xref: /dragonfly/sys/dev/disk/vn/vn.c (revision 2b3f93ea)
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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * from: Utah Hdr: vn.c 1.13 94/04/02
35  *
36  *	from: @(#)vn.c	8.6 (Berkeley) 4/1/94
37  * $FreeBSD: src/sys/dev/vn/vn.c,v 1.105.2.4 2001/11/18 07:11:00 dillon Exp $
38  */
39 
40 /*
41  * Vnode disk driver.
42  *
43  * Block/character interface to a vnode.  Allows one to treat a file
44  * as a disk (e.g. build a filesystem in it, mount it, etc.).
45  *
46  * NOTE 1: There is a security issue involved with this driver.
47  * Once mounted all access to the contents of the "mapped" file via
48  * the special file is controlled by the permissions on the special
49  * file, the protection of the mapped file is ignored (effectively,
50  * by using root credentials in all transactions).
51  *
52  * NOTE 2: Doesn't interact with leases, should it?
53  */
54 
55 #include "use_vn.h"
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/uio.h>
59 #include <sys/kernel.h>
60 #include <sys/proc.h>
61 #include <sys/caps.h>
62 #include <sys/nlookup.h>
63 #include <sys/buf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
68 #include <sys/conf.h>
69 #include <sys/diskslice.h>
70 #include <sys/disk.h>
71 #include <sys/stat.h>
72 #include <sys/module.h>
73 #include <sys/vnioctl.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_pager.h>
79 #include <vm/vm_pageout.h>
80 #include <vm/swap_pager.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_zone.h>
83 #include <sys/devfs.h>
84 
85 static	d_ioctl_t	vnioctl;
86 static	d_open_t	vnopen;
87 static	d_close_t	vnclose;
88 static	d_psize_t	vnsize;
89 static	d_strategy_t	vnstrategy;
90 static	d_clone_t	vnclone;
91 
92 MALLOC_DEFINE(M_VN, "vn_softc", "vn driver structures");
93 DEVFS_DEFINE_CLONE_BITMAP(vn);
94 
95 #if NVN <= 1
96 #define VN_PREALLOCATED_UNITS	4
97 #else
98 #define VN_PREALLOCATED_UNITS	NVN
99 #endif
100 
101 #define VN_BSIZE_BEST	8192
102 
103 /*
104  * dev_ops
105  *	D_DISK		We want to look like a disk
106  *	D_CANFREE	We support BUF_CMD_FREEBLKS
107  *	D_NOEMERGPGR	Too complex for emergency pager
108  */
109 
110 static struct dev_ops vn_ops = {
111 	{ "vn", 0, D_DISK | D_CANFREE | D_NOEMERGPGR },
112 	.d_open =	vnopen,
113 	.d_close =	vnclose,
114 	.d_read =	physread,
115 	.d_write =	physwrite,
116 	.d_ioctl =	vnioctl,
117 	.d_strategy =	vnstrategy,
118 	.d_psize =	vnsize
119 };
120 
121 struct vn_softc {
122 	int		sc_unit;
123 	int		sc_flags;	/* flags 			*/
124 	u_int64_t	sc_size;	/* size of vn, sc_secsize scale	*/
125 	int		sc_secsize;	/* sector size			*/
126 	struct disk	sc_disk;
127 	struct vnode	*sc_vp;		/* vnode if not NULL		*/
128 	vm_object_t	sc_object;	/* backing object if not NULL	*/
129 	struct ucred	*sc_cred;	/* credentials 			*/
130 	int		 sc_maxactive;	/* max # of active requests 	*/
131 	struct buf	 sc_tab;	/* transfer queue 		*/
132 	u_long		 sc_options;	/* options 			*/
133 	cdev_t		 sc_dev;	/* devices that refer to this unit */
134 	SLIST_ENTRY(vn_softc) sc_list;
135 };
136 
137 static SLIST_HEAD(, vn_softc) vn_list;
138 
139 /* sc_flags */
140 #define VNF_INITED	0x01
141 #define	VNF_READONLY	0x02
142 #define VNF_OPENED	0x10
143 #define	VNF_DESTROY	0x20
144 
145 static u_long	vn_options;
146 
147 #define IFOPT(vn,opt) if (((vn)->sc_options|vn_options) & (opt))
148 #define TESTOPT(vn,opt) (((vn)->sc_options|vn_options) & (opt))
149 
150 static int	vnsetcred (struct vn_softc *vn, struct ucred *cred);
151 static void	vnclear (struct vn_softc *vn);
152 static int	vnget (cdev_t dev, struct vn_softc *vn , struct vn_user *vnu);
153 static int	vn_modevent (module_t, int, void *);
154 static int 	vniocattach_file (struct vn_softc *, struct vn_ioctl *, cdev_t dev, int flag, struct ucred *cred);
155 static int 	vniocattach_swap (struct vn_softc *, struct vn_ioctl *, cdev_t dev, int flag, struct ucred *cred);
156 static cdev_t	vn_create(int unit, int clone);
157 
158 static int
vnclone(struct dev_clone_args * ap)159 vnclone(struct dev_clone_args *ap)
160 {
161 	int unit;
162 
163 	unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(vn), 0);
164 	ap->a_dev = vn_create(unit, 1);
165 
166 	return 0;
167 }
168 
169 static	int
vnclose(struct dev_close_args * ap)170 vnclose(struct dev_close_args *ap)
171 {
172 	cdev_t dev = ap->a_head.a_dev;
173 	struct vn_softc *vn;
174 
175 	vn = dev->si_drv1;
176 	KKASSERT(vn != NULL);
177 
178 	vn->sc_flags &= ~VNF_OPENED;
179 
180 	/* The disk has been detached and can now be safely destroyed */
181 	if (vn->sc_flags & VNF_DESTROY) {
182 		KKASSERT(disk_getopencount(&vn->sc_disk) == 0);
183 		disk_destroy(&vn->sc_disk);
184 		devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(vn), dkunit(dev));
185 		SLIST_REMOVE(&vn_list, vn, vn_softc, sc_list);
186 		kfree(vn, M_VN);
187 	}
188 	return (0);
189 }
190 
191 static struct vn_softc *
vncreatevn(void)192 vncreatevn(void)
193 {
194 	struct vn_softc *vn;
195 
196 	vn = kmalloc(sizeof *vn, M_VN, M_WAITOK | M_ZERO);
197 	return vn;
198 }
199 
200 static void
vninitvn(struct vn_softc * vn,cdev_t dev)201 vninitvn(struct vn_softc *vn, cdev_t dev)
202 {
203 	int unit;
204 
205 	KKASSERT(vn != NULL);
206 	KKASSERT(dev != NULL);
207 	unit = dkunit(dev);
208 
209 	vn->sc_unit = unit;
210 	dev->si_drv1 = vn;
211 	vn->sc_dev = dev;
212 
213 	SLIST_INSERT_HEAD(&vn_list, vn, sc_list);
214 }
215 
216 static	int
vnopen(struct dev_open_args * ap)217 vnopen(struct dev_open_args *ap)
218 {
219 	cdev_t dev = ap->a_head.a_dev;
220 	struct vn_softc *vn;
221 
222 	/*
223 	 * Locate preexisting device
224 	 */
225 
226 	vn = dev->si_drv1;
227 	KKASSERT(vn != NULL);
228 
229 	/*
230 	 * Update si_bsize fields for device.  This data will be overridden by
231 	 * the slice/parition code for vn accesses through partitions, and
232 	 * used directly if you open the 'whole disk' device.
233 	 *
234 	 * si_bsize_best must be reinitialized in case VN has been
235 	 * reconfigured, plus make it at least VN_BSIZE_BEST for efficiency.
236 	 */
237 	dev->si_bsize_phys = vn->sc_secsize;
238 	dev->si_bsize_best = vn->sc_secsize;
239 	if (dev->si_bsize_best < VN_BSIZE_BEST)
240 		dev->si_bsize_best = VN_BSIZE_BEST;
241 
242 	if ((ap->a_oflags & FWRITE) && (vn->sc_flags & VNF_READONLY))
243 		return (EACCES);
244 
245 	IFOPT(vn, VN_FOLLOW)
246 		kprintf("vnopen(%s, 0x%x, 0x%x)\n",
247 		    devtoname(dev), ap->a_oflags, ap->a_devtype);
248 
249 	vn->sc_flags |= VNF_OPENED;
250 	return(0);
251 }
252 
253 /*
254  *	vnstrategy:
255  *
256  *	Run strategy routine for VN device.  We use VOP_READ/VOP_WRITE calls
257  *	for vnode-backed vn's, and the swap_pager_strategy() call for
258  *	vm_object-backed vn's.
259  */
260 static int
vnstrategy(struct dev_strategy_args * ap)261 vnstrategy(struct dev_strategy_args *ap)
262 {
263 	cdev_t dev = ap->a_head.a_dev;
264 	struct bio *bio = ap->a_bio;
265 	struct buf *bp;
266 	struct bio *nbio;
267 	int unit;
268 	struct vn_softc *vn;
269 	int error;
270 
271 	unit = dkunit(dev);
272 	vn = dev->si_drv1;
273 	KKASSERT(vn != NULL);
274 
275 	bp = bio->bio_buf;
276 
277 	IFOPT(vn, VN_DEBUG)
278 		kprintf("vnstrategy(%p): unit %d\n", bp, unit);
279 
280 	if ((vn->sc_flags & VNF_INITED) == 0) {
281 		bp->b_error = ENXIO;
282 		bp->b_flags |= B_ERROR;
283 		biodone(bio);
284 		return(0);
285 	}
286 
287 	bp->b_resid = bp->b_bcount;
288 
289 	/*
290 	 * The vnode device is using disk/slice label support.
291 	 *
292 	 * The dscheck() function is called for validating the
293 	 * slices that exist ON the vnode device itself, and
294 	 * translate the "slice-relative" block number, again.
295 	 * dscheck() will call biodone() and return NULL if
296 	 * we are at EOF or beyond the device size.
297 	 */
298 
299 	nbio = bio;
300 
301 	/*
302 	 * Use the translated nbio from this point on
303 	 */
304 	if (vn->sc_vp && bp->b_cmd == BUF_CMD_FREEBLKS) {
305 		/*
306 		 * Freeblks is not handled for vnode-backed elements yet.
307 		 */
308 		bp->b_resid = 0;
309 		/* operation complete */
310 	} else if (vn->sc_vp) {
311 		/*
312 		 * VNODE I/O
313 		 *
314 		 * If an error occurs, we set B_ERROR but we do not set
315 		 * B_INVAL because (for a write anyway), the buffer is
316 		 * still valid.
317 		 */
318 		struct uio auio;
319 		struct iovec aiov;
320 
321 		bzero(&auio, sizeof(auio));
322 
323 		aiov.iov_base = bp->b_data;
324 		aiov.iov_len = bp->b_bcount;
325 		auio.uio_iov = &aiov;
326 		auio.uio_iovcnt = 1;
327 		auio.uio_offset = nbio->bio_offset;
328 		auio.uio_segflg = UIO_SYSSPACE;
329 		if (bp->b_cmd == BUF_CMD_READ)
330 			auio.uio_rw = UIO_READ;
331 		else
332 			auio.uio_rw = UIO_WRITE;
333 		auio.uio_resid = bp->b_bcount;
334 		auio.uio_td = curthread;
335 
336 		/*
337 		 * Don't use IO_DIRECT here, it really gets in the way
338 		 * due to typical blocksize differences between the
339 		 * fs backing the VN device and whatever is running on
340 		 * the VN device.
341 		 */
342 		switch (bp->b_cmd) {
343 		case (BUF_CMD_READ):
344 			vn_lock(vn->sc_vp, LK_SHARED | LK_RETRY);
345 			error = VOP_READ(vn->sc_vp, &auio, IO_RECURSE,
346 					 vn->sc_cred);
347 			break;
348 
349 		case (BUF_CMD_WRITE):
350 			vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
351 			error = VOP_WRITE(vn->sc_vp, &auio, IO_RECURSE,
352 					  vn->sc_cred);
353 			break;
354 
355 		case (BUF_CMD_FLUSH):
356 			auio.uio_resid = 0;
357 			vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
358 			error = VOP_FSYNC(vn->sc_vp, MNT_WAIT, 0);
359 			break;
360 		default:
361 			auio.uio_resid = 0;
362 			error = 0;
363 			goto breakunlocked;
364 		}
365 		vn_unlock(vn->sc_vp);
366 breakunlocked:
367 		bp->b_resid = auio.uio_resid;
368 		if (error) {
369 			bp->b_error = error;
370 			bp->b_flags |= B_ERROR;
371 		}
372 		/* operation complete */
373 	} else if (vn->sc_object) {
374 		/*
375 		 * OBJT_SWAP I/O (handles read, write, freebuf)
376 		 *
377 		 * We have nothing to do if freeing  blocks on a reserved
378 		 * swap area, othrewise execute the op.
379 		 */
380 		if (bp->b_cmd == BUF_CMD_FREEBLKS && TESTOPT(vn, VN_RESERVE)) {
381 			bp->b_resid = 0;
382 			/* operation complete */
383 		} else {
384 			swap_pager_strategy(vn->sc_object, nbio);
385 			return(0);
386 			/* NOT REACHED */
387 		}
388 	} else {
389 		bp->b_resid = bp->b_bcount;
390 		bp->b_flags |= B_ERROR | B_INVAL;
391 		bp->b_error = EINVAL;
392 		/* operation complete */
393 	}
394 	biodone(nbio);
395 	return(0);
396 }
397 
398 /* ARGSUSED */
399 static	int
vnioctl(struct dev_ioctl_args * ap)400 vnioctl(struct dev_ioctl_args *ap)
401 {
402 	cdev_t dev = ap->a_head.a_dev;
403 	struct vn_softc *vn;
404 	struct vn_ioctl *vio;
405 	int error;
406 	u_long *f;
407 
408 	vn = dev->si_drv1;
409 	IFOPT(vn,VN_FOLLOW) {
410 		kprintf("vnioctl(%s, 0x%lx, %p, 0x%x): unit %d\n",
411 		    devtoname(dev), ap->a_cmd, ap->a_data, ap->a_fflag,
412 		    dkunit(dev));
413 	}
414 
415 	switch (ap->a_cmd) {
416 	case VNIOCATTACH:
417 	case VNIOCDETACH:
418 	case VNIOCGSET:
419 	case VNIOCGCLEAR:
420 	case VNIOCGET:
421 	case VNIOCUSET:
422 	case VNIOCUCLEAR:
423 		goto vn_specific;
424 	}
425 
426 #if 0
427 	if (dkslice(dev) != WHOLE_DISK_SLICE ||
428 		dkpart(dev) != WHOLE_SLICE_PART)
429 		return (ENOTTY);
430 #endif
431 
432     vn_specific:
433 
434 	error = caps_priv_check(ap->a_cred, SYSCAP_RESTRICTEDROOT);
435 	if (error)
436 		return (error);
437 
438 	vio = (struct vn_ioctl *)ap->a_data;
439 	f = (u_long*)ap->a_data;
440 
441 	switch (ap->a_cmd) {
442 	case VNIOCATTACH:
443 		if (vn->sc_flags & VNF_INITED)
444 			return(EBUSY);
445 
446 		if (vn->sc_flags & VNF_DESTROY)
447 			return(ENXIO);
448 
449 		if (vio->vn_file == NULL)
450 			error = vniocattach_swap(vn, vio, dev, ap->a_fflag, ap->a_cred);
451 		else
452 			error = vniocattach_file(vn, vio, dev, ap->a_fflag, ap->a_cred);
453 		break;
454 
455 	case VNIOCDETACH:
456 		if ((vn->sc_flags & VNF_INITED) == 0)
457 			return(ENXIO);
458 		/*
459 		 * XXX handle i/o in progress.  Return EBUSY, or wait, or
460 		 * flush the i/o.
461 		 * XXX handle multiple opens of the device.  Return EBUSY,
462 		 * or revoke the fd's.
463 		 * How are these problems handled for removable and failing
464 		 * hardware devices? (Hint: They are not)
465 		 */
466 		if ((disk_getopencount(&vn->sc_disk)) > 1)
467 			return (EBUSY);
468 
469 		vnclear(vn);
470 		IFOPT(vn, VN_FOLLOW)
471 			kprintf("vnioctl: CLRed\n");
472 
473 		if (dkunit(dev) >= VN_PREALLOCATED_UNITS) {
474 			vn->sc_flags |= VNF_DESTROY;
475 		}
476 
477 		break;
478 
479 	case VNIOCGET:
480 		error = vnget(dev, vn, (struct vn_user *) ap->a_data);
481 		break;
482 
483 	case VNIOCGSET:
484 		vn_options |= *f;
485 		*f = vn_options;
486 		break;
487 
488 	case VNIOCGCLEAR:
489 		vn_options &= ~(*f);
490 		*f = vn_options;
491 		break;
492 
493 	case VNIOCUSET:
494 		vn->sc_options |= *f;
495 		*f = vn->sc_options;
496 		break;
497 
498 	case VNIOCUCLEAR:
499 		vn->sc_options &= ~(*f);
500 		*f = vn->sc_options;
501 		break;
502 
503 	default:
504 		error = ENOTTY;
505 		break;
506 	}
507 	return(error);
508 }
509 
510 /*
511  *	vniocattach_file:
512  *
513  *	Attach a file to a VN partition.  Return the size in the vn_size
514  *	field.
515  */
516 
517 static int
vniocattach_file(struct vn_softc * vn,struct vn_ioctl * vio,cdev_t dev,int flag,struct ucred * cred)518 vniocattach_file(struct vn_softc *vn, struct vn_ioctl *vio, cdev_t dev,
519 		 int flag, struct ucred *cred)
520 {
521 	struct vattr vattr;
522 	struct nlookupdata nd;
523 	int error, flags;
524 	struct vnode *vp;
525 	struct disk_info info;
526 
527 	flags = FREAD|FWRITE;
528 	error = nlookup_init(&nd, vio->vn_file,
529 				UIO_USERSPACE, NLC_FOLLOW|NLC_LOCKVP);
530 	if (error)
531 		return (error);
532 	if ((error = vn_open(&nd, NULL, flags, 0)) != 0) {
533 		if (error != EACCES && error != EPERM && error != EROFS)
534 			goto done;
535 		flags &= ~FWRITE;
536 		nlookup_done(&nd);
537 		error = nlookup_init(&nd, vio->vn_file, UIO_USERSPACE, NLC_FOLLOW|NLC_LOCKVP);
538 		if (error)
539 			return (error);
540 		if ((error = vn_open(&nd, NULL, flags, 0)) != 0)
541 			goto done;
542 	}
543 	vp = nd.nl_open_vp;
544 	if (vp->v_type != VREG ||
545 	    (error = VOP_GETATTR(vp, &vattr))) {
546 		if (error == 0)
547 			error = EINVAL;
548 		goto done;
549 	}
550 	vn_unlock(vp);
551 	vn->sc_secsize = DEV_BSIZE;
552 	vn->sc_vp = vp;
553 	nd.nl_open_vp = NULL;
554 
555 	/*
556 	 * If the size is specified, override the file attributes.  Note that
557 	 * the vn_size argument is in PAGE_SIZE sized blocks.
558 	 */
559 	if (vio->vn_size)
560 		vn->sc_size = vio->vn_size * PAGE_SIZE / vn->sc_secsize;
561 	else
562 		vn->sc_size = vattr.va_size / vn->sc_secsize;
563 	error = vnsetcred(vn, cred);
564 	if (error) {
565 		vn->sc_vp = NULL;
566 		vn_close(vp, flags, NULL);
567 		goto done;
568 	}
569 	vn->sc_flags |= VNF_INITED;
570 	if (flags == FREAD)
571 		vn->sc_flags |= VNF_READONLY;
572 
573 	/*
574 	 * Set the disk info so that probing is triggered
575 	 */
576 	bzero(&info, sizeof(struct disk_info));
577 	info.d_media_blksize = vn->sc_secsize;
578 	info.d_media_blocks = vn->sc_size;
579 	/*
580 	 * reserve mbr sector for backwards compatibility
581 	 * when no slices exist.
582 	 */
583 	info.d_dsflags = DSO_COMPATMBR | DSO_RAWPSIZE;
584 	info.d_secpertrack = 32;
585 	info.d_nheads = 64 / (vn->sc_secsize / DEV_BSIZE);
586 	info.d_secpercyl = info.d_secpertrack * info.d_nheads;
587 	info.d_ncylinders = vn->sc_size / info.d_secpercyl;
588 	disk_setdiskinfo_sync(&vn->sc_disk, &info);
589 
590 	error = dev_dopen(dev, flag, S_IFCHR, cred, NULL, NULL);
591 	if (error)
592 		vnclear(vn);
593 
594 	IFOPT(vn, VN_FOLLOW)
595 		kprintf("vnioctl: SET vp %p size %llx blks\n",
596 		       vn->sc_vp, (long long)vn->sc_size);
597 done:
598 	nlookup_done(&nd);
599 	return(error);
600 }
601 
602 /*
603  *	vniocattach_swap:
604  *
605  *	Attach swap backing store to a VN partition of the size specified
606  *	in vn_size.
607  */
608 
609 static int
vniocattach_swap(struct vn_softc * vn,struct vn_ioctl * vio,cdev_t dev,int flag,struct ucred * cred)610 vniocattach_swap(struct vn_softc *vn, struct vn_ioctl *vio, cdev_t dev,
611 		 int flag, struct ucred *cred)
612 {
613 	int error;
614 	struct disk_info info;
615 
616 	/*
617 	 * Range check.  Disallow negative sizes or any size less then the
618 	 * size of a page.  Then round to a page.
619 	 */
620 
621 	if (vio->vn_size <= 0)
622 		return(EDOM);
623 
624 	/*
625 	 * Allocate an OBJT_SWAP object.
626 	 *
627 	 * sc_secsize is PAGE_SIZE'd
628 	 *
629 	 * vio->vn_size is in PAGE_SIZE'd chunks.
630 	 * sc_size must be in PAGE_SIZE'd chunks.
631 	 * Note the truncation.
632 	 */
633 
634 	vn->sc_secsize = PAGE_SIZE;
635 	vn->sc_size = vio->vn_size;
636 	vn->sc_object = swap_pager_alloc(NULL,
637 					 vn->sc_secsize * (off_t)vio->vn_size,
638 					 VM_PROT_DEFAULT, 0);
639 	vm_object_set_flag(vn->sc_object, OBJ_NOPAGEIN);
640 	IFOPT(vn, VN_RESERVE) {
641 		if (swap_pager_reserve(vn->sc_object, 0, vn->sc_size) < 0) {
642 			vm_pager_deallocate(vn->sc_object);
643 			vn->sc_object = NULL;
644 			return(EDOM);
645 		}
646 	}
647 	vn->sc_flags |= VNF_INITED;
648 
649 	error = vnsetcred(vn, cred);
650 	if (error == 0) {
651 		/*
652 		 * Set the disk info so that probing is triggered
653 		 */
654 		bzero(&info, sizeof(struct disk_info));
655 		info.d_media_blksize = vn->sc_secsize;
656 		info.d_media_blocks = vn->sc_size;
657 		/*
658 		 * reserve mbr sector for backwards compatibility
659 		 * when no slices exist.
660 		 */
661 		info.d_dsflags = DSO_COMPATMBR | DSO_RAWPSIZE;
662 		info.d_secpertrack = 32;
663 		info.d_nheads = 64 / (vn->sc_secsize / DEV_BSIZE);
664 		info.d_secpercyl = info.d_secpertrack * info.d_nheads;
665 		info.d_ncylinders = vn->sc_size / info.d_secpercyl;
666 		disk_setdiskinfo_sync(&vn->sc_disk, &info);
667 
668 		error = dev_dopen(dev, flag, S_IFCHR, cred, NULL, NULL);
669 	}
670 	if (error == 0) {
671 		IFOPT(vn, VN_FOLLOW) {
672 			kprintf("vnioctl: SET vp %p size %llx\n",
673 			       vn->sc_vp, (long long)vn->sc_size);
674 		}
675 	}
676 	if (error)
677 		vnclear(vn);
678 	return(error);
679 }
680 
681 /*
682  * Duplicate the current processes' credentials.  Since we are called only
683  * as the result of a SET ioctl and only root can do that, any future access
684  * to this "disk" is essentially as root.  Note that credentials may change
685  * if some other uid can write directly to the mapped file (NFS).
686  */
687 static int
vnsetcred(struct vn_softc * vn,struct ucred * cred)688 vnsetcred(struct vn_softc *vn, struct ucred *cred)
689 {
690 	char *tmpbuf;
691 	int error = 0;
692 
693 	/*
694 	 * Set credits in our softc
695 	 */
696 
697 	if (vn->sc_cred)
698 		crfree(vn->sc_cred);
699 	vn->sc_cred = crdup(cred);
700 
701 	/*
702 	 * Horrible kludge to establish credentials for NFS  XXX.
703 	 */
704 
705 	if (vn->sc_vp) {
706 		struct uio auio;
707 		struct iovec aiov;
708 
709 		tmpbuf = kmalloc(vn->sc_secsize, M_TEMP, M_WAITOK);
710 		bzero(&auio, sizeof(auio));
711 
712 		aiov.iov_base = tmpbuf;
713 		aiov.iov_len = vn->sc_secsize;
714 		auio.uio_iov = &aiov;
715 		auio.uio_iovcnt = 1;
716 		auio.uio_offset = 0;
717 		auio.uio_rw = UIO_READ;
718 		auio.uio_segflg = UIO_SYSSPACE;
719 		auio.uio_resid = aiov.iov_len;
720 		vn_lock(vn->sc_vp, LK_EXCLUSIVE | LK_RETRY);
721 		error = VOP_READ(vn->sc_vp, &auio, 0, vn->sc_cred);
722 		vn_unlock(vn->sc_vp);
723 		kfree(tmpbuf, M_TEMP);
724 	}
725 	return (error);
726 }
727 
728 static void
vnclear(struct vn_softc * vn)729 vnclear(struct vn_softc *vn)
730 {
731 	IFOPT(vn, VN_FOLLOW)
732 		kprintf("vnclear(%p): vp=%p\n", vn, vn->sc_vp);
733 	vn->sc_flags &= ~VNF_INITED;
734 	if (vn->sc_vp != NULL) {
735 		vn_close(vn->sc_vp,
736 		    (vn->sc_flags & VNF_READONLY) ? FREAD : (FREAD|FWRITE),
737 		    NULL);
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 
750 	disk_unprobe(&vn->sc_disk);
751 
752 	vn->sc_size = 0;
753 }
754 
755 /*
756  * 	vnget:
757  *
758  *	populate a struct vn_user for the VNIOCGET ioctl.
759  *	interface conventions defined in sys/sys/vnioctl.h.
760  */
761 
762 static int
vnget(cdev_t dev,struct vn_softc * vn,struct vn_user * vnu)763 vnget(cdev_t dev, struct vn_softc *vn, struct vn_user *vnu)
764 {
765 	int error, found = 0;
766 	char *freepath, *fullpath;
767 	struct vattr vattr;
768 
769 	if (vnu->vnu_unit == -1) {
770 		vnu->vnu_unit = dkunit(dev);
771 	}
772 	else if (vnu->vnu_unit < 0)
773 		return (EINVAL);
774 
775 	SLIST_FOREACH(vn, &vn_list, sc_list) {
776 
777 		if(vn->sc_unit != vnu->vnu_unit)
778 			continue;
779 
780 		found = 1;
781 
782 		if (vn->sc_flags & VNF_INITED && vn->sc_vp != NULL) {
783 
784 			/* note: u_cred checked in vnioctl above */
785 			error = VOP_GETATTR(vn->sc_vp, &vattr);
786 			if (error) {
787 				kprintf("vnget: VOP_GETATTR for %p failed\n",
788 					vn->sc_vp);
789 				return (error);
790 			}
791 
792 			error = vn_fullpath(curproc, vn->sc_vp,
793 						&fullpath, &freepath, 0);
794 
795 			if (error) {
796 				kprintf("vnget: unable to resolve vp %p\n",
797 					vn->sc_vp);
798 				return(error);
799 			}
800 
801 			strlcpy(vnu->vnu_file, fullpath,
802 				sizeof(vnu->vnu_file));
803 			kfree(freepath, M_TEMP);
804 			vnu->vnu_dev = vattr.va_fsid;
805 			vnu->vnu_ino = vattr.va_fileid;
806 
807 		}
808 		else if (vn->sc_flags & VNF_INITED && vn->sc_object != NULL){
809 
810 			strlcpy(vnu->vnu_file, _VN_USER_SWAP,
811 				sizeof(vnu->vnu_file));
812 			vnu->vnu_size = vn->sc_size;
813 			vnu->vnu_secsize = vn->sc_secsize;
814 
815 		} else {
816 
817 			bzero(vnu->vnu_file, sizeof(vnu->vnu_file));
818 			vnu->vnu_dev = 0;
819 			vnu->vnu_ino = 0;
820 
821 		}
822 		break;
823 	}
824 
825 	if (!found)
826 		return(ENXIO);
827 
828 	return(0);
829 }
830 
831 static int
vnsize(struct dev_psize_args * ap)832 vnsize(struct dev_psize_args *ap)
833 {
834 	cdev_t dev = ap->a_head.a_dev;
835 	struct vn_softc *vn;
836 
837 	vn = dev->si_drv1;
838 	if (!vn)
839 		return(ENXIO);
840 	if ((vn->sc_flags & VNF_INITED) == 0)
841 		return(ENXIO);
842 	ap->a_result = (int64_t)vn->sc_size;
843 	return(0);
844 }
845 
846 /*
847  * Returns NULL only if the specified unit cannot be allocated.
848  */
849 static cdev_t
vn_create(int unit,int clone)850 vn_create(int unit, int clone)
851 {
852 	struct vn_softc *vn;
853 	struct disk_info info;
854 	cdev_t dev, ret_dev;
855 
856 	vn = vncreatevn();
857 	if (clone) {
858 		/*
859 		 * For clone devices we need to return the top-level cdev,
860 		 * not the raw dev we'd normally work with.
861 		 */
862 		dev = disk_create_clone(unit, &vn->sc_disk, &vn_ops);
863 		ret_dev = vn->sc_disk.d_cdev;
864 	} else {
865 		ret_dev = dev = disk_create(unit, &vn->sc_disk, &vn_ops);
866 	}
867 	vninitvn(vn, dev);
868 
869 	bzero(&info, sizeof(struct disk_info));
870 	info.d_media_blksize = 512;
871 	info.d_media_blocks = 0;
872 	info.d_dsflags = DSO_MBRQUIET | DSO_RAWPSIZE;
873 	info.d_secpertrack = 32;
874 	info.d_nheads = 64;
875 	info.d_secpercyl = info.d_secpertrack * info.d_nheads;
876 	info.d_ncylinders = 0;
877 	disk_setdiskinfo_sync(&vn->sc_disk, &info);
878 
879 	return ret_dev;
880 }
881 
882 static int
vn_modevent(module_t mod,int type,void * data)883 vn_modevent(module_t mod, int type, void *data)
884 {
885 	struct vn_softc *vn;
886 	static cdev_t dev = NULL;
887 	int i;
888 
889 	switch (type) {
890 	case MOD_LOAD:
891 		dev = make_autoclone_dev(&vn_ops, &DEVFS_CLONE_BITMAP(vn),
892 					 vnclone, UID_ROOT, GID_OPERATOR,
893 					 0640, "vn");
894 		for (i = 0; i < VN_PREALLOCATED_UNITS; i++) {
895 			devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(vn), i);
896 			vn_create(i, 0);
897 		}
898 		break;
899 
900 	case MOD_UNLOAD:
901 	case MOD_SHUTDOWN:
902 		while ((vn = SLIST_FIRST(&vn_list)) != NULL) {
903 			/*
904 			 * XXX: no idea if we can return EBUSY even in the
905 			 *	shutdown case, so err on the side of caution
906 			 *	and just rip stuff out on shutdown.
907 			 */
908 			if (type != MOD_SHUTDOWN) {
909 				if (vn->sc_flags & VNF_OPENED)
910 					return (EBUSY);
911 			}
912 
913 			disk_destroy(&vn->sc_disk);
914 
915 			SLIST_REMOVE_HEAD(&vn_list, sc_list);
916 
917 			if (vn->sc_flags & VNF_INITED)
918 				vnclear(vn);
919 
920 			kfree(vn, M_VN);
921 		}
922 		destroy_autoclone_dev(dev, &DEVFS_CLONE_BITMAP(vn));
923 		dev_ops_remove_all(&vn_ops);
924 		break;
925 	default:
926 		break;
927 	}
928 	return 0;
929 }
930 
931 DEV_MODULE(vn, vn_modevent, 0);
932