xref: /openbsd-src/sys/dev/vnd.c (revision a28daedfc357b214be5c701aa8ba8adb29a7f1c2)
1 /*	$OpenBSD: vnd.c,v 1.90 2008/09/03 23:24:25 krw Exp $	*/
2 /*	$NetBSD: vnd.c,v 1.26 1996/03/30 23:06:11 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1988 University of Utah.
6  * Copyright (c) 1990, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * the Systems Programming Group of the University of Utah Computer
11  * Science Department.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  * from: Utah $Hdr: vn.c 1.13 94/04/02$
38  *
39  *	@(#)vn.c	8.6 (Berkeley) 4/1/94
40  */
41 
42 /*
43  * Vnode disk driver.
44  *
45  * Block/character interface to a vnode.  Allows one to treat a file
46  * as a disk (e.g. build a filesystem in it, mount it, etc.).
47  *
48  * NOTE 1: This uses either the VOP_BMAP/VOP_STRATEGY interface to the
49  * vnode or simple VOP_READ/VOP_WRITE.  The former is suitable for swapping
50  * as it doesn't distort the local buffer cache.  The latter is good for
51  * building disk images as it keeps the cache consistent after the block
52  * device is closed.
53  *
54  * NOTE 2: There is a security issue involved with this driver.
55  * Once mounted all access to the contents of the "mapped" file via
56  * the special file is controlled by the permissions on the special
57  * file, the protection of the mapped file is ignored (effectively,
58  * by using root credentials in all transactions).
59  *
60  * NOTE 3: Doesn't interact with leases, should it?
61  */
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/namei.h>
66 #include <sys/proc.h>
67 #include <sys/errno.h>
68 #include <sys/buf.h>
69 #include <sys/malloc.h>
70 #include <sys/pool.h>
71 #include <sys/ioctl.h>
72 #include <sys/disklabel.h>
73 #include <sys/device.h>
74 #include <sys/disk.h>
75 #include <sys/stat.h>
76 #include <sys/mount.h>
77 #include <sys/vnode.h>
78 #include <sys/file.h>
79 #include <sys/rwlock.h>
80 #include <sys/uio.h>
81 #include <sys/conf.h>
82 
83 #include <crypto/blf.h>
84 
85 #include <miscfs/specfs/specdev.h>
86 
87 #include <dev/vndioctl.h>
88 
89 #ifdef VNDDEBUG
90 int dovndcluster = 1;
91 int vnddebug = 0x00;
92 #define	VDB_FOLLOW	0x01
93 #define	VDB_INIT	0x02
94 #define	VDB_IO		0x04
95 #define	DNPRINTF(f, p...)	do { if ((f) & vnddebug) printf(p); } while (0)
96 #else
97 #define	DNPRINTF(f, p...)	/* nothing */
98 #endif	/* VNDDEBUG */
99 
100 /*
101  * vndunit is a bit weird.  have to reconstitute the dev_t for
102  * DISKUNIT(), but with the minor masked off.
103  */
104 #define	vndunit(x)	DISKUNIT(makedev(major(x), minor(x) & 0x7ff))
105 #define	vndsimple(x)	(minor(x) & 0x800)
106 
107 /* same as MAKEDISKDEV, preserving the vndsimple() property */
108 #define	VNDLABELDEV(dev)	\
109 	makedev(major(dev), DISKMINOR(vndunit(dev), RAW_PART) | \
110 	    (vndsimple(dev) ? 0x800 : 0))
111 
112 struct vndbuf {
113 	struct buf	vb_buf;
114 	struct buf	*vb_obp;
115 };
116 
117 /*
118  * struct vndbuf allocator
119  */
120 struct pool     vndbufpl;
121 
122 #define	getvndbuf()	pool_get(&vndbufpl, PR_WAITOK)
123 #define	putvndbuf(vbp)	pool_put(&vndbufpl, vbp);
124 
125 struct vnd_softc {
126 	struct device	 sc_dev;
127 	struct disk	 sc_dk;
128 
129 	char		 sc_file[VNDNLEN];	/* file we're covering */
130 	int		 sc_flags;		/* flags */
131 	size_t		 sc_size;		/* size of vnd in sectors */
132 	size_t		 sc_secsize;		/* sector size in bytes */
133 	size_t		 sc_nsectors;		/* # of sectors per track */
134 	size_t		 sc_ntracks;		/* # of tracks per cylinder */
135 	struct vnode	*sc_vp;			/* vnode */
136 	struct ucred	*sc_cred;		/* credentials */
137 	struct buf	 sc_tab;		/* transfer queue */
138 	blf_ctx		*sc_keyctx;		/* key context */
139 	struct rwlock	 sc_rwlock;
140 };
141 
142 /* sc_flags */
143 #define	VNF_ALIVE	0x0001
144 #define	VNF_INITED	0x0002
145 #define	VNF_LABELLING	0x0100
146 #define	VNF_WLABEL	0x0200
147 #define	VNF_HAVELABEL	0x0400
148 #define	VNF_SIMPLE	0x1000
149 #define	VNF_READONLY	0x2000
150 
151 #define	VNDRW(v)	((v)->sc_flags & VNF_READONLY ? FREAD : FREAD|FWRITE)
152 
153 struct vnd_softc *vnd_softc;
154 int numvnd = 0;
155 
156 struct dkdriver vnddkdriver = { vndstrategy };
157 
158 /* called by main() at boot time */
159 void	vndattach(int);
160 
161 void	vndclear(struct vnd_softc *);
162 void	vndstart(struct vnd_softc *);
163 int	vndsetcred(struct vnd_softc *, struct ucred *);
164 void	vndiodone(struct buf *);
165 void	vndshutdown(void);
166 void	vndgetdisklabel(dev_t, struct vnd_softc *, struct disklabel *, int);
167 void	vndencrypt(struct vnd_softc *, caddr_t, size_t, daddr64_t, int);
168 size_t	vndbdevsize(struct vnode *, struct proc *);
169 
170 #define vndlock(sc) rw_enter(&sc->sc_rwlock, RW_WRITE|RW_INTR)
171 #define vndunlock(sc) rw_exit_write(&sc->sc_rwlock)
172 
173 void
174 vndencrypt(struct vnd_softc *vnd, caddr_t addr, size_t size, daddr64_t off,
175     int encrypt)
176 {
177 	int i, bsize;
178 	u_char iv[8];
179 
180 	bsize = dbtob(1);
181 	for (i = 0; i < size/bsize; i++) {
182 		bzero(iv, sizeof(iv));
183 		bcopy((u_char *)&off, iv, sizeof(off));
184 		blf_ecb_encrypt(vnd->sc_keyctx, iv, sizeof(iv));
185 		if (encrypt)
186 			blf_cbc_encrypt(vnd->sc_keyctx, iv, addr, bsize);
187 		else
188 			blf_cbc_decrypt(vnd->sc_keyctx, iv, addr, bsize);
189 
190 		addr += bsize;
191 		off++;
192 	}
193 }
194 
195 void
196 vndattach(int num)
197 {
198 	char *mem;
199 	u_long size;
200 	int i;
201 
202 	if (num <= 0)
203 		return;
204 	size = num * sizeof(struct vnd_softc);
205 	mem = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
206 	if (mem == NULL) {
207 		printf("WARNING: no memory for vnode disks\n");
208 		return;
209 	}
210 	vnd_softc = (struct vnd_softc *)mem;
211 	for (i = 0; i < num; i++) {
212 		rw_init(&vnd_softc[i].sc_rwlock, "vndlock");
213 	}
214 	numvnd = num;
215 
216 	pool_init(&vndbufpl, sizeof(struct vndbuf), 0, 0, 0, "vndbufpl", NULL);
217 	pool_setlowat(&vndbufpl, 16);
218 	pool_sethiwat(&vndbufpl, 1024);
219 }
220 
221 int
222 vndopen(dev_t dev, int flags, int mode, struct proc *p)
223 {
224 	int unit = vndunit(dev);
225 	struct vnd_softc *sc;
226 	int error = 0, part, pmask;
227 
228 	DNPRINTF(VDB_FOLLOW, "vndopen(%x, %x, %x, %p)\n", dev, flags, mode, p);
229 
230 	if (unit >= numvnd)
231 		return (ENXIO);
232 	sc = &vnd_softc[unit];
233 
234 	if ((error = vndlock(sc)) != 0)
235 		return (error);
236 
237 	if (!vndsimple(dev) && sc->sc_vp != NULL &&
238 	    (sc->sc_vp->v_type != VREG || sc->sc_keyctx != NULL)) {
239 		error = EINVAL;
240 		goto bad;
241 	}
242 
243 	if ((flags & FWRITE) && (sc->sc_flags & VNF_READONLY)) {
244 		error = EROFS;
245 		goto bad;
246 	}
247 
248 	if ((sc->sc_flags & VNF_INITED) &&
249 	    (sc->sc_flags & VNF_HAVELABEL) == 0) {
250 		sc->sc_flags |= VNF_HAVELABEL;
251 		vndgetdisklabel(dev, sc, sc->sc_dk.dk_label, 0);
252 	}
253 
254 	part = DISKPART(dev);
255 	pmask = 1 << part;
256 
257 	/*
258 	 * If any partition is open, all succeeding openings must be of the
259 	 * same type or read-only.
260 	 */
261 	if (sc->sc_dk.dk_openmask) {
262 		if (((sc->sc_flags & VNF_SIMPLE) != 0) !=
263 		    (vndsimple(dev) != 0) && (flags & FWRITE)) {
264 			error = EBUSY;
265 			goto bad;
266 		}
267 	} else if (vndsimple(dev))
268 		sc->sc_flags |= VNF_SIMPLE;
269 	else
270 		sc->sc_flags &= ~VNF_SIMPLE;
271 
272 	/* Check that the partition exists. */
273 	if (part != RAW_PART &&
274 	    ((sc->sc_flags & VNF_HAVELABEL) == 0 ||
275 	    part >= sc->sc_dk.dk_label->d_npartitions ||
276 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
277 		error = ENXIO;
278 		goto bad;
279 	}
280 
281 	/* Prevent our unit from being unconfigured while open. */
282 	switch (mode) {
283 	case S_IFCHR:
284 		sc->sc_dk.dk_copenmask |= pmask;
285 		break;
286 
287 	case S_IFBLK:
288 		sc->sc_dk.dk_bopenmask |= pmask;
289 		break;
290 	}
291 	sc->sc_dk.dk_openmask =
292 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
293 
294 	error = 0;
295 bad:
296 	vndunlock(sc);
297 	return (error);
298 }
299 
300 /*
301  * Load the label information on the named device
302  */
303 void
304 vndgetdisklabel(dev_t dev, struct vnd_softc *sc, struct disklabel *lp,
305     int spoofonly)
306 {
307 	char *errstring = NULL;
308 
309 	bzero(lp, sizeof(struct disklabel));
310 
311 	lp->d_secsize = sc->sc_secsize;
312 	lp->d_nsectors = sc->sc_nsectors;
313 	lp->d_ntracks = sc->sc_ntracks;
314 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
315 	lp->d_ncylinders = sc->sc_size / lp->d_secpercyl;
316 
317 	strncpy(lp->d_typename, "vnd device", sizeof(lp->d_typename));
318 	lp->d_type = DTYPE_VND;
319 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
320 	DL_SETDSIZE(lp, sc->sc_size);
321 	lp->d_rpm = 3600;
322 	lp->d_interleave = 1;
323 	lp->d_flags = 0;
324 	lp->d_version = 1;
325 
326 	lp->d_magic = DISKMAGIC;
327 	lp->d_magic2 = DISKMAGIC;
328 	lp->d_checksum = dkcksum(lp);
329 
330 	/* Call the generic disklabel extraction routine */
331 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, spoofonly);
332 	if (errstring) {
333 		DNPRINTF(VDB_IO, "%s: %s\n", sc->sc_dev.dv_xname,
334 		    errstring);
335 		return;
336 	}
337 }
338 
339 int
340 vndclose(dev_t dev, int flags, int mode, struct proc *p)
341 {
342 	int unit = vndunit(dev);
343 	struct vnd_softc *sc;
344 	int error = 0, part;
345 
346 	DNPRINTF(VDB_FOLLOW, "vndclose(%x, %x, %x, %p)\n", dev, flags, mode, p);
347 
348 	if (unit >= numvnd)
349 		return (ENXIO);
350 	sc = &vnd_softc[unit];
351 
352 	if ((error = vndlock(sc)) != 0)
353 		return (error);
354 
355 	part = DISKPART(dev);
356 
357 	/* ...that much closer to allowing unconfiguration... */
358 	switch (mode) {
359 	case S_IFCHR:
360 		sc->sc_dk.dk_copenmask &= ~(1 << part);
361 		break;
362 
363 	case S_IFBLK:
364 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
365 		break;
366 	}
367 	sc->sc_dk.dk_openmask =
368 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
369 
370 	vndunlock(sc);
371 	return (0);
372 }
373 
374 /*
375  * Two methods are used, the traditional buffercache bypassing and the
376  * newer, cache-coherent on unmount, one.
377  *
378  * Former method:
379  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
380  * Note that this driver can only be used for swapping over NFS on the hp
381  * since nfs_strategy on the vax cannot handle u-areas and page tables.
382  *
383  * Latter method:
384  * Repack the buffer into an uio structure and use VOP_READ/VOP_WRITE to
385  * access the underlying file.
386  */
387 void
388 vndstrategy(struct buf *bp)
389 {
390 	int unit = vndunit(bp->b_dev);
391 	struct vnd_softc *vnd = &vnd_softc[unit];
392 	struct vndbuf *nbp;
393 	int bsize;
394 	off_t bn;
395 	caddr_t addr;
396 	size_t resid;
397 	int sz, flags, error, s;
398 	struct iovec aiov;
399 	struct uio auio;
400 	struct proc *p = curproc;
401 
402 	DNPRINTF(VDB_FOLLOW, "vndstrategy(%p): unit %d\n", bp, unit);
403 
404 	if ((vnd->sc_flags & VNF_INITED) == 0) {
405 		bp->b_error = ENXIO;
406 		bp->b_flags |= B_ERROR;
407 		s = splbio();
408 		biodone(bp);
409 		splx(s);
410 		return;
411 	}
412 
413 	/* Ensure that the requested block is sector aligned. */
414 	if (bp->b_blkno % DL_BLKSPERSEC(vnd->sc_dk.dk_label) != 0) {
415 		bp->b_error = EINVAL;
416 		bp->b_flags |= B_ERROR;
417 		s = splbio();
418 		biodone(bp);
419 		splx(s);
420 		return;
421 	}
422 
423 	bn = bp->b_blkno;
424 	bp->b_resid = bp->b_bcount;
425 
426 	if (bn < 0) {
427 		bp->b_error = EINVAL;
428 		bp->b_flags |= B_ERROR;
429 		s = splbio();
430 		biodone(bp);
431 		splx(s);
432 		return;
433 	}
434 
435 	/* If we have a label, do a boundary check. */
436 	if (vnd->sc_flags & VNF_HAVELABEL) {
437 		if (bounds_check_with_label(bp, vnd->sc_dk.dk_label, 1) <= 0) {
438 			s = splbio();
439 			biodone(bp);
440 			splx(s);
441 			return;
442 		}
443 
444 		/*
445 		 * bounds_check_with_label() changes bp->b_resid, reset it
446 		 */
447 		bp->b_resid = bp->b_bcount;
448 	}
449 
450 	if (vnd->sc_flags & VNF_HAVELABEL)
451 		sz = howmany(bp->b_bcount, vnd->sc_dk.dk_label->d_secsize);
452 	else
453 		sz = howmany(bp->b_bcount, DEV_BSIZE);
454 
455 	/* No bypassing of buffer cache?  */
456 	if (vndsimple(bp->b_dev)) {
457 		/* Loop until all queued requests are handled.  */
458 		for (;;) {
459 			int part = DISKPART(bp->b_dev);
460 			daddr64_t off = DL_SECTOBLK(vnd->sc_dk.dk_label,
461 			    DL_GETPOFFSET(&vnd->sc_dk.dk_label->d_partitions[part]));
462 			aiov.iov_base = bp->b_data;
463 			auio.uio_resid = aiov.iov_len = bp->b_bcount;
464 			auio.uio_iov = &aiov;
465 			auio.uio_iovcnt = 1;
466 			auio.uio_offset = dbtob((off_t)(bp->b_blkno + off));
467 			auio.uio_segflg = UIO_SYSSPACE;
468 			auio.uio_procp = p;
469 
470 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY, p);
471 			if (bp->b_flags & B_READ) {
472 				auio.uio_rw = UIO_READ;
473 				bp->b_error = VOP_READ(vnd->sc_vp, &auio, 0,
474 				    vnd->sc_cred);
475 				if (vnd->sc_keyctx)
476 					vndencrypt(vnd,	bp->b_data,
477 					   bp->b_bcount, bp->b_blkno, 0);
478 			} else {
479 				if (vnd->sc_keyctx)
480 					vndencrypt(vnd, bp->b_data,
481 					   bp->b_bcount, bp->b_blkno, 1);
482 				auio.uio_rw = UIO_WRITE;
483 				/*
484 				 * Upper layer has already checked I/O for
485 				 * limits, so there is no need to do it again.
486 				 */
487 				bp->b_error = VOP_WRITE(vnd->sc_vp, &auio,
488 				    IO_NOLIMIT, vnd->sc_cred);
489 				/* Data in buffer cache needs to be in clear */
490 				if (vnd->sc_keyctx)
491 					vndencrypt(vnd, bp->b_data,
492 					   bp->b_bcount, bp->b_blkno, 0);
493 			}
494 			VOP_UNLOCK(vnd->sc_vp, 0, p);
495 			if (bp->b_error)
496 				bp->b_flags |= B_ERROR;
497 			bp->b_resid = auio.uio_resid;
498 			s = splbio();
499 			biodone(bp);
500 			splx(s);
501 
502 			/* If nothing more is queued, we are done.  */
503 			if (!vnd->sc_tab.b_active)
504 				return;
505 
506 			/*
507 			 * Dequeue now since lower level strategy
508 			 * routine might queue using same links.
509 			 */
510 			s = splbio();
511 			bp = vnd->sc_tab.b_actf;
512 			vnd->sc_tab.b_actf = bp->b_actf;
513 			vnd->sc_tab.b_active--;
514 			splx(s);
515 		}
516 	}
517 
518 	if (vnd->sc_vp->v_type != VREG || vnd->sc_keyctx != NULL) {
519 		bp->b_error = EINVAL;
520 		bp->b_flags |= B_ERROR;
521 		s = splbio();
522 		biodone(bp);
523 		splx(s);
524 		return;
525 	}
526 
527 	/* The old-style buffercache bypassing method.  */
528 	bn += DL_SECTOBLK(vnd->sc_dk.dk_label,
529 	    DL_GETPOFFSET(&vnd->sc_dk.dk_label->d_partitions[DISKPART(bp->b_dev)]));
530 	bn = dbtob(bn);
531 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
532 	addr = bp->b_data;
533 	flags = bp->b_flags | B_CALL;
534 	for (resid = bp->b_resid; resid; resid -= sz) {
535 		struct vnode *vp;
536 		daddr64_t nbn;
537 		int off, nra;
538 
539 		nra = 0;
540 		vn_lock(vnd->sc_vp, LK_RETRY | LK_EXCLUSIVE, p);
541 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
542 		VOP_UNLOCK(vnd->sc_vp, 0, p);
543 		if (error == 0 && (long)nbn == -1)
544 			error = EIO;
545 #ifdef VNDDEBUG
546 		if (!dovndcluster)
547 			nra = 0;
548 #endif
549 
550 		if ((off = bn % bsize) != 0)
551 			sz = bsize - off;
552 		else
553 			sz = (1 + nra) * bsize;
554 		if (resid < sz)
555 			sz = resid;
556 
557 		DNPRINTF(VDB_IO, "vndstrategy: vp %p/%p bn %x/%x sz %x\n",
558 		    vnd->sc_vp, vp, bn, nbn, sz);
559 
560 		s = splbio();
561 		nbp = getvndbuf();
562 		splx(s);
563 		nbp->vb_buf.b_flags = flags;
564 		nbp->vb_buf.b_bcount = sz;
565 		nbp->vb_buf.b_bufsize = bp->b_bufsize;
566 		nbp->vb_buf.b_error = 0;
567 		if (vp->v_type == VBLK || vp->v_type == VCHR)
568 			nbp->vb_buf.b_dev = vp->v_rdev;
569 		else
570 			nbp->vb_buf.b_dev = NODEV;
571 		nbp->vb_buf.b_data = addr;
572 		nbp->vb_buf.b_blkno = nbn + btodb(off);
573 		nbp->vb_buf.b_proc = bp->b_proc;
574 		nbp->vb_buf.b_iodone = vndiodone;
575 		nbp->vb_buf.b_vp = vp;
576 		nbp->vb_buf.b_dirtyoff = bp->b_dirtyoff;
577 		nbp->vb_buf.b_dirtyend = bp->b_dirtyend;
578 		nbp->vb_buf.b_validoff = bp->b_validoff;
579 		nbp->vb_buf.b_validend = bp->b_validend;
580 		LIST_INIT(&nbp->vb_buf.b_dep);
581 
582 		/* save a reference to the old buffer */
583 		nbp->vb_obp = bp;
584 
585 		/*
586 		 * If there was an error or a hole in the file...punt.
587 		 * Note that we deal with this after the nbp allocation.
588 		 * This ensures that we properly clean up any operations
589 		 * that we have already fired off.
590 		 *
591 		 * XXX we could deal with holes here but it would be
592 		 * a hassle (in the write case).
593 		 * We must still however charge for the write even if there
594 		 * was an error.
595 		 */
596 		if (error) {
597 			nbp->vb_buf.b_error = error;
598 			nbp->vb_buf.b_flags |= B_ERROR;
599 			bp->b_resid -= (resid - sz);
600 			s = splbio();
601 			/* charge for the write */
602 			if ((nbp->vb_buf.b_flags & B_READ) == 0)
603 				nbp->vb_buf.b_vp->v_numoutput++;
604 			biodone(&nbp->vb_buf);
605 			splx(s);
606 			return;
607 		}
608 		/*
609 		 * Just sort by block number
610 		 */
611 		nbp->vb_buf.b_cylinder = nbp->vb_buf.b_blkno;
612 		s = splbio();
613 		disksort(&vnd->sc_tab, &nbp->vb_buf);
614 		vnd->sc_tab.b_active++;
615 		vndstart(vnd);
616 		splx(s);
617 		bn += sz;
618 		addr += sz;
619 	}
620 }
621 
622 /*
623  * Feed requests sequentially.
624  * We do it this way to keep from flooding NFS servers if we are connected
625  * to an NFS file.  This places the burden on the client rather than the
626  * server.
627  */
628 void
629 vndstart(struct vnd_softc *vnd)
630 {
631 	struct buf *bp;
632 
633 	/*
634 	 * Dequeue now since lower level strategy routine might
635 	 * queue using same links
636 	 */
637 	bp = vnd->sc_tab.b_actf;
638 	vnd->sc_tab.b_actf = bp->b_actf;
639 
640 	DNPRINTF(VDB_IO,
641 	    "vndstart(%d): bp %p vp %p blkno %x addr %p cnt %lx\n",
642 	    vnd-vnd_softc, bp, bp->b_vp, bp->b_blkno, bp->b_data,
643 	    bp->b_bcount);
644 
645 	/* Instrumentation. */
646 	disk_busy(&vnd->sc_dk);
647 
648 	if ((bp->b_flags & B_READ) == 0)
649 		bp->b_vp->v_numoutput++;
650 	VOP_STRATEGY(bp);
651 }
652 
653 void
654 vndiodone(struct buf *bp)
655 {
656 	struct vndbuf *vbp = (struct vndbuf *) bp;
657 	struct buf *pbp = vbp->vb_obp;
658 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
659 
660 	splassert(IPL_BIO);
661 
662 	DNPRINTF(VDB_IO,
663 	    "vndiodone(%d): vbp %p vp %p blkno %x addr %p cnt %lx\n",
664 	    vnd-vnd_softc, vbp, vbp->vb_buf.b_vp, vbp->vb_buf.b_blkno,
665 	    vbp->vb_buf.b_data, vbp->vb_buf.b_bcount);
666 
667 	if (vbp->vb_buf.b_error) {
668 		DNPRINTF(VDB_IO, "vndiodone: vbp %p error %d\n", vbp,
669 		    vbp->vb_buf.b_error);
670 
671 		pbp->b_flags |= B_ERROR;
672 		/* XXX does this matter here? */
673 		(&vbp->vb_buf)->b_flags |= B_RAW;
674 		pbp->b_error = biowait(&vbp->vb_buf);
675 	}
676 	pbp->b_resid -= vbp->vb_buf.b_bcount;
677 	putvndbuf(vbp);
678 	if (vnd->sc_tab.b_active) {
679 		disk_unbusy(&vnd->sc_dk, (pbp->b_bcount - pbp->b_resid),
680 		    (pbp->b_flags & B_READ));
681 		if (!vnd->sc_tab.b_actf)
682 			vnd->sc_tab.b_active--;
683 	}
684 	if (pbp->b_resid == 0) {
685 		DNPRINTF(VDB_IO, "vndiodone: pbp %p iodone\n", pbp);
686 		biodone(pbp);
687 	}
688 
689 }
690 
691 /* ARGSUSED */
692 int
693 vndread(dev_t dev, struct uio *uio, int flags)
694 {
695 	int unit = vndunit(dev);
696 	struct vnd_softc *sc;
697 
698 	DNPRINTF(VDB_FOLLOW, "vndread(%x, %p)\n", dev, uio);
699 
700 	if (unit >= numvnd)
701 		return (ENXIO);
702 	sc = &vnd_softc[unit];
703 
704 	if ((sc->sc_flags & VNF_INITED) == 0)
705 		return (ENXIO);
706 
707 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
708 }
709 
710 /* ARGSUSED */
711 int
712 vndwrite(dev_t dev, struct uio *uio, int flags)
713 {
714 	int unit = vndunit(dev);
715 	struct vnd_softc *sc;
716 
717 	DNPRINTF(VDB_FOLLOW, "vndwrite(%x, %p)\n", dev, uio);
718 
719 	if (unit >= numvnd)
720 		return (ENXIO);
721 	sc = &vnd_softc[unit];
722 
723 	if ((sc->sc_flags & VNF_INITED) == 0)
724 		return (ENXIO);
725 
726 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
727 }
728 
729 size_t
730 vndbdevsize(struct vnode *vp, struct proc *p)
731 {
732 	struct partinfo pi;
733 	struct bdevsw *bsw;
734 	dev_t dev;
735 
736 	dev = vp->v_rdev;
737 	bsw = bdevsw_lookup(dev);
738 	if (bsw->d_ioctl == NULL)
739 		return (0);
740 	if (bsw->d_ioctl(dev, DIOCGPART, (caddr_t)&pi, FREAD, p))
741 		return (0);
742 	DNPRINTF(VDB_INIT, "vndbdevsize: size %li secsize %li\n",
743 	    (long)pi.part->p_size,(long)pi.disklab->d_secsize);
744 	return (pi.part->p_size);
745 }
746 
747 /* ARGSUSED */
748 int
749 vndioctl(dev_t dev, u_long cmd, caddr_t addr, int flag, struct proc *p)
750 {
751 	int unit = vndunit(dev);
752 	struct vnd_softc *vnd;
753 	struct vnd_ioctl *vio;
754 	struct vnd_user *vnu;
755 	struct vattr vattr;
756 	struct nameidata nd;
757 	int error, part, pmask, s;
758 
759 	DNPRINTF(VDB_FOLLOW, "vndioctl(%x, %lx, %p, %x, %p): unit %d\n",
760 	    dev, cmd, addr, flag, p, unit);
761 
762 	error = suser(p, 0);
763 	if (error)
764 		return (error);
765 	if (unit >= numvnd)
766 		return (ENXIO);
767 
768 	vnd = &vnd_softc[unit];
769 	vio = (struct vnd_ioctl *)addr;
770 	switch (cmd) {
771 
772 	case VNDIOCSET:
773 		if (vnd->sc_flags & VNF_INITED)
774 			return (EBUSY);
775 		if (!(vnd->sc_flags & VNF_SIMPLE) && vio->vnd_keylen)
776 			return (EINVAL);
777 
778 		if ((error = vndlock(vnd)) != 0)
779 			return (error);
780 
781 		if ((error = copyinstr(vio->vnd_file, vnd->sc_file,
782 		    sizeof(vnd->sc_file), NULL))) {
783 			vndunlock(vnd);
784 			return (error);
785 		}
786 
787 		bzero(vnd->sc_dev.dv_xname, sizeof(vnd->sc_dev.dv_xname));
788 		if (snprintf(vnd->sc_dev.dv_xname, sizeof(vnd->sc_dev.dv_xname),
789 		    "vnd%d", unit) >= sizeof(vnd->sc_dev.dv_xname)) {
790 			printf("VNDIOCSET: device name too long\n");
791 			vndunlock(vnd);
792 			return(ENXIO);
793 		}
794 
795 		/* Set geometry for device. */
796 		vnd->sc_secsize = vio->vnd_secsize;
797 		vnd->sc_ntracks = vio->vnd_ntracks;
798 		vnd->sc_nsectors = vio->vnd_nsectors;
799 
800 		/*
801 		 * Open for read and write first. This lets vn_open() weed out
802 		 * directories, sockets, etc. so we don't have to worry about
803 		 * them.
804 		 */
805 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
806 		vnd->sc_flags &= ~VNF_READONLY;
807 		error = vn_open(&nd, FREAD|FWRITE, 0);
808 		if (error == EROFS) {
809 			vnd->sc_flags |= VNF_READONLY;
810 			error = vn_open(&nd, FREAD, 0);
811 		}
812 		if (error) {
813 			vndunlock(vnd);
814 			return (error);
815 		}
816 
817 		if (nd.ni_vp->v_type != VREG && !vndsimple(dev)) {
818 			VOP_UNLOCK(nd.ni_vp, 0, p);
819 			vn_close(nd.ni_vp, VNDRW(vnd), p->p_ucred, p);
820 			vndunlock(vnd);
821 			return (EINVAL);
822 		}
823 
824 		if (nd.ni_vp->v_type == VBLK)
825 			vnd->sc_size = vndbdevsize(nd.ni_vp, p);
826 		else {
827 			error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
828 			if (error) {
829 				VOP_UNLOCK(nd.ni_vp, 0, p);
830 				vn_close(nd.ni_vp, VNDRW(vnd), p->p_ucred, p);
831 				vndunlock(vnd);
832 				return (error);
833 			}
834 			vnd->sc_size = vattr.va_size / vnd->sc_secsize;
835 		}
836 		VOP_UNLOCK(nd.ni_vp, 0, p);
837 		vnd->sc_vp = nd.ni_vp;
838 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0) {
839 			(void) vn_close(nd.ni_vp, VNDRW(vnd), p->p_ucred, p);
840 			vndunlock(vnd);
841 			return (error);
842 		}
843 
844 		if (vio->vnd_keylen > 0) {
845 			char key[BLF_MAXUTILIZED];
846 
847 			if (vio->vnd_keylen > sizeof(key))
848 				vio->vnd_keylen = sizeof(key);
849 
850 			if ((error = copyin(vio->vnd_key, key,
851 			    vio->vnd_keylen)) != 0) {
852 				(void) vn_close(nd.ni_vp, VNDRW(vnd),
853 				    p->p_ucred, p);
854 				vndunlock(vnd);
855 				return (error);
856 			}
857 
858 			vnd->sc_keyctx = malloc(sizeof(*vnd->sc_keyctx), M_DEVBUF,
859 			    M_WAITOK);
860 			blf_key(vnd->sc_keyctx, key, vio->vnd_keylen);
861 			bzero(key, vio->vnd_keylen);
862 		} else
863 			vnd->sc_keyctx = NULL;
864 
865 		vio->vnd_size = vnd->sc_size * vnd->sc_secsize;
866 		vnd->sc_flags |= VNF_INITED;
867 
868 		DNPRINTF(VDB_INIT, "vndioctl: SET vp %p size %llx\n",
869 		    vnd->sc_vp, (unsigned long long)vnd->sc_size);
870 
871 		/* Attach the disk. */
872 		vnd->sc_dk.dk_driver = &vnddkdriver;
873 		vnd->sc_dk.dk_name = vnd->sc_dev.dv_xname;
874 		disk_attach(&vnd->sc_dk);
875 
876 		vndunlock(vnd);
877 
878 		break;
879 
880 	case VNDIOCCLR:
881 		if ((vnd->sc_flags & VNF_INITED) == 0)
882 			return (ENXIO);
883 
884 		if ((error = vndlock(vnd)) != 0)
885 			return (error);
886 
887 		/*
888 		 * Don't unconfigure if any other partitions are open
889 		 * or if both the character and block flavors of this
890 		 * partition are open.
891 		 */
892 		part = DISKPART(dev);
893 		pmask = (1 << part);
894 		if ((vnd->sc_dk.dk_openmask & ~pmask) ||
895 		    ((vnd->sc_dk.dk_bopenmask & pmask) &&
896 		    (vnd->sc_dk.dk_copenmask & pmask))) {
897 			vndunlock(vnd);
898 			return (EBUSY);
899 		}
900 
901 		vndclear(vnd);
902 		DNPRINTF(VDB_INIT, "vndioctl: CLRed\n");
903 
904 		/* Free crypto key */
905 		if (vnd->sc_keyctx) {
906 			bzero(vnd->sc_keyctx, sizeof(*vnd->sc_keyctx));
907 			free(vnd->sc_keyctx, M_DEVBUF);
908 		}
909 
910 		/* Detach the disk. */
911 		disk_detach(&vnd->sc_dk);
912 
913 		/* This must be atomic. */
914 		s = splhigh();
915 		vndunlock(vnd);
916 		bzero(vnd, sizeof(struct vnd_softc));
917 		splx(s);
918 		break;
919 
920 	case VNDIOCGET:
921 		vnu = (struct vnd_user *)addr;
922 
923 		if (vnu->vnu_unit == -1)
924 			vnu->vnu_unit = unit;
925 		if (vnu->vnu_unit >= numvnd)
926 			return (ENXIO);
927 		if (vnu->vnu_unit < 0)
928 			return (EINVAL);
929 
930 		vnd = &vnd_softc[vnu->vnu_unit];
931 
932 		if (vnd->sc_flags & VNF_INITED) {
933 			error = VOP_GETATTR(vnd->sc_vp, &vattr, p->p_ucred, p);
934 			if (error)
935 				return (error);
936 
937 			strlcpy(vnu->vnu_file, vnd->sc_file,
938 			    sizeof(vnu->vnu_file));
939 			vnu->vnu_dev = vattr.va_fsid;
940 			vnu->vnu_ino = vattr.va_fileid;
941 		} else {
942 			vnu->vnu_dev = 0;
943 			vnu->vnu_ino = 0;
944 		}
945 
946 		break;
947 
948 	case DIOCGPDINFO:
949 		if ((vnd->sc_flags & VNF_HAVELABEL) == 0)
950 			return (ENOTTY);
951 		vndgetdisklabel(dev, vnd, (struct disklabel *)addr, 1);
952 		return (0);
953 
954 	case DIOCGDINFO:
955 		if ((vnd->sc_flags & VNF_HAVELABEL) == 0)
956 			return (ENOTTY);
957 		*(struct disklabel *)addr = *(vnd->sc_dk.dk_label);
958 		return (0);
959 
960 	case DIOCGPART:
961 		if ((vnd->sc_flags & VNF_HAVELABEL) == 0)
962 			return (ENOTTY);
963 		((struct partinfo *)addr)->disklab = vnd->sc_dk.dk_label;
964 		((struct partinfo *)addr)->part =
965 		    &vnd->sc_dk.dk_label->d_partitions[DISKPART(dev)];
966 		return (0);
967 
968 	case DIOCWDINFO:
969 	case DIOCSDINFO:
970 		if ((vnd->sc_flags & VNF_HAVELABEL) == 0)
971 			return (ENOTTY);
972 		if ((flag & FWRITE) == 0)
973 			return (EBADF);
974 
975 		if ((error = vndlock(vnd)) != 0)
976 			return (error);
977 		vnd->sc_flags |= VNF_LABELLING;
978 
979 		error = setdisklabel(vnd->sc_dk.dk_label,
980 		    (struct disklabel *)addr, /*vnd->sc_dk.dk_openmask : */0);
981 		if (error == 0) {
982 			if (cmd == DIOCWDINFO)
983 				error = writedisklabel(VNDLABELDEV(dev),
984 				    vndstrategy, vnd->sc_dk.dk_label);
985 		}
986 
987 		vnd->sc_flags &= ~VNF_LABELLING;
988 		vndunlock(vnd);
989 		return (error);
990 
991 	case DIOCWLABEL:
992 		if ((flag & FWRITE) == 0)
993 			return (EBADF);
994 		if (*(int *)addr)
995 			vnd->sc_flags |= VNF_WLABEL;
996 		else
997 			vnd->sc_flags &= ~VNF_WLABEL;
998 		return (0);
999 
1000 	default:
1001 		return (ENOTTY);
1002 	}
1003 
1004 	return (0);
1005 }
1006 
1007 /*
1008  * Duplicate the current processes' credentials.  Since we are called only
1009  * as the result of a SET ioctl and only root can do that, any future access
1010  * to this "disk" is essentially as root.  Note that credentials may change
1011  * if some other uid can write directly to the mapped file (NFS).
1012  */
1013 int
1014 vndsetcred(struct vnd_softc *vnd, struct ucred *cred)
1015 {
1016 	struct uio auio;
1017 	struct iovec aiov;
1018 	char *tmpbuf;
1019 	int error;
1020 	struct proc *p = curproc;
1021 
1022 	vnd->sc_cred = crdup(cred);
1023 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1024 
1025 	/* XXX: Horrible kludge to establish credentials for NFS */
1026 	aiov.iov_base = tmpbuf;
1027 	aiov.iov_len = MIN(DEV_BSIZE, vnd->sc_size * vnd->sc_secsize);
1028 	auio.uio_iov = &aiov;
1029 	auio.uio_iovcnt = 1;
1030 	auio.uio_offset = 0;
1031 	auio.uio_rw = UIO_READ;
1032 	auio.uio_segflg = UIO_SYSSPACE;
1033 	auio.uio_resid = aiov.iov_len;
1034 	vn_lock(vnd->sc_vp, LK_RETRY | LK_EXCLUSIVE, p);
1035 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1036 	VOP_UNLOCK(vnd->sc_vp, 0, p);
1037 
1038 	free(tmpbuf, M_TEMP);
1039 	return (error);
1040 }
1041 
1042 void
1043 vndshutdown(void)
1044 {
1045 	struct vnd_softc *vnd;
1046 
1047 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1048 		if (vnd->sc_flags & VNF_INITED)
1049 			vndclear(vnd);
1050 }
1051 
1052 void
1053 vndclear(struct vnd_softc *vnd)
1054 {
1055 	struct vnode *vp = vnd->sc_vp;
1056 	struct proc *p = curproc;		/* XXX */
1057 
1058 	DNPRINTF(VDB_FOLLOW, "vndclear(%p): vp %p\n", vnd, vp);
1059 
1060 	vnd->sc_flags &= ~VNF_INITED;
1061 	if (vp == NULL)
1062 		panic("vndioctl: null vp");
1063 	(void) vn_close(vp, VNDRW(vnd), vnd->sc_cred, p);
1064 	crfree(vnd->sc_cred);
1065 	vnd->sc_vp = NULL;
1066 	vnd->sc_cred = NULL;
1067 	vnd->sc_size = 0;
1068 }
1069 
1070 daddr64_t
1071 vndsize(dev_t dev)
1072 {
1073 	int unit = vndunit(dev);
1074 	struct vnd_softc *vnd = &vnd_softc[unit];
1075 
1076 	if (unit >= numvnd || (vnd->sc_flags & VNF_INITED) == 0)
1077 		return (-1);
1078 	return (vnd->sc_size * (vnd->sc_secsize / DEV_BSIZE));
1079 }
1080 
1081 int
1082 vnddump(dev_t dev, daddr64_t blkno, caddr_t va, size_t size)
1083 {
1084 
1085 	/* Not implemented. */
1086 	return (ENXIO);
1087 }
1088