667 		flags &= ~FWRITE;
668 		sc->flags |= MD_READONLY;
669 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, mdio->md_file, p);
670 		error = vn_open(&nd, &flags, 0);
671 		if (error)
672 			return (error);
673 	}
674 	NDFREE(&nd, NDF_ONLY_PNBUF);
675 	if (nd.ni_vp->v_type != VREG ||
676 	    (error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p))) {
677 		VOP_UNLOCK(nd.ni_vp, 0, p);
678 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, p);
679 		return (error ? error : EINVAL);
680 	}
681 	VOP_UNLOCK(nd.ni_vp, 0, p);
682 	sc->secsize = DEV_BSIZE;
683 	sc->vnode = nd.ni_vp;
684 
685 	/*
686 	 * If the size is specified, override the file attributes.
687 	 */
688 	if (mdio->md_size)
689 		sc->nsect = mdio->md_size;
690 	else
691 		sc->nsect = vattr.va_size / sc->secsize; /* XXX: round up ? */
692 	error = mdsetcred(sc, p->p_ucred);
693 	if (error) {
694 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, p);
695 		return(error);
696 	}
697 	mdinit(sc);
698 	return (0);
699 }
700 
701 static int
702 mddestroy(struct md_s *sc, struct md_ioctl *mdio, struct proc *p)
703 {
704 	unsigned u;
705 
706 	if (sc->dev != NULL) {
707 		devstat_remove_entry(&sc->stats);
708 		disk_destroy(sc->dev);
709 	}
710 	if (sc->vnode != NULL)
711 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?  FREAD : (FREAD|FWRITE), sc->cred, p);
712 	if (sc->cred != NULL)
713 		crfree(sc->cred);
714 	if (sc->object != NULL) {
715 		mtx_lock(&vm_mtx);
716 		vm_pager_deallocate(sc->object);
717 		mtx_unlock(&vm_mtx);
718 	}
719 	if (sc->secp != NULL) {
720 		for (u = 0; u < sc->nsect; u++)
721 			if ((uintptr_t)sc->secp[u] > 255)
722 				FREE(sc->secp[u], M_MDSECT);
723 		FREE(sc->secp, M_MD);
724 	}
725 
726 	/* XXX: LOCK(unique unit numbers) */
727 	LIST_REMOVE(sc, list);
728 	/* XXX: UNLOCK(unique unit numbers) */
729 	FREE(sc, M_MD);
730 	return (0);
731 }
732 
733 static int
734 mdcreate_swap(struct md_ioctl *mdio, struct proc *p)
735 {
736 	int error;
737 	struct md_s *sc;
738 
739 	if (mdio->md_options & MD_AUTOUNIT) {
740 		sc = mdnew(-1);
741 		mdio->md_unit = sc->unit;
742 	} else {
743 		sc = mdnew(mdio->md_unit);
744 	}
745 	if (sc == NULL)
746 		return (EBUSY);
747 
748 	sc->type = MD_SWAP;
749 
750 	/*
751 	 * Range check.  Disallow negative sizes or any size less then the
752 	 * size of a page.  Then round to a page.
753 	 */
754 
755 	if (mdio->md_size == 0) {
756 		mddestroy(sc, mdio, p);
757 		return(EDOM);
758 	}
759 
760 	/*
761 	 * Allocate an OBJT_SWAP object.
762 	 *
763 	 * sc_secsize is PAGE_SIZE'd
764 	 *
765 	 * mdio->size is in DEV_BSIZE'd chunks.
766 	 * Note the truncation.
767 	 */
768 
769 	mtx_lock(&vm_mtx);
770 	sc->secsize = PAGE_SIZE;
771 	sc->nsect = mdio->md_size / (PAGE_SIZE / DEV_BSIZE);
772 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, sc->secsize * (vm_offset_t)sc->nsect, VM_PROT_DEFAULT, 0);
773 	if (mdio->md_options & MD_RESERVE) {
774 		if (swap_pager_reserve(sc->object, 0, sc->nsect) < 0) {
775 			vm_pager_deallocate(sc->object);
776 			mtx_unlock(&vm_mtx);
777 			sc->object = NULL;
778 			mddestroy(sc, mdio, p);
779 			return(EDOM);
780 		}
781 	}
782 	mtx_unlock(&vm_mtx);
783 	error = mdsetcred(sc, p->p_ucred);
784 	if (error)
785 		mddestroy(sc, mdio, p);
786 	else
787 		mdinit(sc);
788 	return(error);
789 }
790 
791 static int
792 mdctlioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct proc *p)
793 {
794 	struct md_ioctl *mdio;
795 	struct md_s *sc;
796 
797 	if (md_debug)
798 		printf("mdctlioctl(%s %lx %p %x %p)\n",
799 			devtoname(dev), cmd, addr, flags, p);
800 
801 	mdio = (struct md_ioctl *)addr;
802 	switch (cmd) {
803 	case MDIOCATTACH:
804 		switch (mdio->md_type) {
805 		case MD_MALLOC:
806 			return(mdcreate_malloc(mdio));
807 		case MD_PRELOAD:
808 			return(mdcreate_preload(mdio));
809 		case MD_VNODE:
810 			return(mdcreate_vnode(mdio, p));
811 		case MD_SWAP:
812 			return(mdcreate_swap(mdio, p));
813 		default:
814 			return (EINVAL);
815 		}
816 	case MDIOCDETACH:
817 		if (mdio->md_file != NULL)
818 			return(EINVAL);
819 		if (mdio->md_size != 0)
820 			return(EINVAL);
821 		if (mdio->md_options != 0)
822 			return(EINVAL);
823 		sc = mdfind(mdio->md_unit);
824 		if (sc == NULL)
825 			return (ENOENT);
826 		switch(sc->type) {
827 		case MD_VNODE:
828 		case MD_SWAP:
829 		case MD_MALLOC:
830 		case MD_PRELOAD:
831 			return(mddestroy(sc, mdio, p));
832 		default:
833 			return (EOPNOTSUPP);
834 		}
835 	case MDIOCQUERY:
836 		sc = mdfind(mdio->md_unit);
837 		if (sc == NULL)
838 			return (ENOENT);
839 		mdio->md_type = sc->type;
840 		mdio->md_options = sc->flags;
841 		switch (sc->type) {
842 		case MD_MALLOC:
843 			mdio->md_size = sc->nsect;
844 			break;
845 		case MD_PRELOAD:
846 			mdio->md_size = sc->nsect;
847 			(u_char *)(uintptr_t)mdio->md_base = sc->pl_ptr;
848 			break;
849 		case MD_SWAP:
850 			mdio->md_size = sc->nsect * (PAGE_SIZE / DEV_BSIZE);
851 			break;
852 		case MD_VNODE:
853 			mdio->md_size = sc->nsect;
854 			/* XXX fill this in */
855 			mdio->md_file = NULL;
856 			break;
857 		}
858 		return (0);
859 	default:
860 		return (ENOIOCTL);
861 	};
862 	return (ENOIOCTL);
863 }
864 
865 static void
866 md_preloaded(u_char *image, unsigned length)
867 {
868 	struct md_s *sc;
869 
870 	sc = mdnew(-1);
871 	if (sc == NULL)
872 		return;
873 	sc->type = MD_PRELOAD;
874 	sc->secsize = DEV_BSIZE;
875 	sc->nsect = length / DEV_BSIZE;
876 	sc->pl_ptr = image;
877 	sc->pl_len = length;
878 	if (sc->unit == 0)
879 		mdrootready = 1;
880 	mdinit(sc);
881 }
882 
883 static void
884 md_drvinit(void *unused)
885 {
886 
887 	caddr_t mod;
888 	caddr_t c;
889 	u_char *ptr, *name, *type;
890 	unsigned len;
891 
892 #ifdef MD_ROOT_SIZE
893 	md_preloaded(mfs_root, MD_ROOT_SIZE*1024);
894 #endif
895 	mod = NULL;
896 	while ((mod = preload_search_next_name(mod)) != NULL) {
897 		name = (char *)preload_search_info(mod, MODINFO_NAME);
898 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
899 		if (name == NULL)
900 			continue;
901 		if (type == NULL)
902 			continue;
903 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
904 			continue;
905 		c = preload_search_info(mod, MODINFO_ADDR);
906 		ptr = *(u_char **)c;
907 		c = preload_search_info(mod, MODINFO_SIZE);
908 		len = *(unsigned *)c;
909 		printf("md%d: Preloaded image <%s> %d bytes at %p\n",
910 		   mdunits, name, len, ptr);
911 		md_preloaded(ptr, len);
912 	}
913 	status_dev = make_dev(&mdctl_cdevsw, 0xffff00ff, UID_ROOT, GID_WHEEL, 0600, "mdctl");
914 }
915 
916 static int
917 md_modevent(module_t mod, int type, void *data)
918 {
919         switch (type) {
920         case MOD_LOAD:
921 		md_drvinit(NULL);
922                 break;
923         case MOD_UNLOAD:
924 		if (!LIST_EMPTY(&md_softc_list))
925 			return EBUSY;
926                 if (status_dev)
927                         destroy_dev(status_dev);
928                 status_dev = 0;
929                 break;
930         default:
931                 break;
932         }
933         return 0;
934 }
935 
936 static moduledata_t md_mod = {
937         "md",
938         md_modevent,
939         NULL
940 };
941 DECLARE_MODULE(md, md_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR);
942 MODULE_VERSION(md, MD_MODVER);
943 
944 
945 #ifdef MD_ROOT
946 static void
947 md_takeroot(void *junk)
948 {
949 	if (mdrootready)
950 		rootdevnames[0] = "ufs:/dev/md0c";
951 }
952 
953 SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL);
954 #endif
955 
956