vnd.c revision 1.219 1 /* $NetBSD: vnd.c,v 1.219 2011/10/14 09:23:30 hannken Exp $ */
2
3 /*-
4 * Copyright (c) 1996, 1997, 1998, 2008 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe.
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 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 /*
33 * Copyright (c) 1988 University of Utah.
34 * Copyright (c) 1990, 1993
35 * The Regents of the University of California. All rights reserved.
36 *
37 * This code is derived from software contributed to Berkeley by
38 * the Systems Programming Group of the University of Utah Computer
39 * Science Department.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 * notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 * notice, this list of conditions and the following disclaimer in the
48 * documentation and/or other materials provided with the distribution.
49 * 3. Neither the name of the University nor the names of its contributors
50 * may be used to endorse or promote products derived from this software
51 * without specific prior written permission.
52 *
53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63 * SUCH DAMAGE.
64 *
65 * from: Utah $Hdr: vn.c 1.13 94/04/02$
66 *
67 * @(#)vn.c 8.9 (Berkeley) 5/14/95
68 */
69
70 /*
71 * Vnode disk driver.
72 *
73 * Block/character interface to a vnode. Allows one to treat a file
74 * as a disk (e.g. build a filesystem in it, mount it, etc.).
75 *
76 * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
77 * this uses them to avoid distorting the local buffer cache. If those
78 * block-level operations are not available, this falls back to the regular
79 * read and write calls. Using these may distort the cache in some cases
80 * but better have the driver working than preventing it to work on file
81 * systems where the block-level operations are not implemented for
82 * whatever reason.
83 *
84 * NOTE 2: There is a security issue involved with this driver.
85 * Once mounted all access to the contents of the "mapped" file via
86 * the special file is controlled by the permissions on the special
87 * file, the protection of the mapped file is ignored (effectively,
88 * by using root credentials in all transactions).
89 *
90 * NOTE 3: Doesn't interact with leases, should it?
91 */
92
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.219 2011/10/14 09:23:30 hannken Exp $");
95
96 #if defined(_KERNEL_OPT)
97 #include "opt_vnd.h"
98 #include "opt_compat_netbsd.h"
99 #endif
100
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/namei.h>
104 #include <sys/proc.h>
105 #include <sys/kthread.h>
106 #include <sys/errno.h>
107 #include <sys/buf.h>
108 #include <sys/bufq.h>
109 #include <sys/malloc.h>
110 #include <sys/ioctl.h>
111 #include <sys/disklabel.h>
112 #include <sys/device.h>
113 #include <sys/disk.h>
114 #include <sys/stat.h>
115 #include <sys/mount.h>
116 #include <sys/vnode.h>
117 #include <sys/file.h>
118 #include <sys/uio.h>
119 #include <sys/conf.h>
120 #include <sys/kauth.h>
121
122 #include <net/zlib.h>
123
124 #include <miscfs/genfs/genfs.h>
125 #include <miscfs/specfs/specdev.h>
126
127 #include <dev/dkvar.h>
128 #include <dev/vndvar.h>
129
130 #include <prop/proplib.h>
131
132 #if defined(VNDDEBUG) && !defined(DEBUG)
133 #define DEBUG
134 #endif
135
136 #ifdef DEBUG
137 int dovndcluster = 1;
138 #define VDB_FOLLOW 0x01
139 #define VDB_INIT 0x02
140 #define VDB_IO 0x04
141 #define VDB_LABEL 0x08
142 int vnddebug = 0x00;
143 #endif
144
145 #define vndunit(x) DISKUNIT(x)
146
147 struct vndxfer {
148 struct buf vx_buf;
149 struct vnd_softc *vx_vnd;
150 };
151 #define VND_BUFTOXFER(bp) ((struct vndxfer *)(void *)bp)
152
153 #define VND_GETXFER(vnd) pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
154 #define VND_PUTXFER(vnd, vx) pool_put(&(vnd)->sc_vxpool, (vx))
155
156 #define VNDLABELDEV(dev) \
157 (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
158
159 /* called by main() at boot time */
160 void vndattach(int);
161
162 static void vndclear(struct vnd_softc *, int);
163 static int vnddoclear(struct vnd_softc *, int, int, bool);
164 static int vndsetcred(struct vnd_softc *, kauth_cred_t);
165 static void vndthrottle(struct vnd_softc *, struct vnode *);
166 static void vndiodone(struct buf *);
167 #if 0
168 static void vndshutdown(void);
169 #endif
170
171 static void vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
172 static void vndgetdisklabel(dev_t, struct vnd_softc *);
173
174 static int vndlock(struct vnd_softc *);
175 static void vndunlock(struct vnd_softc *);
176 #ifdef VND_COMPRESSION
177 static void compstrategy(struct buf *, off_t);
178 static void *vnd_alloc(void *, u_int, u_int);
179 static void vnd_free(void *, void *);
180 #endif /* VND_COMPRESSION */
181
182 static void vndthread(void *);
183 static bool vnode_has_op(const struct vnode *, int);
184 static void handle_with_rdwr(struct vnd_softc *, const struct buf *,
185 struct buf *);
186 static void handle_with_strategy(struct vnd_softc *, const struct buf *,
187 struct buf *);
188 static void vnd_set_properties(struct vnd_softc *);
189
190 static dev_type_open(vndopen);
191 static dev_type_close(vndclose);
192 static dev_type_read(vndread);
193 static dev_type_write(vndwrite);
194 static dev_type_ioctl(vndioctl);
195 static dev_type_strategy(vndstrategy);
196 static dev_type_dump(vnddump);
197 static dev_type_size(vndsize);
198
199 const struct bdevsw vnd_bdevsw = {
200 vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
201 };
202
203 const struct cdevsw vnd_cdevsw = {
204 vndopen, vndclose, vndread, vndwrite, vndioctl,
205 nostop, notty, nopoll, nommap, nokqfilter, D_DISK
206 };
207
208 static int vnd_match(device_t, cfdata_t, void *);
209 static void vnd_attach(device_t, device_t, void *);
210 static int vnd_detach(device_t, int);
211
212 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
213 vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
214 extern struct cfdriver vnd_cd;
215
216 static struct vnd_softc *vnd_spawn(int);
217 int vnd_destroy(device_t);
218
219 static struct dkdriver vnddkdriver = { vndstrategy, minphys };
220
221 void
222 vndattach(int num)
223 {
224 int error;
225
226 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
227 if (error)
228 aprint_error("%s: unable to register cfattach\n",
229 vnd_cd.cd_name);
230 }
231
232 static int
233 vnd_match(device_t self, cfdata_t cfdata, void *aux)
234 {
235
236 return 1;
237 }
238
239 static void
240 vnd_attach(device_t parent, device_t self, void *aux)
241 {
242 struct vnd_softc *sc = device_private(self);
243
244 sc->sc_dev = self;
245 sc->sc_comp_offsets = NULL;
246 sc->sc_comp_buff = NULL;
247 sc->sc_comp_decombuf = NULL;
248 bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
249 disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
250 if (!pmf_device_register(self, NULL, NULL))
251 aprint_error_dev(self, "couldn't establish power handler\n");
252 }
253
254 static int
255 vnd_detach(device_t self, int flags)
256 {
257 int error;
258 struct vnd_softc *sc = device_private(self);
259
260 if (sc->sc_flags & VNF_INITED) {
261 error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
262 if (error != 0)
263 return error;
264 }
265
266 pmf_device_deregister(self);
267 bufq_free(sc->sc_tab);
268 disk_destroy(&sc->sc_dkdev);
269
270 return 0;
271 }
272
273 static struct vnd_softc *
274 vnd_spawn(int unit)
275 {
276 cfdata_t cf;
277
278 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
279 cf->cf_name = vnd_cd.cd_name;
280 cf->cf_atname = vnd_cd.cd_name;
281 cf->cf_unit = unit;
282 cf->cf_fstate = FSTATE_STAR;
283
284 return device_private(config_attach_pseudo(cf));
285 }
286
287 int
288 vnd_destroy(device_t dev)
289 {
290 int error;
291 cfdata_t cf;
292
293 cf = device_cfdata(dev);
294 error = config_detach(dev, DETACH_QUIET);
295 if (error)
296 return error;
297 free(cf, M_DEVBUF);
298 return 0;
299 }
300
301 static int
302 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
303 {
304 int unit = vndunit(dev);
305 struct vnd_softc *sc;
306 int error = 0, part, pmask;
307 struct disklabel *lp;
308
309 #ifdef DEBUG
310 if (vnddebug & VDB_FOLLOW)
311 printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
312 #endif
313 sc = device_lookup_private(&vnd_cd, unit);
314 if (sc == NULL) {
315 sc = vnd_spawn(unit);
316 if (sc == NULL)
317 return ENOMEM;
318 }
319
320 if ((error = vndlock(sc)) != 0)
321 return error;
322
323 if ((sc->sc_flags & VNF_CLEARING) != 0) {
324 error = ENXIO;
325 goto done;
326 }
327
328 lp = sc->sc_dkdev.dk_label;
329
330 part = DISKPART(dev);
331 pmask = (1 << part);
332
333 /*
334 * If we're initialized, check to see if there are any other
335 * open partitions. If not, then it's safe to update the
336 * in-core disklabel. Only read the disklabel if it is
337 * not already valid.
338 */
339 if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
340 sc->sc_dkdev.dk_openmask == 0)
341 vndgetdisklabel(dev, sc);
342
343 /* Check that the partitions exists. */
344 if (part != RAW_PART) {
345 if (((sc->sc_flags & VNF_INITED) == 0) ||
346 ((part >= lp->d_npartitions) ||
347 (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
348 error = ENXIO;
349 goto done;
350 }
351 }
352
353 /* Prevent our unit from being unconfigured while open. */
354 switch (mode) {
355 case S_IFCHR:
356 sc->sc_dkdev.dk_copenmask |= pmask;
357 break;
358
359 case S_IFBLK:
360 sc->sc_dkdev.dk_bopenmask |= pmask;
361 break;
362 }
363 sc->sc_dkdev.dk_openmask =
364 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
365
366 done:
367 vndunlock(sc);
368 return error;
369 }
370
371 static int
372 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
373 {
374 int unit = vndunit(dev);
375 struct vnd_softc *sc;
376 int error = 0, part;
377
378 #ifdef DEBUG
379 if (vnddebug & VDB_FOLLOW)
380 printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
381 #endif
382 sc = device_lookup_private(&vnd_cd, unit);
383 if (sc == NULL)
384 return ENXIO;
385
386 if ((error = vndlock(sc)) != 0)
387 return error;
388
389 part = DISKPART(dev);
390
391 /* ...that much closer to allowing unconfiguration... */
392 switch (mode) {
393 case S_IFCHR:
394 sc->sc_dkdev.dk_copenmask &= ~(1 << part);
395 break;
396
397 case S_IFBLK:
398 sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
399 break;
400 }
401 sc->sc_dkdev.dk_openmask =
402 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
403
404 vndunlock(sc);
405
406 if ((sc->sc_flags & VNF_INITED) == 0) {
407 if ((error = vnd_destroy(sc->sc_dev)) != 0) {
408 aprint_error_dev(sc->sc_dev,
409 "unable to detach instance\n");
410 return error;
411 }
412 }
413
414 return 0;
415 }
416
417 /*
418 * Queue the request, and wakeup the kernel thread to handle it.
419 */
420 static void
421 vndstrategy(struct buf *bp)
422 {
423 int unit = vndunit(bp->b_dev);
424 struct vnd_softc *vnd =
425 device_lookup_private(&vnd_cd, unit);
426 struct disklabel *lp;
427 daddr_t blkno;
428 int s = splbio();
429
430 if (vnd == NULL) {
431 bp->b_error = ENXIO;
432 goto done;
433 }
434 lp = vnd->sc_dkdev.dk_label;
435
436 if ((vnd->sc_flags & VNF_INITED) == 0) {
437 bp->b_error = ENXIO;
438 goto done;
439 }
440
441 /*
442 * The transfer must be a whole number of blocks.
443 */
444 if ((bp->b_bcount % lp->d_secsize) != 0) {
445 bp->b_error = EINVAL;
446 goto done;
447 }
448
449 /*
450 * check if we're read-only.
451 */
452 if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
453 bp->b_error = EACCES;
454 goto done;
455 }
456
457 /* If it's a nil transfer, wake up the top half now. */
458 if (bp->b_bcount == 0) {
459 goto done;
460 }
461
462 /*
463 * Do bounds checking and adjust transfer. If there's an error,
464 * the bounds check will flag that for us.
465 */
466 if (DISKPART(bp->b_dev) == RAW_PART) {
467 if (bounds_check_with_mediasize(bp, DEV_BSIZE,
468 vnd->sc_size) <= 0)
469 goto done;
470 } else {
471 if (bounds_check_with_label(&vnd->sc_dkdev,
472 bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
473 goto done;
474 }
475
476 /*
477 * Put the block number in terms of the logical blocksize
478 * of the "device".
479 */
480
481 blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
482
483 /*
484 * Translate the partition-relative block number to an absolute.
485 */
486 if (DISKPART(bp->b_dev) != RAW_PART) {
487 struct partition *pp;
488
489 pp = &vnd->sc_dkdev.dk_label->d_partitions[
490 DISKPART(bp->b_dev)];
491 blkno += pp->p_offset;
492 }
493 bp->b_rawblkno = blkno;
494
495 #ifdef DEBUG
496 if (vnddebug & VDB_FOLLOW)
497 printf("vndstrategy(%p): unit %d\n", bp, unit);
498 #endif
499 bufq_put(vnd->sc_tab, bp);
500 wakeup(&vnd->sc_tab);
501 splx(s);
502 return;
503
504 done:
505 bp->b_resid = bp->b_bcount;
506 biodone(bp);
507 splx(s);
508 }
509
510 static bool
511 vnode_has_strategy(struct vnd_softc *vnd)
512 {
513 return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
514 vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
515 }
516
517 /* XXX this function needs a reliable check to detect
518 * sparse files. Otherwise, bmap/strategy may be used
519 * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
520 * works on sparse files.
521 */
522 #if notyet
523 static bool
524 vnode_strategy_probe(struct vnd_softc *vnd)
525 {
526 int error;
527 daddr_t nbn;
528
529 if (!vnode_has_strategy(vnd))
530 return false;
531
532 /* Convert the first logical block number to its
533 * physical block number.
534 */
535 error = 0;
536 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
537 error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
538 VOP_UNLOCK(vnd->sc_vp);
539
540 /* Test if that worked. */
541 if (error == 0 && (long)nbn == -1)
542 return false;
543
544 return true;
545 }
546 #endif
547
548 static void
549 vndthread(void *arg)
550 {
551 struct vnd_softc *vnd = arg;
552 int s;
553
554 /* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
555 * directly access the backing vnode. If we can, use these two
556 * operations to avoid messing with the local buffer cache.
557 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
558 * which are guaranteed to work with any file system. */
559 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
560 ! vnode_has_strategy(vnd))
561 vnd->sc_flags |= VNF_USE_VN_RDWR;
562
563 #ifdef DEBUG
564 if (vnddebug & VDB_INIT)
565 printf("vndthread: vp %p, %s\n", vnd->sc_vp,
566 (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
567 "using bmap/strategy operations" :
568 "using read/write operations");
569 #endif
570
571 s = splbio();
572 vnd->sc_flags |= VNF_KTHREAD;
573 wakeup(&vnd->sc_kthread);
574
575 /*
576 * Dequeue requests and serve them depending on the available
577 * vnode operations.
578 */
579 while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
580 struct vndxfer *vnx;
581 int flags;
582 struct buf *obp;
583 struct buf *bp;
584
585 obp = bufq_get(vnd->sc_tab);
586 if (obp == NULL) {
587 tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
588 continue;
589 };
590 splx(s);
591 flags = obp->b_flags;
592 #ifdef DEBUG
593 if (vnddebug & VDB_FOLLOW)
594 printf("vndthread(%p)\n", obp);
595 #endif
596
597 if (vnd->sc_vp->v_mount == NULL) {
598 obp->b_error = ENXIO;
599 goto done;
600 }
601 #ifdef VND_COMPRESSION
602 /* handle a compressed read */
603 if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
604 off_t bn;
605
606 /* Convert to a byte offset within the file. */
607 bn = obp->b_rawblkno *
608 vnd->sc_dkdev.dk_label->d_secsize;
609
610 compstrategy(obp, bn);
611 goto done;
612 }
613 #endif /* VND_COMPRESSION */
614
615 /*
616 * Allocate a header for this transfer and link it to the
617 * buffer
618 */
619 s = splbio();
620 vnx = VND_GETXFER(vnd);
621 splx(s);
622 vnx->vx_vnd = vnd;
623
624 s = splbio();
625 while (vnd->sc_active >= vnd->sc_maxactive) {
626 tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
627 }
628 vnd->sc_active++;
629 splx(s);
630
631 /* Instrumentation. */
632 disk_busy(&vnd->sc_dkdev);
633
634 bp = &vnx->vx_buf;
635 buf_init(bp);
636 bp->b_flags = (obp->b_flags & B_READ);
637 bp->b_oflags = obp->b_oflags;
638 bp->b_cflags = obp->b_cflags;
639 bp->b_iodone = vndiodone;
640 bp->b_private = obp;
641 bp->b_vp = vnd->sc_vp;
642 bp->b_objlock = bp->b_vp->v_interlock;
643 bp->b_data = obp->b_data;
644 bp->b_bcount = obp->b_bcount;
645 BIO_COPYPRIO(bp, obp);
646
647 /* Handle the request using the appropriate operations. */
648 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
649 handle_with_strategy(vnd, obp, bp);
650 else
651 handle_with_rdwr(vnd, obp, bp);
652
653 s = splbio();
654 continue;
655
656 done:
657 biodone(obp);
658 s = splbio();
659 }
660
661 vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
662 wakeup(&vnd->sc_kthread);
663 splx(s);
664 kthread_exit(0);
665 }
666
667 /*
668 * Checks if the given vnode supports the requested operation.
669 * The operation is specified the offset returned by VOFFSET.
670 *
671 * XXX The test below used to determine this is quite fragile
672 * because it relies on the file system to use genfs to specify
673 * unimplemented operations. There might be another way to do
674 * it more cleanly.
675 */
676 static bool
677 vnode_has_op(const struct vnode *vp, int opoffset)
678 {
679 int (*defaultp)(void *);
680 int (*opp)(void *);
681
682 defaultp = vp->v_op[VOFFSET(vop_default)];
683 opp = vp->v_op[opoffset];
684
685 return opp != defaultp && opp != genfs_eopnotsupp &&
686 opp != genfs_badop && opp != genfs_nullop;
687 }
688
689 /*
690 * Handes the read/write request given in 'bp' using the vnode's VOP_READ
691 * and VOP_WRITE operations.
692 *
693 * 'obp' is a pointer to the original request fed to the vnd device.
694 */
695 static void
696 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
697 {
698 bool doread;
699 off_t offset;
700 size_t len, resid;
701 struct vnode *vp;
702
703 doread = bp->b_flags & B_READ;
704 offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
705 len = bp->b_bcount;
706 vp = vnd->sc_vp;
707
708 #if defined(DEBUG)
709 if (vnddebug & VDB_IO)
710 printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
711 ", secsize %d, offset %" PRIu64
712 ", bcount %d\n",
713 vp, doread ? "read" : "write", obp->b_rawblkno,
714 vnd->sc_dkdev.dk_label->d_secsize, offset,
715 bp->b_bcount);
716 #endif
717
718 /* Issue the read or write operation. */
719 bp->b_error =
720 vn_rdwr(doread ? UIO_READ : UIO_WRITE,
721 vp, bp->b_data, len, offset, UIO_SYSSPACE,
722 IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
723 bp->b_resid = resid;
724
725 /* Keep mapped pages below threshold. */
726 mutex_enter(vp->v_interlock);
727 if (vp->v_uobj.uo_npages > 1024*1024 / PAGE_SIZE)
728 (void) VOP_PUTPAGES(vp, 0, 0,
729 PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
730 else
731 mutex_exit(vp->v_interlock);
732
733 /* We need to increase the number of outputs on the vnode if
734 * there was any write to it. */
735 if (!doread) {
736 mutex_enter(vp->v_interlock);
737 vp->v_numoutput++;
738 mutex_exit(vp->v_interlock);
739 }
740
741 biodone(bp);
742 }
743
744 /*
745 * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
746 * and VOP_STRATEGY operations.
747 *
748 * 'obp' is a pointer to the original request fed to the vnd device.
749 */
750 static void
751 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
752 struct buf *bp)
753 {
754 int bsize, error, flags, skipped;
755 size_t resid, sz;
756 off_t bn, offset;
757 struct vnode *vp;
758
759 flags = obp->b_flags;
760
761 if (!(flags & B_READ)) {
762 vp = bp->b_vp;
763 mutex_enter(vp->v_interlock);
764 vp->v_numoutput++;
765 mutex_exit(vp->v_interlock);
766 }
767
768 /* convert to a byte offset within the file. */
769 bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
770
771 bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
772 skipped = 0;
773
774 /*
775 * Break the request into bsize pieces and feed them
776 * sequentially using VOP_BMAP/VOP_STRATEGY.
777 * We do it this way to keep from flooding NFS servers if we
778 * are connected to an NFS file. This places the burden on
779 * the client rather than the server.
780 */
781 error = 0;
782 bp->b_resid = bp->b_bcount;
783 for (offset = 0, resid = bp->b_resid; resid;
784 resid -= sz, offset += sz) {
785 struct buf *nbp;
786 daddr_t nbn;
787 int off, nra;
788
789 nra = 0;
790 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
791 error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
792 VOP_UNLOCK(vnd->sc_vp);
793
794 if (error == 0 && (long)nbn == -1)
795 error = EIO;
796
797 /*
798 * If there was an error or a hole in the file...punt.
799 * Note that we may have to wait for any operations
800 * that we have already fired off before releasing
801 * the buffer.
802 *
803 * XXX we could deal with holes here but it would be
804 * a hassle (in the write case).
805 */
806 if (error) {
807 skipped += resid;
808 break;
809 }
810
811 #ifdef DEBUG
812 if (!dovndcluster)
813 nra = 0;
814 #endif
815
816 off = bn % bsize;
817 sz = MIN(((off_t)1 + nra) * bsize - off, resid);
818 #ifdef DEBUG
819 if (vnddebug & VDB_IO)
820 printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
821 " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
822 nbn, sz);
823 #endif
824
825 nbp = getiobuf(vp, true);
826 nestiobuf_setup(bp, nbp, offset, sz);
827 nbp->b_blkno = nbn + btodb(off);
828
829 #if 0 /* XXX #ifdef DEBUG */
830 if (vnddebug & VDB_IO)
831 printf("vndstart(%ld): bp %p vp %p blkno "
832 "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
833 (long) (vnd-vnd_softc), &nbp->vb_buf,
834 nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
835 nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
836 nbp->vb_buf.b_bcount);
837 #endif
838 VOP_STRATEGY(vp, nbp);
839 bn += sz;
840 }
841 nestiobuf_done(bp, skipped, error);
842 }
843
844 static void
845 vndiodone(struct buf *bp)
846 {
847 struct vndxfer *vnx = VND_BUFTOXFER(bp);
848 struct vnd_softc *vnd = vnx->vx_vnd;
849 struct buf *obp = bp->b_private;
850 int s = splbio();
851
852 KASSERT(&vnx->vx_buf == bp);
853 KASSERT(vnd->sc_active > 0);
854 #ifdef DEBUG
855 if (vnddebug & VDB_IO) {
856 printf("vndiodone1: bp %p iodone: error %d\n",
857 bp, bp->b_error);
858 }
859 #endif
860 disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
861 (bp->b_flags & B_READ));
862 vnd->sc_active--;
863 if (vnd->sc_active == 0) {
864 wakeup(&vnd->sc_tab);
865 }
866 splx(s);
867 obp->b_error = bp->b_error;
868 obp->b_resid = bp->b_resid;
869 buf_destroy(bp);
870 VND_PUTXFER(vnd, vnx);
871 biodone(obp);
872 }
873
874 /* ARGSUSED */
875 static int
876 vndread(dev_t dev, struct uio *uio, int flags)
877 {
878 int unit = vndunit(dev);
879 struct vnd_softc *sc;
880
881 #ifdef DEBUG
882 if (vnddebug & VDB_FOLLOW)
883 printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
884 #endif
885
886 sc = device_lookup_private(&vnd_cd, unit);
887 if (sc == NULL)
888 return ENXIO;
889
890 if ((sc->sc_flags & VNF_INITED) == 0)
891 return ENXIO;
892
893 return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
894 }
895
896 /* ARGSUSED */
897 static int
898 vndwrite(dev_t dev, struct uio *uio, int flags)
899 {
900 int unit = vndunit(dev);
901 struct vnd_softc *sc;
902
903 #ifdef DEBUG
904 if (vnddebug & VDB_FOLLOW)
905 printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
906 #endif
907
908 sc = device_lookup_private(&vnd_cd, unit);
909 if (sc == NULL)
910 return ENXIO;
911
912 if ((sc->sc_flags & VNF_INITED) == 0)
913 return ENXIO;
914
915 return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
916 }
917
918 static int
919 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
920 {
921 int error;
922 struct vnd_softc *vnd;
923
924 if (*un == -1)
925 *un = unit;
926 if (*un < 0)
927 return EINVAL;
928
929 vnd = device_lookup_private(&vnd_cd, *un);
930 if (vnd == NULL)
931 return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
932
933 if ((vnd->sc_flags & VNF_INITED) == 0)
934 return -1;
935
936 vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
937 error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
938 VOP_UNLOCK(vnd->sc_vp);
939 return error;
940 }
941
942 static int
943 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
944 {
945 int error;
946
947 if ((error = vndlock(vnd)) != 0)
948 return error;
949
950 /*
951 * Don't unconfigure if any other partitions are open
952 * or if both the character and block flavors of this
953 * partition are open.
954 */
955 if (DK_BUSY(vnd, pmask) && !force) {
956 vndunlock(vnd);
957 return EBUSY;
958 }
959
960 /*
961 * XXX vndclear() might call vndclose() implicitly;
962 * release lock to avoid recursion
963 *
964 * Set VNF_CLEARING to prevent vndopen() from
965 * sneaking in after we vndunlock().
966 */
967 vnd->sc_flags |= VNF_CLEARING;
968 vndunlock(vnd);
969 vndclear(vnd, minor);
970 #ifdef DEBUG
971 if (vnddebug & VDB_INIT)
972 printf("vndioctl: CLRed\n");
973 #endif
974
975 /* Destroy the xfer and buffer pools. */
976 pool_destroy(&vnd->sc_vxpool);
977
978 /* Detach the disk. */
979 disk_detach(&vnd->sc_dkdev);
980
981 return 0;
982 }
983
984 /* ARGSUSED */
985 static int
986 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
987 {
988 bool force;
989 int unit = vndunit(dev);
990 struct vnd_softc *vnd;
991 struct vnd_ioctl *vio;
992 struct vattr vattr;
993 struct pathbuf *pb;
994 struct nameidata nd;
995 int error, part, pmask;
996 size_t geomsize;
997 int fflags;
998 #ifdef __HAVE_OLD_DISKLABEL
999 struct disklabel newlabel;
1000 #endif
1001 struct dkwedge_info *dkw;
1002 struct dkwedge_list *dkwl;
1003
1004 #ifdef DEBUG
1005 if (vnddebug & VDB_FOLLOW)
1006 printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1007 dev, cmd, data, flag, l->l_proc, unit);
1008 #endif
1009 vnd = device_lookup_private(&vnd_cd, unit);
1010 if (vnd == NULL &&
1011 #ifdef COMPAT_30
1012 cmd != VNDIOCGET30 &&
1013 #endif
1014 #ifdef COMPAT_50
1015 cmd != VNDIOCGET50 &&
1016 #endif
1017 cmd != VNDIOCGET)
1018 return ENXIO;
1019 vio = (struct vnd_ioctl *)data;
1020
1021 /* Must be open for writes for these commands... */
1022 switch (cmd) {
1023 case VNDIOCSET:
1024 case VNDIOCCLR:
1025 #ifdef COMPAT_50
1026 case VNDIOCSET50:
1027 case VNDIOCCLR50:
1028 #endif
1029 case DIOCSDINFO:
1030 case DIOCWDINFO:
1031 #ifdef __HAVE_OLD_DISKLABEL
1032 case ODIOCSDINFO:
1033 case ODIOCWDINFO:
1034 #endif
1035 case DIOCKLABEL:
1036 case DIOCWLABEL:
1037 if ((flag & FWRITE) == 0)
1038 return EBADF;
1039 }
1040
1041 /* Must be initialized for these... */
1042 switch (cmd) {
1043 case VNDIOCCLR:
1044 #ifdef VNDIOCCLR50
1045 case VNDIOCCLR50:
1046 #endif
1047 case DIOCGDINFO:
1048 case DIOCSDINFO:
1049 case DIOCWDINFO:
1050 case DIOCGPART:
1051 case DIOCKLABEL:
1052 case DIOCWLABEL:
1053 case DIOCGDEFLABEL:
1054 case DIOCCACHESYNC:
1055 #ifdef __HAVE_OLD_DISKLABEL
1056 case ODIOCGDINFO:
1057 case ODIOCSDINFO:
1058 case ODIOCWDINFO:
1059 case ODIOCGDEFLABEL:
1060 #endif
1061 if ((vnd->sc_flags & VNF_INITED) == 0)
1062 return ENXIO;
1063 }
1064
1065 switch (cmd) {
1066 #ifdef VNDIOCSET50
1067 case VNDIOCSET50:
1068 #endif
1069 case VNDIOCSET:
1070 if (vnd->sc_flags & VNF_INITED)
1071 return EBUSY;
1072
1073 if ((error = vndlock(vnd)) != 0)
1074 return error;
1075
1076 fflags = FREAD;
1077 if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1078 fflags |= FWRITE;
1079 error = pathbuf_copyin(vio->vnd_file, &pb);
1080 if (error) {
1081 goto unlock_and_exit;
1082 }
1083 NDINIT(&nd, LOOKUP, FOLLOW, pb);
1084 if ((error = vn_open(&nd, fflags, 0)) != 0) {
1085 pathbuf_destroy(pb);
1086 goto unlock_and_exit;
1087 }
1088 KASSERT(l);
1089 error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1090 if (!error && nd.ni_vp->v_type != VREG)
1091 error = EOPNOTSUPP;
1092 if (!error && vattr.va_bytes < vattr.va_size)
1093 /* File is definitely sparse, use vn_rdwr() */
1094 vnd->sc_flags |= VNF_USE_VN_RDWR;
1095 if (error) {
1096 VOP_UNLOCK(nd.ni_vp);
1097 goto close_and_exit;
1098 }
1099
1100 /* If using a compressed file, initialize its info */
1101 /* (or abort with an error if kernel has no compression) */
1102 if (vio->vnd_flags & VNF_COMP) {
1103 #ifdef VND_COMPRESSION
1104 struct vnd_comp_header *ch;
1105 int i;
1106 u_int32_t comp_size;
1107 u_int32_t comp_maxsize;
1108
1109 /* allocate space for compresed file header */
1110 ch = malloc(sizeof(struct vnd_comp_header),
1111 M_TEMP, M_WAITOK);
1112
1113 /* read compressed file header */
1114 error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1115 sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1116 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1117 if (error) {
1118 free(ch, M_TEMP);
1119 VOP_UNLOCK(nd.ni_vp);
1120 goto close_and_exit;
1121 }
1122
1123 /* save some header info */
1124 vnd->sc_comp_blksz = ntohl(ch->block_size);
1125 /* note last offset is the file byte size */
1126 vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1127 free(ch, M_TEMP);
1128 if (vnd->sc_comp_blksz == 0 ||
1129 vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1130 VOP_UNLOCK(nd.ni_vp);
1131 error = EINVAL;
1132 goto close_and_exit;
1133 }
1134 if (sizeof(struct vnd_comp_header) +
1135 sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1136 vattr.va_size) {
1137 VOP_UNLOCK(nd.ni_vp);
1138 error = EINVAL;
1139 goto close_and_exit;
1140 }
1141
1142 /* set decompressed file size */
1143 vattr.va_size =
1144 ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1145 (u_quad_t)vnd->sc_comp_blksz;
1146
1147 /* allocate space for all the compressed offsets */
1148 vnd->sc_comp_offsets =
1149 malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1150 M_DEVBUF, M_WAITOK);
1151
1152 /* read in the offsets */
1153 error = vn_rdwr(UIO_READ, nd.ni_vp,
1154 (void *)vnd->sc_comp_offsets,
1155 sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1156 sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1157 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1158 if (error) {
1159 VOP_UNLOCK(nd.ni_vp);
1160 goto close_and_exit;
1161 }
1162 /*
1163 * find largest block size (used for allocation limit).
1164 * Also convert offset to native byte order.
1165 */
1166 comp_maxsize = 0;
1167 for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1168 vnd->sc_comp_offsets[i] =
1169 be64toh(vnd->sc_comp_offsets[i]);
1170 comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1171 - vnd->sc_comp_offsets[i];
1172 if (comp_size > comp_maxsize)
1173 comp_maxsize = comp_size;
1174 }
1175 vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1176 be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1177
1178 /* create compressed data buffer */
1179 vnd->sc_comp_buff = malloc(comp_maxsize,
1180 M_DEVBUF, M_WAITOK);
1181
1182 /* create decompressed buffer */
1183 vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1184 M_DEVBUF, M_WAITOK);
1185 vnd->sc_comp_buffblk = -1;
1186
1187 /* Initialize decompress stream */
1188 memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1189 vnd->sc_comp_stream.zalloc = vnd_alloc;
1190 vnd->sc_comp_stream.zfree = vnd_free;
1191 error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1192 if (error) {
1193 if (vnd->sc_comp_stream.msg)
1194 printf("vnd%d: compressed file, %s\n",
1195 unit, vnd->sc_comp_stream.msg);
1196 VOP_UNLOCK(nd.ni_vp);
1197 error = EINVAL;
1198 goto close_and_exit;
1199 }
1200
1201 vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1202 #else /* !VND_COMPRESSION */
1203 VOP_UNLOCK(nd.ni_vp);
1204 error = EOPNOTSUPP;
1205 goto close_and_exit;
1206 #endif /* VND_COMPRESSION */
1207 }
1208
1209 VOP_UNLOCK(nd.ni_vp);
1210 vnd->sc_vp = nd.ni_vp;
1211 vnd->sc_size = btodb(vattr.va_size); /* note truncation */
1212
1213 /*
1214 * Use pseudo-geometry specified. If none was provided,
1215 * use "standard" Adaptec fictitious geometry.
1216 */
1217 if (vio->vnd_flags & VNDIOF_HASGEOM) {
1218
1219 memcpy(&vnd->sc_geom, &vio->vnd_geom,
1220 sizeof(vio->vnd_geom));
1221
1222 /*
1223 * Sanity-check the sector size.
1224 * XXX Don't allow secsize < DEV_BSIZE. Should
1225 * XXX we?
1226 */
1227 if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1228 (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1229 vnd->sc_geom.vng_ncylinders == 0 ||
1230 (vnd->sc_geom.vng_ntracks *
1231 vnd->sc_geom.vng_nsectors) == 0) {
1232 error = EINVAL;
1233 goto close_and_exit;
1234 }
1235
1236 /*
1237 * Compute the size (in DEV_BSIZE blocks) specified
1238 * by the geometry.
1239 */
1240 geomsize = (vnd->sc_geom.vng_nsectors *
1241 vnd->sc_geom.vng_ntracks *
1242 vnd->sc_geom.vng_ncylinders) *
1243 (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1244
1245 /*
1246 * Sanity-check the size against the specified
1247 * geometry.
1248 */
1249 if (vnd->sc_size < geomsize) {
1250 error = EINVAL;
1251 goto close_and_exit;
1252 }
1253 } else if (vnd->sc_size >= (32 * 64)) {
1254 /*
1255 * Size must be at least 2048 DEV_BSIZE blocks
1256 * (1M) in order to use this geometry.
1257 */
1258 vnd->sc_geom.vng_secsize = DEV_BSIZE;
1259 vnd->sc_geom.vng_nsectors = 32;
1260 vnd->sc_geom.vng_ntracks = 64;
1261 vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1262 } else {
1263 vnd->sc_geom.vng_secsize = DEV_BSIZE;
1264 vnd->sc_geom.vng_nsectors = 1;
1265 vnd->sc_geom.vng_ntracks = 1;
1266 vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1267 }
1268
1269 vnd_set_properties(vnd);
1270
1271 if (vio->vnd_flags & VNDIOF_READONLY) {
1272 vnd->sc_flags |= VNF_READONLY;
1273 }
1274
1275 if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1276 goto close_and_exit;
1277
1278 vndthrottle(vnd, vnd->sc_vp);
1279 vio->vnd_osize = dbtob(vnd->sc_size);
1280 #ifdef VNDIOCSET50
1281 if (cmd != VNDIOCSET50)
1282 #endif
1283 vio->vnd_size = dbtob(vnd->sc_size);
1284 vnd->sc_flags |= VNF_INITED;
1285
1286 /* create the kernel thread, wait for it to be up */
1287 error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1288 &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1289 if (error)
1290 goto close_and_exit;
1291 while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1292 tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1293 }
1294 #ifdef DEBUG
1295 if (vnddebug & VDB_INIT)
1296 printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1297 vnd->sc_vp, (unsigned long) vnd->sc_size,
1298 vnd->sc_geom.vng_secsize,
1299 vnd->sc_geom.vng_nsectors,
1300 vnd->sc_geom.vng_ntracks,
1301 vnd->sc_geom.vng_ncylinders);
1302 #endif
1303
1304 /* Attach the disk. */
1305 disk_attach(&vnd->sc_dkdev);
1306 disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
1307
1308 /* Initialize the xfer and buffer pools. */
1309 pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1310 0, 0, "vndxpl", NULL, IPL_BIO);
1311
1312 /* Try and read the disklabel. */
1313 vndgetdisklabel(dev, vnd);
1314
1315 vndunlock(vnd);
1316
1317 pathbuf_destroy(pb);
1318 break;
1319
1320 close_and_exit:
1321 (void) vn_close(nd.ni_vp, fflags, l->l_cred);
1322 pathbuf_destroy(pb);
1323 unlock_and_exit:
1324 #ifdef VND_COMPRESSION
1325 /* free any allocated memory (for compressed file) */
1326 if (vnd->sc_comp_offsets) {
1327 free(vnd->sc_comp_offsets, M_DEVBUF);
1328 vnd->sc_comp_offsets = NULL;
1329 }
1330 if (vnd->sc_comp_buff) {
1331 free(vnd->sc_comp_buff, M_DEVBUF);
1332 vnd->sc_comp_buff = NULL;
1333 }
1334 if (vnd->sc_comp_decombuf) {
1335 free(vnd->sc_comp_decombuf, M_DEVBUF);
1336 vnd->sc_comp_decombuf = NULL;
1337 }
1338 #endif /* VND_COMPRESSION */
1339 vndunlock(vnd);
1340 return error;
1341
1342 #ifdef VNDIOCCLR50
1343 case VNDIOCCLR50:
1344 #endif
1345 case VNDIOCCLR:
1346 part = DISKPART(dev);
1347 pmask = (1 << part);
1348 force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1349
1350 if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1351 return error;
1352
1353 break;
1354
1355 #ifdef COMPAT_30
1356 case VNDIOCGET30: {
1357 struct vnd_user30 *vnu;
1358 struct vattr va;
1359 vnu = (struct vnd_user30 *)data;
1360 KASSERT(l);
1361 switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1362 case 0:
1363 vnu->vnu_dev = va.va_fsid;
1364 vnu->vnu_ino = va.va_fileid;
1365 break;
1366 case -1:
1367 /* unused is not an error */
1368 vnu->vnu_dev = 0;
1369 vnu->vnu_ino = 0;
1370 break;
1371 default:
1372 return error;
1373 }
1374 break;
1375 }
1376 #endif
1377
1378 #ifdef COMPAT_50
1379 case VNDIOCGET50: {
1380 struct vnd_user50 *vnu;
1381 struct vattr va;
1382 vnu = (struct vnd_user50 *)data;
1383 KASSERT(l);
1384 switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1385 case 0:
1386 vnu->vnu_dev = va.va_fsid;
1387 vnu->vnu_ino = va.va_fileid;
1388 break;
1389 case -1:
1390 /* unused is not an error */
1391 vnu->vnu_dev = 0;
1392 vnu->vnu_ino = 0;
1393 break;
1394 default:
1395 return error;
1396 }
1397 break;
1398 }
1399 #endif
1400
1401 case VNDIOCGET: {
1402 struct vnd_user *vnu;
1403 struct vattr va;
1404 vnu = (struct vnd_user *)data;
1405 KASSERT(l);
1406 switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1407 case 0:
1408 vnu->vnu_dev = va.va_fsid;
1409 vnu->vnu_ino = va.va_fileid;
1410 break;
1411 case -1:
1412 /* unused is not an error */
1413 vnu->vnu_dev = 0;
1414 vnu->vnu_ino = 0;
1415 break;
1416 default:
1417 return error;
1418 }
1419 break;
1420 }
1421
1422 case DIOCGDINFO:
1423 *(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1424 break;
1425
1426 #ifdef __HAVE_OLD_DISKLABEL
1427 case ODIOCGDINFO:
1428 newlabel = *(vnd->sc_dkdev.dk_label);
1429 if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1430 return ENOTTY;
1431 memcpy(data, &newlabel, sizeof (struct olddisklabel));
1432 break;
1433 #endif
1434
1435 case DIOCGPART:
1436 ((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1437 ((struct partinfo *)data)->part =
1438 &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1439 break;
1440
1441 case DIOCWDINFO:
1442 case DIOCSDINFO:
1443 #ifdef __HAVE_OLD_DISKLABEL
1444 case ODIOCWDINFO:
1445 case ODIOCSDINFO:
1446 #endif
1447 {
1448 struct disklabel *lp;
1449
1450 if ((error = vndlock(vnd)) != 0)
1451 return error;
1452
1453 vnd->sc_flags |= VNF_LABELLING;
1454
1455 #ifdef __HAVE_OLD_DISKLABEL
1456 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1457 memset(&newlabel, 0, sizeof newlabel);
1458 memcpy(&newlabel, data, sizeof (struct olddisklabel));
1459 lp = &newlabel;
1460 } else
1461 #endif
1462 lp = (struct disklabel *)data;
1463
1464 error = setdisklabel(vnd->sc_dkdev.dk_label,
1465 lp, 0, vnd->sc_dkdev.dk_cpulabel);
1466 if (error == 0) {
1467 if (cmd == DIOCWDINFO
1468 #ifdef __HAVE_OLD_DISKLABEL
1469 || cmd == ODIOCWDINFO
1470 #endif
1471 )
1472 error = writedisklabel(VNDLABELDEV(dev),
1473 vndstrategy, vnd->sc_dkdev.dk_label,
1474 vnd->sc_dkdev.dk_cpulabel);
1475 }
1476
1477 vnd->sc_flags &= ~VNF_LABELLING;
1478
1479 vndunlock(vnd);
1480
1481 if (error)
1482 return error;
1483 break;
1484 }
1485
1486 case DIOCKLABEL:
1487 if (*(int *)data != 0)
1488 vnd->sc_flags |= VNF_KLABEL;
1489 else
1490 vnd->sc_flags &= ~VNF_KLABEL;
1491 break;
1492
1493 case DIOCWLABEL:
1494 if (*(int *)data != 0)
1495 vnd->sc_flags |= VNF_WLABEL;
1496 else
1497 vnd->sc_flags &= ~VNF_WLABEL;
1498 break;
1499
1500 case DIOCGDEFLABEL:
1501 vndgetdefaultlabel(vnd, (struct disklabel *)data);
1502 break;
1503
1504 #ifdef __HAVE_OLD_DISKLABEL
1505 case ODIOCGDEFLABEL:
1506 vndgetdefaultlabel(vnd, &newlabel);
1507 if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1508 return ENOTTY;
1509 memcpy(data, &newlabel, sizeof (struct olddisklabel));
1510 break;
1511 #endif
1512
1513 case DIOCCACHESYNC:
1514 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1515 error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1516 FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1517 VOP_UNLOCK(vnd->sc_vp);
1518 return error;
1519
1520 case DIOCAWEDGE:
1521 dkw = (void *) data;
1522
1523 if ((flag & FWRITE) == 0)
1524 return EBADF;
1525
1526 /* If the ioctl happens here, the parent is us. */
1527 strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1528 sizeof(dkw->dkw_parent));
1529 return dkwedge_add(dkw);
1530
1531 case DIOCDWEDGE:
1532 dkw = (void *) data;
1533
1534 if ((flag & FWRITE) == 0)
1535 return EBADF;
1536
1537 /* If the ioctl happens here, the parent is us. */
1538 strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1539 sizeof(dkw->dkw_parent));
1540 return dkwedge_del(dkw);
1541
1542 case DIOCLWEDGES:
1543 dkwl = (void *) data;
1544
1545 return dkwedge_list(&vnd->sc_dkdev, dkwl, l);
1546
1547 default:
1548 return ENOTTY;
1549 }
1550
1551 return 0;
1552 }
1553
1554 /*
1555 * Duplicate the current processes' credentials. Since we are called only
1556 * as the result of a SET ioctl and only root can do that, any future access
1557 * to this "disk" is essentially as root. Note that credentials may change
1558 * if some other uid can write directly to the mapped file (NFS).
1559 */
1560 static int
1561 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1562 {
1563 struct uio auio;
1564 struct iovec aiov;
1565 char *tmpbuf;
1566 int error;
1567
1568 vnd->sc_cred = kauth_cred_dup(cred);
1569 tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1570
1571 /* XXX: Horrible kludge to establish credentials for NFS */
1572 aiov.iov_base = tmpbuf;
1573 aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1574 auio.uio_iov = &aiov;
1575 auio.uio_iovcnt = 1;
1576 auio.uio_offset = 0;
1577 auio.uio_rw = UIO_READ;
1578 auio.uio_resid = aiov.iov_len;
1579 UIO_SETUP_SYSSPACE(&auio);
1580 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1581 error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1582 if (error == 0) {
1583 /*
1584 * Because vnd does all IO directly through the vnode
1585 * we need to flush (at least) the buffer from the above
1586 * VOP_READ from the buffer cache to prevent cache
1587 * incoherencies. Also, be careful to write dirty
1588 * buffers back to stable storage.
1589 */
1590 error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1591 curlwp, 0, 0);
1592 }
1593 VOP_UNLOCK(vnd->sc_vp);
1594
1595 free(tmpbuf, M_TEMP);
1596 return error;
1597 }
1598
1599 /*
1600 * Set maxactive based on FS type
1601 */
1602 static void
1603 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1604 {
1605
1606 if (vp->v_tag == VT_NFS)
1607 vnd->sc_maxactive = 2;
1608 else
1609 vnd->sc_maxactive = 8;
1610
1611 if (vnd->sc_maxactive < 1)
1612 vnd->sc_maxactive = 1;
1613 }
1614
1615 #if 0
1616 static void
1617 vndshutdown(void)
1618 {
1619 struct vnd_softc *vnd;
1620
1621 for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1622 if (vnd->sc_flags & VNF_INITED)
1623 vndclear(vnd);
1624 }
1625 #endif
1626
1627 static void
1628 vndclear(struct vnd_softc *vnd, int myminor)
1629 {
1630 struct vnode *vp = vnd->sc_vp;
1631 int fflags = FREAD;
1632 int bmaj, cmaj, i, mn;
1633 int s;
1634
1635 #ifdef DEBUG
1636 if (vnddebug & VDB_FOLLOW)
1637 printf("vndclear(%p): vp %p\n", vnd, vp);
1638 #endif
1639 /* locate the major number */
1640 bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1641 cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1642
1643 /* Nuke the vnodes for any open instances */
1644 for (i = 0; i < MAXPARTITIONS; i++) {
1645 mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1646 vdevgone(bmaj, mn, mn, VBLK);
1647 if (mn != myminor) /* XXX avoid to kill own vnode */
1648 vdevgone(cmaj, mn, mn, VCHR);
1649 }
1650
1651 if ((vnd->sc_flags & VNF_READONLY) == 0)
1652 fflags |= FWRITE;
1653
1654 s = splbio();
1655 bufq_drain(vnd->sc_tab);
1656 splx(s);
1657
1658 vnd->sc_flags |= VNF_VUNCONF;
1659 wakeup(&vnd->sc_tab);
1660 while (vnd->sc_flags & VNF_KTHREAD)
1661 tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1662
1663 #ifdef VND_COMPRESSION
1664 /* free the compressed file buffers */
1665 if (vnd->sc_flags & VNF_COMP) {
1666 if (vnd->sc_comp_offsets) {
1667 free(vnd->sc_comp_offsets, M_DEVBUF);
1668 vnd->sc_comp_offsets = NULL;
1669 }
1670 if (vnd->sc_comp_buff) {
1671 free(vnd->sc_comp_buff, M_DEVBUF);
1672 vnd->sc_comp_buff = NULL;
1673 }
1674 if (vnd->sc_comp_decombuf) {
1675 free(vnd->sc_comp_decombuf, M_DEVBUF);
1676 vnd->sc_comp_decombuf = NULL;
1677 }
1678 }
1679 #endif /* VND_COMPRESSION */
1680 vnd->sc_flags &=
1681 ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1682 | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1683 if (vp == NULL)
1684 panic("vndclear: null vp");
1685 (void) vn_close(vp, fflags, vnd->sc_cred);
1686 kauth_cred_free(vnd->sc_cred);
1687 vnd->sc_vp = NULL;
1688 vnd->sc_cred = NULL;
1689 vnd->sc_size = 0;
1690 }
1691
1692 static int
1693 vndsize(dev_t dev)
1694 {
1695 struct vnd_softc *sc;
1696 struct disklabel *lp;
1697 int part, unit, omask;
1698 int size;
1699
1700 unit = vndunit(dev);
1701 sc = device_lookup_private(&vnd_cd, unit);
1702 if (sc == NULL)
1703 return -1;
1704
1705 if ((sc->sc_flags & VNF_INITED) == 0)
1706 return -1;
1707
1708 part = DISKPART(dev);
1709 omask = sc->sc_dkdev.dk_openmask & (1 << part);
1710 lp = sc->sc_dkdev.dk_label;
1711
1712 if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp)) /* XXX */
1713 return -1;
1714
1715 if (lp->d_partitions[part].p_fstype != FS_SWAP)
1716 size = -1;
1717 else
1718 size = lp->d_partitions[part].p_size *
1719 (lp->d_secsize / DEV_BSIZE);
1720
1721 if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp)) /* XXX */
1722 return -1;
1723
1724 return size;
1725 }
1726
1727 static int
1728 vnddump(dev_t dev, daddr_t blkno, void *va,
1729 size_t size)
1730 {
1731
1732 /* Not implemented. */
1733 return ENXIO;
1734 }
1735
1736 static void
1737 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1738 {
1739 struct vndgeom *vng = &sc->sc_geom;
1740 struct partition *pp;
1741
1742 memset(lp, 0, sizeof(*lp));
1743
1744 lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1745 lp->d_secsize = vng->vng_secsize;
1746 lp->d_nsectors = vng->vng_nsectors;
1747 lp->d_ntracks = vng->vng_ntracks;
1748 lp->d_ncylinders = vng->vng_ncylinders;
1749 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1750
1751 strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1752 lp->d_type = DTYPE_VND;
1753 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1754 lp->d_rpm = 3600;
1755 lp->d_interleave = 1;
1756 lp->d_flags = 0;
1757
1758 pp = &lp->d_partitions[RAW_PART];
1759 pp->p_offset = 0;
1760 pp->p_size = lp->d_secperunit;
1761 pp->p_fstype = FS_UNUSED;
1762 lp->d_npartitions = RAW_PART + 1;
1763
1764 lp->d_magic = DISKMAGIC;
1765 lp->d_magic2 = DISKMAGIC;
1766 lp->d_checksum = dkcksum(lp);
1767 }
1768
1769 /*
1770 * Read the disklabel from a vnd. If one is not present, create a fake one.
1771 */
1772 static void
1773 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1774 {
1775 const char *errstring;
1776 struct disklabel *lp = sc->sc_dkdev.dk_label;
1777 struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1778 int i;
1779
1780 memset(clp, 0, sizeof(*clp));
1781
1782 vndgetdefaultlabel(sc, lp);
1783
1784 /*
1785 * Call the generic disklabel extraction routine.
1786 */
1787 errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1788 if (errstring) {
1789 /*
1790 * Lack of disklabel is common, but we print the warning
1791 * anyway, since it might contain other useful information.
1792 */
1793 aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1794
1795 /*
1796 * For historical reasons, if there's no disklabel
1797 * present, all partitions must be FS_BSDFFS and
1798 * occupy the entire disk.
1799 */
1800 for (i = 0; i < MAXPARTITIONS; i++) {
1801 /*
1802 * Don't wipe out port specific hack (such as
1803 * dos partition hack of i386 port).
1804 */
1805 if (lp->d_partitions[i].p_size != 0)
1806 continue;
1807
1808 lp->d_partitions[i].p_size = lp->d_secperunit;
1809 lp->d_partitions[i].p_offset = 0;
1810 lp->d_partitions[i].p_fstype = FS_BSDFFS;
1811 }
1812
1813 strncpy(lp->d_packname, "default label",
1814 sizeof(lp->d_packname));
1815
1816 lp->d_npartitions = MAXPARTITIONS;
1817 lp->d_checksum = dkcksum(lp);
1818 }
1819
1820 /* In-core label now valid. */
1821 sc->sc_flags |= VNF_VLABEL;
1822 }
1823
1824 /*
1825 * Wait interruptibly for an exclusive lock.
1826 *
1827 * XXX
1828 * Several drivers do this; it should be abstracted and made MP-safe.
1829 */
1830 static int
1831 vndlock(struct vnd_softc *sc)
1832 {
1833 int error;
1834
1835 while ((sc->sc_flags & VNF_LOCKED) != 0) {
1836 sc->sc_flags |= VNF_WANTED;
1837 if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1838 return error;
1839 }
1840 sc->sc_flags |= VNF_LOCKED;
1841 return 0;
1842 }
1843
1844 /*
1845 * Unlock and wake up any waiters.
1846 */
1847 static void
1848 vndunlock(struct vnd_softc *sc)
1849 {
1850
1851 sc->sc_flags &= ~VNF_LOCKED;
1852 if ((sc->sc_flags & VNF_WANTED) != 0) {
1853 sc->sc_flags &= ~VNF_WANTED;
1854 wakeup(sc);
1855 }
1856 }
1857
1858 #ifdef VND_COMPRESSION
1859 /* compressed file read */
1860 static void
1861 compstrategy(struct buf *bp, off_t bn)
1862 {
1863 int error;
1864 int unit = vndunit(bp->b_dev);
1865 struct vnd_softc *vnd =
1866 device_lookup_private(&vnd_cd, unit);
1867 u_int32_t comp_block;
1868 struct uio auio;
1869 char *addr;
1870 int s;
1871
1872 /* set up constants for data move */
1873 auio.uio_rw = UIO_READ;
1874 UIO_SETUP_SYSSPACE(&auio);
1875
1876 /* read, and transfer the data */
1877 addr = bp->b_data;
1878 bp->b_resid = bp->b_bcount;
1879 s = splbio();
1880 while (bp->b_resid > 0) {
1881 unsigned length;
1882 size_t length_in_buffer;
1883 u_int32_t offset_in_buffer;
1884 struct iovec aiov;
1885
1886 /* calculate the compressed block number */
1887 comp_block = bn / (off_t)vnd->sc_comp_blksz;
1888
1889 /* check for good block number */
1890 if (comp_block >= vnd->sc_comp_numoffs) {
1891 bp->b_error = EINVAL;
1892 splx(s);
1893 return;
1894 }
1895
1896 /* read in the compressed block, if not in buffer */
1897 if (comp_block != vnd->sc_comp_buffblk) {
1898 length = vnd->sc_comp_offsets[comp_block + 1] -
1899 vnd->sc_comp_offsets[comp_block];
1900 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1901 error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1902 length, vnd->sc_comp_offsets[comp_block],
1903 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1904 NULL, NULL);
1905 if (error) {
1906 bp->b_error = error;
1907 VOP_UNLOCK(vnd->sc_vp);
1908 splx(s);
1909 return;
1910 }
1911 /* uncompress the buffer */
1912 vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1913 vnd->sc_comp_stream.avail_in = length;
1914 vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1915 vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1916 inflateReset(&vnd->sc_comp_stream);
1917 error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1918 if (error != Z_STREAM_END) {
1919 if (vnd->sc_comp_stream.msg)
1920 aprint_normal_dev(vnd->sc_dev,
1921 "compressed file, %s\n",
1922 vnd->sc_comp_stream.msg);
1923 bp->b_error = EBADMSG;
1924 VOP_UNLOCK(vnd->sc_vp);
1925 splx(s);
1926 return;
1927 }
1928 vnd->sc_comp_buffblk = comp_block;
1929 VOP_UNLOCK(vnd->sc_vp);
1930 }
1931
1932 /* transfer the usable uncompressed data */
1933 offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1934 length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1935 if (length_in_buffer > bp->b_resid)
1936 length_in_buffer = bp->b_resid;
1937 auio.uio_iov = &aiov;
1938 auio.uio_iovcnt = 1;
1939 aiov.iov_base = addr;
1940 aiov.iov_len = length_in_buffer;
1941 auio.uio_resid = aiov.iov_len;
1942 auio.uio_offset = 0;
1943 error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1944 length_in_buffer, &auio);
1945 if (error) {
1946 bp->b_error = error;
1947 splx(s);
1948 return;
1949 }
1950
1951 bn += length_in_buffer;
1952 addr += length_in_buffer;
1953 bp->b_resid -= length_in_buffer;
1954 }
1955 splx(s);
1956 }
1957
1958 /* compression memory allocation routines */
1959 static void *
1960 vnd_alloc(void *aux, u_int items, u_int siz)
1961 {
1962 return malloc(items * siz, M_TEMP, M_NOWAIT);
1963 }
1964
1965 static void
1966 vnd_free(void *aux, void *ptr)
1967 {
1968 free(ptr, M_TEMP);
1969 }
1970 #endif /* VND_COMPRESSION */
1971
1972 static void
1973 vnd_set_properties(struct vnd_softc *vnd)
1974 {
1975 prop_dictionary_t disk_info, odisk_info, geom;
1976
1977 disk_info = prop_dictionary_create();
1978
1979 geom = prop_dictionary_create();
1980
1981 prop_dictionary_set_uint64(geom, "sectors-per-unit",
1982 vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1983 vnd->sc_geom.vng_ncylinders);
1984
1985 prop_dictionary_set_uint32(geom, "sector-size",
1986 vnd->sc_geom.vng_secsize);
1987
1988 prop_dictionary_set_uint16(geom, "sectors-per-track",
1989 vnd->sc_geom.vng_nsectors);
1990
1991 prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1992 vnd->sc_geom.vng_ntracks);
1993
1994 prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1995 vnd->sc_geom.vng_ncylinders);
1996
1997 prop_dictionary_set(disk_info, "geometry", geom);
1998 prop_object_release(geom);
1999
2000 prop_dictionary_set(device_properties(vnd->sc_dev),
2001 "disk-info", disk_info);
2002
2003 /*
2004 * Don't release disk_info here; we keep a reference to it.
2005 * disk_detach() will release it when we go away.
2006 */
2007
2008 odisk_info = vnd->sc_dkdev.dk_info;
2009 vnd->sc_dkdev.dk_info = disk_info;
2010 if (odisk_info)
2011 prop_object_release(odisk_info);
2012 }
2013
2014 #ifdef _MODULE
2015
2016 #include <sys/module.h>
2017
2018 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
2019 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2020
2021 static int
2022 vnd_modcmd(modcmd_t cmd, void *arg)
2023 {
2024 int bmajor = -1, cmajor = -1, error = 0;
2025
2026 switch (cmd) {
2027 case MODULE_CMD_INIT:
2028 error = config_cfdriver_attach(&vnd_cd);
2029 if (error)
2030 break;
2031
2032 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2033 if (error) {
2034 config_cfdriver_detach(&vnd_cd);
2035 aprint_error("%s: unable to register cfattach\n",
2036 vnd_cd.cd_name);
2037 break;
2038 }
2039
2040 error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2041 &vnd_cdevsw, &cmajor);
2042 if (error) {
2043 config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2044 config_cfdriver_detach(&vnd_cd);
2045 break;
2046 }
2047
2048 break;
2049
2050 case MODULE_CMD_FINI:
2051 error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2052 if (error)
2053 break;
2054 config_cfdriver_detach(&vnd_cd);
2055 devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2056 break;
2057
2058 case MODULE_CMD_STAT:
2059 return ENOTTY;
2060
2061 default:
2062 return ENOTTY;
2063 }
2064
2065 return error;
2066 }
2067
2068 #endif
2069