xref: /dragonfly/sys/dev/disk/vn/vn.c (revision 9dbf638f)
1 /*
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1990, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * from: Utah Hdr: vn.c 1.13 94/04/02
39  *
40  *	from: @(#)vn.c	8.6 (Berkeley) 4/1/94
41  * $FreeBSD: src/sys/dev/vn/vn.c,v 1.105.2.4 2001/11/18 07:11:00 dillon Exp $
42  * $DragonFly: src/sys/dev/disk/vn/vn.c,v 1.6 2003/07/21 05:50:38 dillon Exp $
43  */
44 
45 /*
46  * Vnode disk driver.
47  *
48  * Block/character interface to a vnode.  Allows one to treat a file
49  * as a disk (e.g. build a filesystem in it, mount it, etc.).
50  *
51  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
52  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
53  * local buffer cache.
54  *
55  * NOTE 2: There is a security issue involved with this driver.
56  * Once mounted all access to the contents of the "mapped" file via
57  * the special file is controlled by the permissions on the special
58  * file, the protection of the mapped file is ignored (effectively,
59  * by using root credentials in all transactions).
60  *
61  * NOTE 3: Doesn't interact with leases, should it?
62  */
63 
64 #include <sys/param.h>
65 #include <sys/systm.h>
66 #include <sys/kernel.h>
67 #include <sys/proc.h>
68 #include <sys/namei.h>
69 #include <sys/buf.h>
70 #include <sys/malloc.h>
71 #include <sys/mount.h>
72 #include <sys/vnode.h>
73 #include <sys/fcntl.h>
74 #include <sys/conf.h>
75 #include <sys/disklabel.h>
76 #include <sys/diskslice.h>
77 #include <sys/stat.h>
78 #include <sys/conf.h>
79 #include <sys/module.h>
80 #include <sys/vnioctl.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_pager.h>
86 #include <vm/vm_pageout.h>
87 #include <vm/swap_pager.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_zone.h>
90 
91 static	d_ioctl_t	vnioctl;
92 static	d_open_t	vnopen;
93 static	d_close_t	vnclose;
94 static	d_psize_t	vnsize;
95 static	d_strategy_t	vnstrategy;
96 
97 #define CDEV_MAJOR 43
98 
99 #define VN_BSIZE_BEST	8192
100 
101 /*
102  * cdevsw
103  *	D_DISK		we want to look like a disk
104  *	D_CANFREE	We support B_FREEBUF
105  */
106 
107 static struct cdevsw vn_cdevsw = {
108 	/* name */	"vn",
109 	/* maj */	CDEV_MAJOR,
110 	/* flags */	D_DISK|D_CANFREE,
111 	/* port */	NULL,
112 	/* autoq */	0,
113 
114 	/* open */	vnopen,
115 	/* close */	vnclose,
116 	/* read */	physread,
117 	/* write */	physwrite,
118 	/* ioctl */	vnioctl,
119 	/* poll */	nopoll,
120 	/* mmap */	nommap,
121 	/* strategy */	vnstrategy,
122 	/* dump */	nodump,
123 	/* psize */	vnsize
124 };
125 
126 #define	getvnbuf()	\
127 	((struct buf *)malloc(sizeof(struct buf), M_DEVBUF, M_WAITOK))
128 
129 #define putvnbuf(bp)	\
130 	free((caddr_t)(bp), M_DEVBUF)
131 
132 struct vn_softc {
133 	int		sc_unit;
134 	int		sc_flags;	/* flags 			*/
135 	int		sc_size;	/* size of vn, sc_secsize scale	*/
136 	int		sc_secsize;	/* sector size			*/
137 	struct diskslices *sc_slices;
138 	struct vnode	*sc_vp;		/* vnode if not NULL		*/
139 	vm_object_t	sc_object;	/* backing object if not NULL	*/
140 	struct ucred	*sc_cred;	/* credentials 			*/
141 	int		 sc_maxactive;	/* max # of active requests 	*/
142 	struct buf	 sc_tab;	/* transfer queue 		*/
143 	u_long		 sc_options;	/* options 			*/
144 	dev_t		 sc_devlist;	/* devices that refer to this unit */
145 	SLIST_ENTRY(vn_softc) sc_list;
146 };
147 
148 static SLIST_HEAD(, vn_softc) vn_list;
149 
150 /* sc_flags */
151 #define VNF_INITED	0x01
152 #define	VNF_READONLY	0x02
153 
154 static u_long	vn_options;
155 
156 #define IFOPT(vn,opt) if (((vn)->sc_options|vn_options) & (opt))
157 #define TESTOPT(vn,opt) (((vn)->sc_options|vn_options) & (opt))
158 
159 static int	vnsetcred (struct vn_softc *vn, struct ucred *cred);
160 static void	vnclear (struct vn_softc *vn);
161 static int	vn_modevent (module_t, int, void *);
162 static int 	vniocattach_file (struct vn_softc *, struct vn_ioctl *, dev_t dev, int flag, struct thread *p);
163 static int 	vniocattach_swap (struct vn_softc *, struct vn_ioctl *, dev_t dev, int flag, struct thread *p);
164 
165 static	int
166 vnclose(dev_t dev, int flags, int mode, struct thread *td)
167 {
168 	struct vn_softc *vn = dev->si_drv1;
169 
170 	IFOPT(vn, VN_LABELS)
171 		if (vn->sc_slices != NULL)
172 			dsclose(dev, mode, vn->sc_slices);
173 	return (0);
174 }
175 
176 static struct vn_softc *
177 vnfindvn(dev_t dev)
178 {
179 	int unit;
180 	struct vn_softc *vn;
181 
182 	unit = dkunit(dev);
183 	vn = dev->si_drv1;
184 	if (!vn) {
185 		SLIST_FOREACH(vn, &vn_list, sc_list) {
186 			if (vn->sc_unit == unit) {
187 				dev->si_drv2 = vn->sc_devlist;
188 				vn->sc_devlist = dev;
189 				dev->si_drv1 = vn;
190 				break;
191 			}
192 		}
193 	}
194 	if (!vn) {
195 		vn = malloc(sizeof *vn, M_DEVBUF, M_WAITOK);
196 		if (!vn)
197 			return (NULL);
198 		bzero(vn, sizeof *vn);
199 		vn->sc_unit = unit;
200 		dev->si_drv1 = vn;
201 		vn->sc_devlist = make_dev(&vn_cdevsw, 0,
202 		    UID_ROOT, GID_OPERATOR, 0640, "vn%d", unit);
203 		vn->sc_devlist->si_drv1 = vn;
204 		vn->sc_devlist->si_drv2 = NULL;
205 		if (vn->sc_devlist != dev) {
206 			dev->si_drv2 = vn->sc_devlist;
207 			vn->sc_devlist = dev;
208 		}
209 		SLIST_INSERT_HEAD(&vn_list, vn, sc_list);
210 	}
211 	return (vn);
212 }
213 
214 static	int
215 vnopen(dev_t dev, int flags, int mode, struct thread *td)
216 {
217 	struct vn_softc *vn;
218 
219 	/*
220 	 * Locate preexisting device
221 	 */
222 
223 	if ((vn = dev->si_drv1) == NULL)
224 		vn = vnfindvn(dev);
225 
226 	/*
227 	 * Update si_bsize fields for device.  This data will be overriden by
228 	 * the slice/parition code for vn accesses through partitions, and
229 	 * used directly if you open the 'whole disk' device.
230 	 *
231 	 * si_bsize_best must be reinitialized in case VN has been
232 	 * reconfigured, plus make it at least VN_BSIZE_BEST for efficiency.
233 	 */
234 	dev->si_bsize_phys = vn->sc_secsize;
235 	dev->si_bsize_best = vn->sc_secsize;
236 	if (dev->si_bsize_best < VN_BSIZE_BEST)
237 		dev->si_bsize_best = VN_BSIZE_BEST;
238 
239 	if ((flags & FWRITE) && (vn->sc_flags & VNF_READONLY))
240 		return (EACCES);
241 
242 	IFOPT(vn, VN_FOLLOW)
243 		printf("vnopen(%s, 0x%x, 0x%x, %p)\n",
244 		    devtoname(dev), flags, mode, (void *)td);
245 
246 	/*
247 	 * Initialize label
248 	 */
249 
250 	IFOPT(vn, VN_LABELS) {
251 		if (vn->sc_flags & VNF_INITED) {
252 			struct disklabel label;
253 
254 			/* Build label for whole disk. */
255 			bzero(&label, sizeof label);
256 			label.d_secsize = vn->sc_secsize;
257 			label.d_nsectors = 32;
258 			label.d_ntracks = 64 / (vn->sc_secsize / DEV_BSIZE);
259 			label.d_secpercyl = label.d_nsectors * label.d_ntracks;
260 			label.d_ncylinders = vn->sc_size / label.d_secpercyl;
261 			label.d_secperunit = vn->sc_size;
262 			label.d_partitions[RAW_PART].p_size = vn->sc_size;
263 
264 			return (dsopen(dev, mode, 0, &vn->sc_slices, &label));
265 		}
266 		if (dkslice(dev) != WHOLE_DISK_SLICE ||
267 		    dkpart(dev) != RAW_PART ||
268 		    mode != S_IFCHR) {
269 			return (ENXIO);
270 		}
271 	}
272 	return(0);
273 }
274 
275 /*
276  *	vnstrategy:
277  *
278  *	Run strategy routine for VN device.  We use VOP_READ/VOP_WRITE calls
279  *	for vnode-backed vn's, and the new vm_pager_strategy() call for
280  *	vm_object-backed vn's.
281  *
282  *	Currently B_ASYNC is only partially handled - for OBJT_SWAP I/O only.
283  *
284  *	NOTE: bp->b_blkno is DEV_BSIZE'd.  We must generate bp->b_pblkno for
285  *	our uio or vn_pager_strategy() call that is vn->sc_secsize'd
286  */
287 
288 static	void
289 vnstrategy(struct buf *bp)
290 {
291 	int unit;
292 	struct vn_softc *vn;
293 	int error;
294 
295 	unit = dkunit(bp->b_dev);
296 	vn = bp->b_dev->si_drv1;
297 	if (!vn)
298 		vn = vnfindvn(bp->b_dev);
299 
300 	IFOPT(vn, VN_DEBUG)
301 		printf("vnstrategy(%p): unit %d\n", bp, unit);
302 
303 	if ((vn->sc_flags & VNF_INITED) == 0) {
304 		bp->b_error = ENXIO;
305 		bp->b_flags |= B_ERROR;
306 		biodone(bp);
307 		return;
308 	}
309 
310 	bp->b_resid = bp->b_bcount;
311 
312 	IFOPT(vn, VN_LABELS) {
313 		if (vn->sc_slices != NULL && dscheck(bp, vn->sc_slices) <= 0) {
314 			bp->b_flags |= B_INVAL;
315 			biodone(bp);
316 			return;
317 		}
318 	} else {
319 		int pbn;	/* in sc_secsize chunks */
320 		long sz;	/* in sc_secsize chunks */
321 
322 		/*
323 		 * Check for required alignment.  Transfers must be a valid
324 		 * multiple of the sector size.
325 		 */
326 		if (bp->b_bcount % vn->sc_secsize != 0 ||
327 		    bp->b_blkno % (vn->sc_secsize / DEV_BSIZE) != 0) {
328 			bp->b_error = EINVAL;
329 			bp->b_flags |= B_ERROR | B_INVAL;
330 			biodone(bp);
331 			return;
332 		}
333 
334 		pbn = bp->b_blkno / (vn->sc_secsize / DEV_BSIZE);
335 		sz = howmany(bp->b_bcount, vn->sc_secsize);
336 
337 		/*
338 		 * If out of bounds return an error.  If at the EOF point,
339 		 * simply read or write less.
340 		 */
341 		if (pbn < 0 || pbn >= vn->sc_size) {
342 			if (pbn != vn->sc_size) {
343 				bp->b_error = EINVAL;
344 				bp->b_flags |= B_ERROR | B_INVAL;
345 			}
346 			biodone(bp);
347 			return;
348 		}
349 
350 		/*
351 		 * If the request crosses EOF, truncate the request.
352 		 */
353 		if (pbn + sz > vn->sc_size) {
354 			bp->b_bcount = (vn->sc_size - pbn) * vn->sc_secsize;
355 			bp->b_resid = bp->b_bcount;
356 		}
357 		bp->b_pblkno = pbn;
358 	}
359 
360 	if (vn->sc_vp && (bp->b_flags & B_FREEBUF)) {
361 		/*
362 		 * Not handled for vnode-backed element yet.
363 		 */
364 		biodone(bp);
365 	} else if (vn->sc_vp) {
366 		/*
367 		 * VNODE I/O
368 		 *
369 		 * If an error occurs, we set B_ERROR but we do not set
370 		 * B_INVAL because (for a write anyway), the buffer is
371 		 * still valid.
372 		 */
373 		struct uio auio;
374 		struct iovec aiov;
375 
376 		bzero(&auio, sizeof(auio));
377 
378 		aiov.iov_base = bp->b_data;
379 		aiov.iov_len = bp->b_bcount;
380 		auio.uio_iov = &aiov;
381 		auio.uio_iovcnt = 1;
382 		auio.uio_offset = (vm_ooffset_t)bp->b_pblkno * vn->sc_secsize;
383 		auio.uio_segflg = UIO_SYSSPACE;
384 		if( bp->b_flags & B_READ)
385 			auio.uio_rw = UIO_READ;
386 		else
387 			auio.uio_rw = UIO_WRITE;
388 		auio.uio_resid = bp->b_bcount;
389 		auio.uio_td = curthread;
390 		vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY, curthread);
391 		if (bp->b_flags & B_READ)
392 			error = VOP_READ(vn->sc_vp, &auio, IO_DIRECT, vn->sc_cred);
393 		else
394 			error = VOP_WRITE(vn->sc_vp, &auio, IO_NOWDRAIN, vn->sc_cred);
395 		VOP_UNLOCK(vn->sc_vp, 0, curthread);
396 		bp->b_resid = auio.uio_resid;
397 
398 		if (error) {
399 			bp->b_error = error;
400 			bp->b_flags |= B_ERROR;
401 		}
402 		biodone(bp);
403 	} else if (vn->sc_object) {
404 		/*
405 		 * OBJT_SWAP I/O
406 		 *
407 		 * ( handles read, write, freebuf )
408 		 *
409 		 * Note: if we pre-reserved swap, B_FREEBUF is disabled
410 		 */
411 		KASSERT((bp->b_bufsize & (vn->sc_secsize - 1)) == 0,
412 		    ("vnstrategy: buffer %p too small for physio", bp));
413 
414 		if ((bp->b_flags & B_FREEBUF) && TESTOPT(vn, VN_RESERVE)) {
415 			biodone(bp);
416 		} else {
417 			vm_pager_strategy(vn->sc_object, bp);
418 		}
419 	} else {
420 		bp->b_flags |= B_ERROR;
421 		bp->b_error = EINVAL;
422 		biodone(bp);
423 	}
424 }
425 
426 /* ARGSUSED */
427 static	int
428 vnioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct thread *td)
429 {
430 	struct vn_softc *vn;
431 	struct vn_ioctl *vio;
432 	int error;
433 	u_long *f;
434 
435 	vn = dev->si_drv1;
436 	IFOPT(vn,VN_FOLLOW)
437 		printf("vnioctl(%s, 0x%lx, %p, 0x%x, %p): unit %d\n",
438 		    devtoname(dev), cmd, (void *)data, flag, (void *)td,
439 		    dkunit(dev));
440 
441 	switch (cmd) {
442 	case VNIOCATTACH:
443 	case VNIOCDETACH:
444 	case VNIOCGSET:
445 	case VNIOCGCLEAR:
446 	case VNIOCUSET:
447 	case VNIOCUCLEAR:
448 		goto vn_specific;
449 	}
450 
451 	IFOPT(vn,VN_LABELS) {
452 		if (vn->sc_slices != NULL) {
453 			error = dsioctl(dev, cmd, data, flag, &vn->sc_slices);
454 			if (error != ENOIOCTL)
455 				return (error);
456 		}
457 		if (dkslice(dev) != WHOLE_DISK_SLICE ||
458 		    dkpart(dev) != RAW_PART)
459 			return (ENOTTY);
460 	}
461 
462     vn_specific:
463 
464 	error = suser(td);
465 	if (error)
466 		return (error);
467 
468 	vio = (struct vn_ioctl *)data;
469 	f = (u_long*)data;
470 	switch (cmd) {
471 
472 	case VNIOCATTACH:
473 		if (vn->sc_flags & VNF_INITED)
474 			return(EBUSY);
475 
476 		if (vio->vn_file == NULL)
477 			error = vniocattach_swap(vn, vio, dev, flag, td);
478 		else
479 			error = vniocattach_file(vn, vio, dev, flag, td);
480 		break;
481 
482 	case VNIOCDETACH:
483 		if ((vn->sc_flags & VNF_INITED) == 0)
484 			return(ENXIO);
485 		/*
486 		 * XXX handle i/o in progress.  Return EBUSY, or wait, or
487 		 * flush the i/o.
488 		 * XXX handle multiple opens of the device.  Return EBUSY,
489 		 * or revoke the fd's.
490 		 * How are these problems handled for removable and failing
491 		 * hardware devices? (Hint: They are not)
492 		 */
493 		vnclear(vn);
494 		IFOPT(vn, VN_FOLLOW)
495 			printf("vnioctl: CLRed\n");
496 		break;
497 
498 	case VNIOCGSET:
499 		vn_options |= *f;
500 		*f = vn_options;
501 		break;
502 
503 	case VNIOCGCLEAR:
504 		vn_options &= ~(*f);
505 		*f = vn_options;
506 		break;
507 
508 	case VNIOCUSET:
509 		vn->sc_options |= *f;
510 		*f = vn->sc_options;
511 		break;
512 
513 	case VNIOCUCLEAR:
514 		vn->sc_options &= ~(*f);
515 		*f = vn->sc_options;
516 		break;
517 
518 	default:
519 		error = ENOTTY;
520 		break;
521 	}
522 	return(error);
523 }
524 
525 /*
526  *	vniocattach_file:
527  *
528  *	Attach a file to a VN partition.  Return the size in the vn_size
529  *	field.
530  */
531 
532 static int
533 vniocattach_file(vn, vio, dev, flag, td)
534 	struct vn_softc *vn;
535 	struct vn_ioctl *vio;
536 	dev_t dev;
537 	int flag;
538 	struct thread *td;
539 {
540 	struct vattr vattr;
541 	struct nameidata nd;
542 	int error, flags;
543 	struct proc *p = td->td_proc;
544 
545 	KKASSERT(p != NULL);
546 
547 	flags = FREAD|FWRITE;
548 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vn_file, td);
549 	error = vn_open(&nd, flags, 0);
550 	if (error) {
551 		if (error != EACCES && error != EPERM && error != EROFS)
552 			return (error);
553 		flags &= ~FWRITE;
554 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vn_file, td);
555 		error = vn_open(&nd, flags, 0);
556 		if (error)
557 			return (error);
558 	}
559 	NDFREE(&nd, NDF_ONLY_PNBUF);
560 	if (nd.ni_vp->v_type != VREG ||
561 	    (error = VOP_GETATTR(nd.ni_vp, &vattr, td))) {
562 		VOP_UNLOCK(nd.ni_vp, 0, td);
563 		(void) vn_close(nd.ni_vp, flags, td);
564 		return (error ? error : EINVAL);
565 	}
566 	VOP_UNLOCK(nd.ni_vp, 0, td);
567 	vn->sc_secsize = DEV_BSIZE;
568 	vn->sc_vp = nd.ni_vp;
569 
570 	/*
571 	 * If the size is specified, override the file attributes.  Note that
572 	 * the vn_size argument is in PAGE_SIZE sized blocks.
573 	 */
574 	if (vio->vn_size)
575 		vn->sc_size = (quad_t)vio->vn_size * PAGE_SIZE / vn->sc_secsize;
576 	else
577 		vn->sc_size = vattr.va_size / vn->sc_secsize;
578 	error = vnsetcred(vn, p->p_ucred);
579 	if (error) {
580 		(void) vn_close(nd.ni_vp, flags, td);
581 		return(error);
582 	}
583 	vn->sc_flags |= VNF_INITED;
584 	if (flags == FREAD)
585 		vn->sc_flags |= VNF_READONLY;
586 	IFOPT(vn, VN_LABELS) {
587 		/*
588 		 * Reopen so that `ds' knows which devices are open.
589 		 * If this is the first VNIOCSET, then we've
590 		 * guaranteed that the device is the cdev and that
591 		 * no other slices or labels are open.  Otherwise,
592 		 * we rely on VNIOCCLR not being abused.
593 		 */
594 		error = vnopen(dev, flag, S_IFCHR, td);
595 		if (error)
596 			vnclear(vn);
597 	}
598 	IFOPT(vn, VN_FOLLOW)
599 		printf("vnioctl: SET vp %p size %x blks\n",
600 		       vn->sc_vp, vn->sc_size);
601 	return(0);
602 }
603 
604 /*
605  *	vniocattach_swap:
606  *
607  *	Attach swap backing store to a VN partition of the size specified
608  *	in vn_size.
609  */
610 
611 static int
612 vniocattach_swap(vn, vio, dev, flag, td)
613 	struct vn_softc *vn;
614 	struct vn_ioctl *vio;
615 	dev_t dev;
616 	int flag;
617 	struct thread *td;
618 {
619 	int error;
620 	struct proc *p = td->td_proc;
621 
622 	KKASSERT(p != NULL);
623 	/*
624 	 * Range check.  Disallow negative sizes or any size less then the
625 	 * size of a page.  Then round to a page.
626 	 */
627 
628 	if (vio->vn_size <= 0)
629 		return(EDOM);
630 
631 	/*
632 	 * Allocate an OBJT_SWAP object.
633 	 *
634 	 * sc_secsize is PAGE_SIZE'd
635 	 *
636 	 * vio->vn_size is in PAGE_SIZE'd chunks.
637 	 * sc_size must be in PAGE_SIZE'd chunks.
638 	 * Note the truncation.
639 	 */
640 
641 	vn->sc_secsize = PAGE_SIZE;
642 	vn->sc_size = vio->vn_size;
643 	vn->sc_object =
644 	 vm_pager_allocate(OBJT_SWAP, NULL, vn->sc_secsize * (vm_ooffset_t)vio->vn_size, VM_PROT_DEFAULT, 0);
645 	IFOPT(vn, VN_RESERVE) {
646 		if (swap_pager_reserve(vn->sc_object, 0, vn->sc_size) < 0) {
647 			vm_pager_deallocate(vn->sc_object);
648 			vn->sc_object = NULL;
649 			return(EDOM);
650 		}
651 	}
652 	vn->sc_flags |= VNF_INITED;
653 
654 	error = vnsetcred(vn, p->p_ucred);
655 	if (error == 0) {
656 		IFOPT(vn, VN_LABELS) {
657 			/*
658 			 * Reopen so that `ds' knows which devices are open.
659 			 * If this is the first VNIOCSET, then we've
660 			 * guaranteed that the device is the cdev and that
661 			 * no other slices or labels are open.  Otherwise,
662 			 * we rely on VNIOCCLR not being abused.
663 			 */
664 			error = vnopen(dev, flag, S_IFCHR, td);
665 		}
666 	}
667 	if (error == 0) {
668 		IFOPT(vn, VN_FOLLOW) {
669 			printf("vnioctl: SET vp %p size %x\n",
670 			       vn->sc_vp, vn->sc_size);
671 		}
672 	}
673 	if (error)
674 		vnclear(vn);
675 	return(error);
676 }
677 
678 /*
679  * Duplicate the current processes' credentials.  Since we are called only
680  * as the result of a SET ioctl and only root can do that, any future access
681  * to this "disk" is essentially as root.  Note that credentials may change
682  * if some other uid can write directly to the mapped file (NFS).
683  */
684 int
685 vnsetcred(struct vn_softc *vn, struct ucred *cred)
686 {
687 	char *tmpbuf;
688 	int error = 0;
689 
690 	/*
691 	 * Set credits in our softc
692 	 */
693 
694 	if (vn->sc_cred)
695 		crfree(vn->sc_cred);
696 	vn->sc_cred = crdup(cred);
697 
698 	/*
699 	 * Horrible kludge to establish credentials for NFS  XXX.
700 	 */
701 
702 	if (vn->sc_vp) {
703 		struct uio auio;
704 		struct iovec aiov;
705 
706 		tmpbuf = malloc(vn->sc_secsize, M_TEMP, M_WAITOK);
707 		bzero(&auio, sizeof(auio));
708 
709 		aiov.iov_base = tmpbuf;
710 		aiov.iov_len = vn->sc_secsize;
711 		auio.uio_iov = &aiov;
712 		auio.uio_iovcnt = 1;
713 		auio.uio_offset = 0;
714 		auio.uio_rw = UIO_READ;
715 		auio.uio_segflg = UIO_SYSSPACE;
716 		auio.uio_resid = aiov.iov_len;
717 		vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY, curthread);
718 		error = VOP_READ(vn->sc_vp, &auio, 0, vn->sc_cred);
719 		VOP_UNLOCK(vn->sc_vp, 0, curthread);
720 		free(tmpbuf, M_TEMP);
721 	}
722 	return (error);
723 }
724 
725 void
726 vnclear(struct vn_softc *vn)
727 {
728 	struct thread *td = curthread;		/* XXX */
729 
730 	IFOPT(vn, VN_FOLLOW)
731 		printf("vnclear(%p): vp=%p\n", vn, vn->sc_vp);
732 	if (vn->sc_slices != NULL)
733 		dsgone(&vn->sc_slices);
734 	vn->sc_flags &= ~VNF_INITED;
735 	if (vn->sc_vp != NULL) {
736 		(void)vn_close(vn->sc_vp, vn->sc_flags & VNF_READONLY ?
737 		    FREAD : (FREAD|FWRITE), td);
738 		vn->sc_vp = NULL;
739 	}
740 	vn->sc_flags &= ~VNF_READONLY;
741 	if (vn->sc_cred) {
742 		crfree(vn->sc_cred);
743 		vn->sc_cred = NULL;
744 	}
745 	if (vn->sc_object != NULL) {
746 		vm_pager_deallocate(vn->sc_object);
747 		vn->sc_object = NULL;
748 	}
749 	vn->sc_size = 0;
750 }
751 
752 static	int
753 vnsize(dev_t dev)
754 {
755 	struct vn_softc *vn;
756 
757 	vn = dev->si_drv1;
758 	if (!vn)
759 		return(-1);
760 	if ((vn->sc_flags & VNF_INITED) == 0)
761 		return(-1);
762 
763 	return(vn->sc_size);
764 }
765 
766 static int
767 vn_modevent(module_t mod, int type, void *data)
768 {
769 	struct vn_softc *vn;
770 	dev_t dev;
771 
772 	switch (type) {
773 	case MOD_LOAD:
774 		cdevsw_add(&vn_cdevsw);
775 		break;
776 
777 	case MOD_UNLOAD:
778 		/* fall through */
779 	case MOD_SHUTDOWN:
780 		for (;;) {
781 			vn = SLIST_FIRST(&vn_list);
782 			if (!vn)
783 				break;
784 			SLIST_REMOVE_HEAD(&vn_list, sc_list);
785 			if (vn->sc_flags & VNF_INITED)
786 				vnclear(vn);
787 			/* Cleanup all dev_t's that refer to this unit */
788 			while ((dev = vn->sc_devlist) != NULL) {
789 				vn->sc_devlist = dev->si_drv2;
790 				dev->si_drv1 = dev->si_drv2 = NULL;
791 				/* If the last one, destroy it. */
792 				if (vn->sc_devlist == NULL)
793 					destroy_dev(dev);
794 			}
795 			free(vn, M_DEVBUF);
796 		}
797 		cdevsw_remove(&vn_cdevsw);
798 		break;
799 	default:
800 		break;
801 	}
802 	return 0;
803 }
804 
805 DEV_MODULE(vn, vn_modevent, 0);
806