FreeBSD/Linux Kernel Cross Reference
sys/dev/vnd.c
1 /* $NetBSD: vnd.c,v 1.160.2.2 2008/04/08 20:19:27 jdc Exp $ */
2
3 /*-
4 * Copyright (c) 1996, 1997, 1998 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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 /*
40 * Copyright (c) 1990, 1993
41 * The Regents of the University of California. All rights reserved.
42 *
43 * This code is derived from software contributed to Berkeley by
44 * the Systems Programming Group of the University of Utah Computer
45 * Science Department.
46 *
47 * Redistribution and use in source and binary forms, with or without
48 * modification, are permitted provided that the following conditions
49 * are met:
50 * 1. Redistributions of source code must retain the above copyright
51 * notice, this list of conditions and the following disclaimer.
52 * 2. Redistributions in binary form must reproduce the above copyright
53 * notice, this list of conditions and the following disclaimer in the
54 * documentation and/or other materials provided with the distribution.
55 * 3. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
58 *
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69 * SUCH DAMAGE.
70 *
71 * from: Utah $Hdr: vn.c 1.13 94/04/02$
72 *
73 * @(#)vn.c 8.9 (Berkeley) 5/14/95
74 */
75
76 /*
77 * Copyright (c) 1988 University of Utah.
78 *
79 * This code is derived from software contributed to Berkeley by
80 * the Systems Programming Group of the University of Utah Computer
81 * Science Department.
82 *
83 * Redistribution and use in source and binary forms, with or without
84 * modification, are permitted provided that the following conditions
85 * are met:
86 * 1. Redistributions of source code must retain the above copyright
87 * notice, this list of conditions and the following disclaimer.
88 * 2. Redistributions in binary form must reproduce the above copyright
89 * notice, this list of conditions and the following disclaimer in the
90 * documentation and/or other materials provided with the distribution.
91 * 3. All advertising materials mentioning features or use of this software
92 * must display the following acknowledgement:
93 * This product includes software developed by the University of
94 * California, Berkeley and its contributors.
95 * 4. Neither the name of the University nor the names of its contributors
96 * may be used to endorse or promote products derived from this software
97 * without specific prior written permission.
98 *
99 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
100 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
101 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
102 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
103 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
104 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
105 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
106 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
107 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
108 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
109 * SUCH DAMAGE.
110 *
111 * from: Utah $Hdr: vn.c 1.13 94/04/02$
112 *
113 * @(#)vn.c 8.9 (Berkeley) 5/14/95
114 */
115
116 /*
117 * Vnode disk driver.
118 *
119 * Block/character interface to a vnode. Allows one to treat a file
120 * as a disk (e.g. build a filesystem in it, mount it, etc.).
121 *
122 * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
123 * this uses them to avoid distorting the local buffer cache. If those
124 * block-level operations are not available, this falls back to the regular
125 * read and write calls. Using these may distort the cache in some cases
126 * but better have the driver working than preventing it to work on file
127 * systems where the block-level operations are not implemented for
128 * whatever reason.
129 *
130 * NOTE 2: There is a security issue involved with this driver.
131 * Once mounted all access to the contents of the "mapped" file via
132 * the special file is controlled by the permissions on the special
133 * file, the protection of the mapped file is ignored (effectively,
134 * by using root credentials in all transactions).
135 *
136 * NOTE 3: Doesn't interact with leases, should it?
137 */
138
139 #include <sys/cdefs.h>
140 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.160.2.2 2008/04/08 20:19:27 jdc Exp $");
141
142 #if defined(_KERNEL_OPT)
143 #include "fs_nfs.h"
144 #include "opt_vnd.h"
145 #endif
146
147 #include <sys/param.h>
148 #include <sys/systm.h>
149 #include <sys/namei.h>
150 #include <sys/proc.h>
151 #include <sys/kthread.h>
152 #include <sys/errno.h>
153 #include <sys/buf.h>
154 #include <sys/bufq.h>
155 #include <sys/malloc.h>
156 #include <sys/ioctl.h>
157 #include <sys/disklabel.h>
158 #include <sys/device.h>
159 #include <sys/disk.h>
160 #include <sys/stat.h>
161 #include <sys/mount.h>
162 #include <sys/vnode.h>
163 #include <sys/file.h>
164 #include <sys/uio.h>
165 #include <sys/conf.h>
166 #include <sys/kauth.h>
167
168 #include <net/zlib.h>
169
170 #include <miscfs/genfs/genfs.h>
171 #include <miscfs/specfs/specdev.h>
172
173 #include <dev/vndvar.h>
174
175 #include <prop/proplib.h>
176
177 #if defined(VNDDEBUG) && !defined(DEBUG)
178 #define DEBUG
179 #endif
180
181 #ifdef DEBUG
182 int dovndcluster = 1;
183 #define VDB_FOLLOW 0x01
184 #define VDB_INIT 0x02
185 #define VDB_IO 0x04
186 #define VDB_LABEL 0x08
187 int vnddebug = 0x00;
188 #endif
189
190 #define vndunit(x) DISKUNIT(x)
191
192 struct vndxfer {
193 struct buf vx_buf;
194 struct vnd_softc *vx_vnd;
195 };
196 #define VND_BUFTOXFER(bp) ((struct vndxfer *)(void *)bp)
197
198 #define VND_GETXFER(vnd) pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
199 #define VND_PUTXFER(vnd, vx) pool_put(&(vnd)->sc_vxpool, (vx))
200
201 #define VNDLABELDEV(dev) \
202 (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
203
204 /* called by main() at boot time (XXX: and the LKM driver) */
205 void vndattach(int);
206
207 static void vndclear(struct vnd_softc *, int);
208 static int vndsetcred(struct vnd_softc *, kauth_cred_t);
209 static void vndthrottle(struct vnd_softc *, struct vnode *);
210 static void vndiodone(struct buf *);
211 #if 0
212 static void vndshutdown(void);
213 #endif
214
215 static void vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
216 static void vndgetdisklabel(dev_t, struct vnd_softc *);
217
218 static int vndlock(struct vnd_softc *);
219 static void vndunlock(struct vnd_softc *);
220 #ifdef VND_COMPRESSION
221 static void compstrategy(struct buf *, off_t);
222 static void *vnd_alloc(void *, u_int, u_int);
223 static void vnd_free(void *, void *);
224 #endif /* VND_COMPRESSION */
225
226 static void vndthread(void *);
227 static boolean_t vnode_has_op(const struct vnode *, int);
228 static void handle_with_rdwr(struct vnd_softc *, const struct buf *,
229 struct buf *);
230 static void handle_with_strategy(struct vnd_softc *, const struct buf *,
231 struct buf *);
232 static void vnd_set_properties(struct vnd_softc *);
233
234 static dev_type_open(vndopen);
235 static dev_type_close(vndclose);
236 static dev_type_read(vndread);
237 static dev_type_write(vndwrite);
238 static dev_type_ioctl(vndioctl);
239 static dev_type_strategy(vndstrategy);
240 static dev_type_dump(vnddump);
241 static dev_type_size(vndsize);
242
243 const struct bdevsw vnd_bdevsw = {
244 vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
245 };
246
247 const struct cdevsw vnd_cdevsw = {
248 vndopen, vndclose, vndread, vndwrite, vndioctl,
249 nostop, notty, nopoll, nommap, nokqfilter, D_DISK
250 };
251
252 static int vnd_match(struct device *, struct cfdata *, void *);
253 static void vnd_attach(struct device *, struct device *, void *);
254 static int vnd_detach(struct device *, int);
255
256 CFATTACH_DECL(vnd, sizeof(struct vnd_softc),
257 vnd_match, vnd_attach, vnd_detach, NULL);
258 extern struct cfdriver vnd_cd;
259
260 static struct vnd_softc *vnd_spawn(int);
261 int vnd_destroy(struct device *);
262
263 void
264 vndattach(int num)
265 {
266 int error;
267
268 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
269 if (error)
270 aprint_error("%s: unable to register cfattach\n",
271 vnd_cd.cd_name);
272 }
273
274 static int
275 vnd_match(struct device *self, struct cfdata *cfdata,
276 void *aux)
277 {
278 return 1;
279 }
280
281 static void
282 vnd_attach(struct device *parent, struct device *self,
283 void *aux)
284 {
285 struct vnd_softc *sc = (struct vnd_softc *)self;
286
287 sc->sc_comp_offsets = NULL;
288 sc->sc_comp_buff = NULL;
289 sc->sc_comp_decombuf = NULL;
290 bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
291 pseudo_disk_init(&sc->sc_dkdev);
292 }
293
294 static int
295 vnd_detach(struct device *self, int flags)
296 {
297 struct vnd_softc *sc = (struct vnd_softc *)self;
298 if (sc->sc_flags & VNF_INITED)
299 return EBUSY;
300
301 bufq_free(sc->sc_tab);
302
303 return 0;
304 }
305
306 static struct vnd_softc *
307 vnd_spawn(int unit)
308 {
309 struct cfdata *cf;
310
311 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
312 cf->cf_name = vnd_cd.cd_name;
313 cf->cf_atname = vnd_cd.cd_name;
314 cf->cf_unit = unit;
315 cf->cf_fstate = FSTATE_STAR;
316
317 return (struct vnd_softc *)config_attach_pseudo(cf);
318 }
319
320 int
321 vnd_destroy(struct device *dev)
322 {
323 int error;
324 struct cfdata *cf;
325
326 cf = device_cfdata(dev);
327 error = config_detach(dev, DETACH_QUIET);
328 if (error)
329 return error;
330 free(cf, M_DEVBUF);
331 return 0;
332 }
333
334 static int
335 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
336 {
337 int unit = vndunit(dev);
338 struct vnd_softc *sc;
339 int error = 0, part, pmask;
340 struct disklabel *lp;
341
342 #ifdef DEBUG
343 if (vnddebug & VDB_FOLLOW)
344 printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
345 #endif
346 sc = device_lookup(&vnd_cd, unit);
347 if (sc == NULL) {
348 sc = vnd_spawn(unit);
349 if (sc == NULL)
350 return ENOMEM;
351 }
352
353 if ((error = vndlock(sc)) != 0)
354 return (error);
355
356 lp = sc->sc_dkdev.dk_label;
357
358 part = DISKPART(dev);
359 pmask = (1 << part);
360
361 /*
362 * If we're initialized, check to see if there are any other
363 * open partitions. If not, then it's safe to update the
364 * in-core disklabel. Only read the disklabel if it is
365 * not already valid.
366 */
367 if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
368 sc->sc_dkdev.dk_openmask == 0)
369 vndgetdisklabel(dev, sc);
370
371 /* Check that the partitions exists. */
372 if (part != RAW_PART) {
373 if (((sc->sc_flags & VNF_INITED) == 0) ||
374 ((part >= lp->d_npartitions) ||
375 (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
376 error = ENXIO;
377 goto done;
378 }
379 }
380
381 /* Prevent our unit from being unconfigured while open. */
382 switch (mode) {
383 case S_IFCHR:
384 sc->sc_dkdev.dk_copenmask |= pmask;
385 break;
386
387 case S_IFBLK:
388 sc->sc_dkdev.dk_bopenmask |= pmask;
389 break;
390 }
391 sc->sc_dkdev.dk_openmask =
392 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
393
394 done:
395 vndunlock(sc);
396 return (error);
397 }
398
399 static int
400 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
401 {
402 int unit = vndunit(dev);
403 struct vnd_softc *sc;
404 int error = 0, part;
405
406 #ifdef DEBUG
407 if (vnddebug & VDB_FOLLOW)
408 printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
409 #endif
410 sc = device_lookup(&vnd_cd, unit);
411 if (sc == NULL)
412 return ENXIO;
413
414 if ((error = vndlock(sc)) != 0)
415 return (error);
416
417 part = DISKPART(dev);
418
419 /* ...that much closer to allowing unconfiguration... */
420 switch (mode) {
421 case S_IFCHR:
422 sc->sc_dkdev.dk_copenmask &= ~(1 << part);
423 break;
424
425 case S_IFBLK:
426 sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
427 break;
428 }
429 sc->sc_dkdev.dk_openmask =
430 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
431
432 vndunlock(sc);
433
434 if ((sc->sc_flags & VNF_INITED) == 0) {
435 if ((error = vnd_destroy((struct device *)sc)) != 0) {
436 aprint_error("%s: unable to detach instance\n",
437 sc->sc_dev.dv_xname);
438 return error;
439 }
440 }
441
442 return (0);
443 }
444
445 /*
446 * Queue the request, and wakeup the kernel thread to handle it.
447 */
448 static void
449 vndstrategy(struct buf *bp)
450 {
451 int unit = vndunit(bp->b_dev);
452 struct vnd_softc *vnd =
453 (struct vnd_softc *)device_lookup(&vnd_cd, unit);
454 struct disklabel *lp = vnd->sc_dkdev.dk_label;
455 daddr_t blkno;
456 int s = splbio();
457
458 if ((vnd->sc_flags & VNF_INITED) == 0) {
459 bp->b_error = ENXIO;
460 goto bad;
461 }
462
463 /*
464 * The transfer must be a whole number of blocks.
465 */
466 if ((bp->b_bcount % lp->d_secsize) != 0) {
467 bp->b_error = EINVAL;
468 goto bad;
469 }
470
471 /*
472 * check if we're read-only.
473 */
474 if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
475 bp->b_error = EACCES;
476 bp->b_flags |= B_ERROR;
477 goto done;
478 }
479
480 /*
481 * Do bounds checking and adjust transfer. If there's an error,
482 * the bounds check will flag that for us.
483 */
484 if (DISKPART(bp->b_dev) == RAW_PART) {
485 if (bounds_check_with_mediasize(bp, DEV_BSIZE,
486 vnd->sc_size) <= 0)
487 goto done;
488 } else {
489 if (bounds_check_with_label(&vnd->sc_dkdev,
490 bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
491 goto done;
492 }
493
494 /*
495 * Put the block number in terms of the logical blocksize
496 * of the "device".
497 */
498
499 blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
500
501 /*
502 * Translate the partition-relative block number to an absolute.
503 */
504 if (DISKPART(bp->b_dev) != RAW_PART) {
505 struct partition *pp;
506
507 pp = &vnd->sc_dkdev.dk_label->d_partitions[
508 DISKPART(bp->b_dev)];
509 blkno += pp->p_offset;
510 }
511 bp->b_rawblkno = blkno;
512
513 #ifdef DEBUG
514 if (vnddebug & VDB_FOLLOW)
515 printf("vndstrategy(%p): unit %d\n", bp, unit);
516 #endif
517 BUFQ_PUT(vnd->sc_tab, bp);
518 wakeup(&vnd->sc_tab);
519 splx(s);
520 return;
521 bad:
522 bp->b_flags |= B_ERROR;
523 done:
524 bp->b_resid = bp->b_bcount;
525 biodone(bp);
526 splx(s);
527 }
528
529 static void
530 vndthread(void *arg)
531 {
532 struct vnd_softc *vnd = arg;
533 boolean_t usestrategy;
534 int s;
535
536 /* Determine whether we can use VOP_BMAP and VOP_STRATEGY to
537 * directly access the backing vnode. If we can, use these two
538 * operations to avoid messing with the local buffer cache.
539 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
540 * which are guaranteed to work with any file system. */
541 usestrategy = vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
542 vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
543
544 #ifdef DEBUG
545 if (vnddebug & VDB_INIT)
546 printf("vndthread: vp %p, %s\n", vnd->sc_vp,
547 usestrategy ?
548 "using bmap/strategy operations" :
549 "using read/write operations");
550 #endif
551
552 s = splbio();
553 vnd->sc_flags |= VNF_KTHREAD;
554 wakeup(&vnd->sc_kthread);
555
556 /*
557 * Dequeue requests and serve them depending on the available
558 * vnode operations.
559 */
560 while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
561 struct vndxfer *vnx;
562 int flags;
563 struct buf *obp;
564 struct buf *bp;
565
566 obp = BUFQ_GET(vnd->sc_tab);
567 if (obp == NULL) {
568 tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
569 continue;
570 };
571 splx(s);
572 flags = obp->b_flags;
573 #ifdef DEBUG
574 if (vnddebug & VDB_FOLLOW)
575 printf("vndthread(%p\n", obp);
576 #endif
577
578 if (vnd->sc_vp->v_mount == NULL) {
579 obp->b_error = ENXIO;
580 obp->b_flags |= B_ERROR;
581 goto done;
582 }
583 #ifdef VND_COMPRESSION
584 /* handle a compressed read */
585 if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
586 off_t bn;
587
588 /* Convert to a byte offset within the file. */
589 bn = obp->b_rawblkno *
590 vnd->sc_dkdev.dk_label->d_secsize;
591
592 compstrategy(obp, bn);
593 goto done;
594 }
595 #endif /* VND_COMPRESSION */
596
597 /*
598 * Allocate a header for this transfer and link it to the
599 * buffer
600 */
601 s = splbio();
602 vnx = VND_GETXFER(vnd);
603 splx(s);
604 vnx->vx_vnd = vnd;
605
606 s = splbio();
607 while (vnd->sc_active >= vnd->sc_maxactive) {
608 tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
609 }
610 vnd->sc_active++;
611 splx(s);
612
613 /* Instrumentation. */
614 disk_busy(&vnd->sc_dkdev);
615
616 bp = &vnx->vx_buf;
617 BUF_INIT(bp);
618 bp->b_flags = (obp->b_flags & B_READ) | B_CALL;
619 bp->b_iodone = vndiodone;
620 bp->b_private = obp;
621 bp->b_vp = vnd->sc_vp;
622 bp->b_data = obp->b_data;
623 bp->b_bcount = obp->b_bcount;
624 BIO_COPYPRIO(bp, obp);
625
626 /* Handle the request using the appropriate operations. */
627 if (usestrategy)
628 handle_with_strategy(vnd, obp, bp);
629 else
630 handle_with_rdwr(vnd, obp, bp);
631
632 s = splbio();
633 continue;
634
635 done:
636 biodone(obp);
637 s = splbio();
638 }
639
640 vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
641 wakeup(&vnd->sc_kthread);
642 splx(s);
643 kthread_exit(0);
644 }
645
646 /*
647 * Checks if the given vnode supports the requested operation.
648 * The operation is specified the offset returned by VOFFSET.
649 *
650 * XXX The test below used to determine this is quite fragile
651 * because it relies on the file system to use genfs to specify
652 * unimplemented operations. There might be another way to do
653 * it more cleanly.
654 */
655 static boolean_t
656 vnode_has_op(const struct vnode *vp, int opoffset)
657 {
658 int (*defaultp)(void *);
659 int (*opp)(void *);
660
661 defaultp = vp->v_op[VOFFSET(vop_default)];
662 opp = vp->v_op[opoffset];
663
664 return opp != defaultp && opp != genfs_eopnotsupp &&
665 opp != genfs_badop && opp != genfs_nullop;
666 }
667
668 /*
669 * Handes the read/write request given in 'bp' using the vnode's VOP_READ
670 * and VOP_WRITE operations.
671 *
672 * 'obp' is a pointer to the original request fed to the vnd device.
673 */
674 static void
675 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
676 {
677 boolean_t doread;
678 off_t offset;
679 size_t resid;
680 struct vnode *vp;
681
682 doread = bp->b_flags & B_READ;
683 offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
684 vp = vnd->sc_vp;
685
686 #if defined(DEBUG)
687 if (vnddebug & VDB_IO)
688 printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
689 ", secsize %d, offset %" PRIu64
690 ", bcount %d, resid %d\n",
691 vp, doread ? "read" : "write", obp->b_rawblkno,
692 vnd->sc_dkdev.dk_label->d_secsize, offset,
693 bp->b_bcount, bp->b_resid);
694 #endif
695
696 /* Issue the read or write operation. */
697 bp->b_error =
698 vn_rdwr(doread ? UIO_READ : UIO_WRITE,
699 vp, bp->b_data, bp->b_bcount, offset,
700 UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
701 bp->b_resid = resid;
702 if (bp->b_error != 0)
703 bp->b_flags |= B_ERROR;
704 else
705 KASSERT(!(bp->b_flags & B_ERROR));
706
707 /* We need to increase the number of outputs on the vnode if
708 * there was any write to it. */
709 if (!doread)
710 V_INCR_NUMOUTPUT(vp);
711
712 biodone(bp);
713 }
714
715 /*
716 * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
717 * and VOP_STRATEGY operations.
718 *
719 * 'obp' is a pointer to the original request fed to the vnd device.
720 */
721 static void
722 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
723 struct buf *bp)
724 {
725 int bsize, error, flags, skipped;
726 size_t resid, sz;
727 off_t bn, offset;
728 struct mount *mp;
729
730 flags = obp->b_flags;
731
732 mp = NULL;
733 if (!(flags & B_READ)) {
734 int s;
735
736 s = splbio();
737 V_INCR_NUMOUTPUT(bp->b_vp);
738 splx(s);
739
740 vn_start_write(vnd->sc_vp, &mp, V_WAIT);
741 KASSERT(mp != NULL);
742 }
743
744 /* convert to a byte offset within the file. */
745 bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
746
747 bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
748 skipped = 0;
749
750 /*
751 * Break the request into bsize pieces and feed them
752 * sequentially using VOP_BMAP/VOP_STRATEGY.
753 * We do it this way to keep from flooding NFS servers if we
754 * are connected to an NFS file. This places the burden on
755 * the client rather than the server.
756 */
757 error = 0;
758 bp->b_resid = bp->b_bcount;
759 for (offset = 0, resid = bp->b_resid; resid;
760 resid -= sz, offset += sz) {
761 struct buf *nbp;
762 struct vnode *vp;
763 daddr_t nbn;
764 int off, nra;
765
766 nra = 0;
767 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
768 error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
769 VOP_UNLOCK(vnd->sc_vp, 0);
770
771 if (error == 0 && (long)nbn == -1)
772 error = EIO;
773
774 /*
775 * If there was an error or a hole in the file...punt.
776 * Note that we may have to wait for any operations
777 * that we have already fired off before releasing
778 * the buffer.
779 *
780 * XXX we could deal with holes here but it would be
781 * a hassle (in the write case).
782 */
783 if (error) {
784 skipped += resid;
785 break;
786 }
787
788 #ifdef DEBUG
789 if (!dovndcluster)
790 nra = 0;
791 #endif
792
793 off = bn % bsize;
794 sz = MIN(((off_t)1 + nra) * bsize - off, resid);
795 #ifdef DEBUG
796 if (vnddebug & VDB_IO)
797 printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
798 " sz 0x%zx\n",
799 vnd->sc_vp, vp, (long long)bn, nbn, sz);
800 #endif
801
802 nbp = getiobuf();
803 nestiobuf_setup(bp, nbp, offset, sz);
804 nbp->b_blkno = nbn + btodb(off);
805
806 #if 0 /* XXX #ifdef DEBUG */
807 if (vnddebug & VDB_IO)
808 printf("vndstart(%ld): bp %p vp %p blkno "
809 "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
810 (long) (vnd-vnd_softc), &nbp->vb_buf,
811 nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
812 nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
813 nbp->vb_buf.b_bcount);
814 #endif
815 VOP_STRATEGY(vp, nbp);
816 bn += sz;
817 }
818 nestiobuf_done(bp, skipped, error);
819
820 if (!(flags & B_READ)) {
821 KASSERT(mp != NULL);
822 vn_finished_write(mp, 0);
823 }
824 }
825
826 static void
827 vndiodone(struct buf *bp)
828 {
829 struct vndxfer *vnx = VND_BUFTOXFER(bp);
830 struct vnd_softc *vnd = vnx->vx_vnd;
831 struct buf *obp = bp->b_private;
832
833 KASSERT(&vnx->vx_buf == bp);
834 KASSERT(vnd->sc_active > 0);
835 #ifdef DEBUG
836 if (vnddebug & VDB_IO) {
837 printf("vndiodone1: bp %p iodone: error %d\n",
838 bp, (bp->b_flags & B_ERROR) != 0 ? bp->b_error : 0);
839 }
840 #endif
841 disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
842 (bp->b_flags & B_READ));
843 vnd->sc_active--;
844 if (vnd->sc_active == 0) {
845 wakeup(&vnd->sc_tab);
846 }
847 obp->b_flags |= bp->b_flags & B_ERROR;
848 obp->b_error = bp->b_error;
849 obp->b_resid = bp->b_resid;
850 VND_PUTXFER(vnd, vnx);
851 biodone(obp);
852 }
853
854 /* ARGSUSED */
855 static int
856 vndread(dev_t dev, struct uio *uio, int flags)
857 {
858 int unit = vndunit(dev);
859 struct vnd_softc *sc;
860
861 #ifdef DEBUG
862 if (vnddebug & VDB_FOLLOW)
863 printf("vndread(0x%x, %p)\n", dev, uio);
864 #endif
865
866 sc = device_lookup(&vnd_cd, unit);
867 if (sc == NULL)
868 return ENXIO;
869
870 if ((sc->sc_flags & VNF_INITED) == 0)
871 return (ENXIO);
872
873 return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
874 }
875
876 /* ARGSUSED */
877 static int
878 vndwrite(dev_t dev, struct uio *uio, int flags)
879 {
880 int unit = vndunit(dev);
881 struct vnd_softc *sc;
882
883 #ifdef DEBUG
884 if (vnddebug & VDB_FOLLOW)
885 printf("vndwrite(0x%x, %p)\n", dev, uio);
886 #endif
887
888 sc = device_lookup(&vnd_cd, unit);
889 if (sc == NULL)
890 return ENXIO;
891
892 if ((sc->sc_flags & VNF_INITED) == 0)
893 return (ENXIO);
894
895 return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
896 }
897
898 static int
899 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
900 {
901 struct vnd_softc *vnd;
902
903 if (*un == -1)
904 *un = unit;
905 if (*un < 0)
906 return EINVAL;
907
908 vnd = device_lookup(&vnd_cd, *un);
909 if (vnd == NULL)
910 return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
911
912 if ((vnd->sc_flags & VNF_INITED) == 0)
913 return -1;
914
915 return VOP_GETATTR(vnd->sc_vp, va, l->l_cred, l);
916 }
917
918 /* ARGSUSED */
919 static int
920 vndioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
921 {
922 int unit = vndunit(dev);
923 struct vnd_softc *vnd;
924 struct vnd_ioctl *vio;
925 struct vattr vattr;
926 struct nameidata nd;
927 int error, part, pmask;
928 size_t geomsize;
929 int fflags;
930 #ifdef __HAVE_OLD_DISKLABEL
931 struct disklabel newlabel;
932 #endif
933
934 #ifdef DEBUG
935 if (vnddebug & VDB_FOLLOW)
936 printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
937 dev, cmd, data, flag, l->l_proc, unit);
938 #endif
939 vnd = device_lookup(&vnd_cd, unit);
940 if (vnd == NULL &&
941 #ifdef COMPAT_30
942 cmd != VNDIOOCGET &&
943 #endif
944 cmd != VNDIOCGET)
945 return ENXIO;
946 vio = (struct vnd_ioctl *)data;
947
948 /* Must be open for writes for these commands... */
949 switch (cmd) {
950 case VNDIOCSET:
951 case VNDIOCCLR:
952 case DIOCSDINFO:
953 case DIOCWDINFO:
954 #ifdef __HAVE_OLD_DISKLABEL
955 case ODIOCSDINFO:
956 case ODIOCWDINFO:
957 #endif
958 case DIOCKLABEL:
959 case DIOCWLABEL:
960 if ((flag & FWRITE) == 0)
961 return (EBADF);
962 }
963
964 /* Must be initialized for these... */
965 switch (cmd) {
966 case VNDIOCCLR:
967 case DIOCGDINFO:
968 case DIOCSDINFO:
969 case DIOCWDINFO:
970 case DIOCGPART:
971 case DIOCKLABEL:
972 case DIOCWLABEL:
973 case DIOCGDEFLABEL:
974 #ifdef __HAVE_OLD_DISKLABEL
975 case ODIOCGDINFO:
976 case ODIOCSDINFO:
977 case ODIOCWDINFO:
978 case ODIOCGDEFLABEL:
979 #endif
980 if ((vnd->sc_flags & VNF_INITED) == 0)
981 return (ENXIO);
982 }
983
984 switch (cmd) {
985 case VNDIOCSET:
986 if (vnd->sc_flags & VNF_INITED)
987 return (EBUSY);
988
989 if ((error = vndlock(vnd)) != 0)
990 return (error);
991
992 fflags = FREAD;
993 if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
994 fflags |= FWRITE;
995 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, l);
996 if ((error = vn_open(&nd, fflags, 0)) != 0)
997 goto unlock_and_exit;
998 KASSERT(l);
999 error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred, l);
1000 if (!error && nd.ni_vp->v_type != VREG)
1001 error = EOPNOTSUPP;
1002 if (error) {
1003 VOP_UNLOCK(nd.ni_vp, 0);
1004 goto close_and_exit;
1005 }
1006
1007 /* If using a compressed file, initialize its info */
1008 /* (or abort with an error if kernel has no compression) */
1009 if (vio->vnd_flags & VNF_COMP) {
1010 #ifdef VND_COMPRESSION
1011 struct vnd_comp_header *ch;
1012 int i;
1013 u_int32_t comp_size;
1014 u_int32_t comp_maxsize;
1015
1016 /* allocate space for compresed file header */
1017 ch = malloc(sizeof(struct vnd_comp_header),
1018 M_TEMP, M_WAITOK);
1019
1020 /* read compressed file header */
1021 error = vn_rdwr(UIO_READ, nd.ni_vp, (caddr_t)ch,
1022 sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1023 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1024 if(error) {
1025 free(ch, M_TEMP);
1026 VOP_UNLOCK(nd.ni_vp, 0);
1027 goto close_and_exit;
1028 }
1029
1030 /* save some header info */
1031 vnd->sc_comp_blksz = ntohl(ch->block_size);
1032 /* note last offset is the file byte size */
1033 vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1034 free(ch, M_TEMP);
1035 if (vnd->sc_comp_blksz == 0 ||
1036 vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1037 VOP_UNLOCK(nd.ni_vp, 0);
1038 error = EINVAL;
1039 goto close_and_exit;
1040 }
1041 if(sizeof(struct vnd_comp_header) +
1042 sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1043 vattr.va_size) {
1044 VOP_UNLOCK(nd.ni_vp, 0);
1045 error = EINVAL;
1046 goto close_and_exit;
1047 }
1048
1049 /* set decompressed file size */
1050 vattr.va_size =
1051 ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1052 (u_quad_t)vnd->sc_comp_blksz;
1053
1054 /* allocate space for all the compressed offsets */
1055 vnd->sc_comp_offsets =
1056 malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1057 M_DEVBUF, M_WAITOK);
1058
1059 /* read in the offsets */
1060 error = vn_rdwr(UIO_READ, nd.ni_vp,
1061 (caddr_t)vnd->sc_comp_offsets,
1062 sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1063 sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1064 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1065 if(error) {
1066 VOP_UNLOCK(nd.ni_vp, 0);
1067 goto close_and_exit;
1068 }
1069 /*
1070 * find largest block size (used for allocation limit).
1071 * Also convert offset to native byte order.
1072 */
1073 comp_maxsize = 0;
1074 for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1075 vnd->sc_comp_offsets[i] =
1076 be64toh(vnd->sc_comp_offsets[i]);
1077 comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1078 - vnd->sc_comp_offsets[i];
1079 if (comp_size > comp_maxsize)
1080 comp_maxsize = comp_size;
1081 }
1082 vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1083 be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1084
1085 /* create compressed data buffer */
1086 vnd->sc_comp_buff = malloc(comp_maxsize,
1087 M_DEVBUF, M_WAITOK);
1088
1089 /* create decompressed buffer */
1090 vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1091 M_DEVBUF, M_WAITOK);
1092 vnd->sc_comp_buffblk = -1;
1093
1094 /* Initialize decompress stream */
1095 bzero(&vnd->sc_comp_stream, sizeof(z_stream));
1096 vnd->sc_comp_stream.zalloc = vnd_alloc;
1097 vnd->sc_comp_stream.zfree = vnd_free;
1098 error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1099 if(error) {
1100 if(vnd->sc_comp_stream.msg)
1101 printf("vnd%d: compressed file, %s\n",
1102 unit, vnd->sc_comp_stream.msg);
1103 VOP_UNLOCK(nd.ni_vp, 0);
1104 error = EINVAL;
1105 goto close_and_exit;
1106 }
1107
1108 vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1109 #else /* !VND_COMPRESSION */
1110 VOP_UNLOCK(nd.ni_vp, 0);
1111 error = EOPNOTSUPP;
1112 goto close_and_exit;
1113 #endif /* VND_COMPRESSION */
1114 }
1115
1116 VOP_UNLOCK(nd.ni_vp, 0);
1117 vnd->sc_vp = nd.ni_vp;
1118 vnd->sc_size = btodb(vattr.va_size); /* note truncation */
1119
1120 /*
1121 * Use pseudo-geometry specified. If none was provided,
1122 * use "standard" Adaptec fictitious geometry.
1123 */
1124 if (vio->vnd_flags & VNDIOF_HASGEOM) {
1125
1126 memcpy(&vnd->sc_geom, &vio->vnd_geom,
1127 sizeof(vio->vnd_geom));
1128
1129 /*
1130 * Sanity-check the sector size.
1131 * XXX Don't allow secsize < DEV_BSIZE. Should
1132 * XXX we?
1133 */
1134 if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1135 (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1136 vnd->sc_geom.vng_ncylinders == 0 ||
1137 (vnd->sc_geom.vng_ntracks *
1138 vnd->sc_geom.vng_nsectors) == 0) {
1139 error = EINVAL;
1140 goto close_and_exit;
1141 }
1142
1143 /*
1144 * Compute the size (in DEV_BSIZE blocks) specified
1145 * by the geometry.
1146 */
1147 geomsize = (vnd->sc_geom.vng_nsectors *
1148 vnd->sc_geom.vng_ntracks *
1149 vnd->sc_geom.vng_ncylinders) *
1150 (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1151
1152 /*
1153 * Sanity-check the size against the specified
1154 * geometry.
1155 */
1156 if (vnd->sc_size < geomsize) {
1157 error = EINVAL;
1158 goto close_and_exit;
1159 }
1160 } else if (vnd->sc_size >= (32 * 64)) {
1161 /*
1162 * Size must be at least 2048 DEV_BSIZE blocks
1163 * (1M) in order to use this geometry.
1164 */
1165 vnd->sc_geom.vng_secsize = DEV_BSIZE;
1166 vnd->sc_geom.vng_nsectors = 32;
1167 vnd->sc_geom.vng_ntracks = 64;
1168 vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1169 } else {
1170 vnd->sc_geom.vng_secsize = DEV_BSIZE;
1171 vnd->sc_geom.vng_nsectors = 1;
1172 vnd->sc_geom.vng_ntracks = 1;
1173 vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1174 }
1175
1176 vnd_set_properties(vnd);
1177
1178 if (vio->vnd_flags & VNDIOF_READONLY) {
1179 vnd->sc_flags |= VNF_READONLY;
1180 }
1181
1182 if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1183 goto close_and_exit;
1184
1185 vndthrottle(vnd, vnd->sc_vp);
1186 vio->vnd_size = dbtob(vnd->sc_size);
1187 vnd->sc_flags |= VNF_INITED;
1188
1189 /* create the kernel thread, wait for it to be up */
1190 error = kthread_create1(vndthread, vnd, &vnd->sc_kthread,
1191 vnd->sc_dev.dv_xname);
1192 if (error)
1193 goto close_and_exit;
1194 while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1195 tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1196 }
1197 #ifdef DEBUG
1198 if (vnddebug & VDB_INIT)
1199 printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1200 vnd->sc_vp, (unsigned long) vnd->sc_size,
1201 vnd->sc_geom.vng_secsize,
1202 vnd->sc_geom.vng_nsectors,
1203 vnd->sc_geom.vng_ntracks,
1204 vnd->sc_geom.vng_ncylinders);
1205 #endif
1206
1207 /* Attach the disk. */
1208 vnd->sc_dkdev.dk_name = vnd->sc_dev.dv_xname;
1209 pseudo_disk_attach(&vnd->sc_dkdev);
1210
1211 /* Initialize the xfer and buffer pools. */
1212 pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1213 0, 0, "vndxpl", NULL);
1214
1215 /* Try and read the disklabel. */
1216 vndgetdisklabel(dev, vnd);
1217
1218 vndunlock(vnd);
1219
1220 break;
1221
1222 close_and_exit:
1223 (void) vn_close(nd.ni_vp, fflags, l->l_cred, l);
1224 unlock_and_exit:
1225 #ifdef VND_COMPRESSION
1226 /* free any allocated memory (for compressed file) */
1227 if(vnd->sc_comp_offsets) {
1228 free(vnd->sc_comp_offsets, M_DEVBUF);
1229 vnd->sc_comp_offsets = NULL;
1230 }
1231 if(vnd->sc_comp_buff) {
1232 free(vnd->sc_comp_buff, M_DEVBUF);
1233 vnd->sc_comp_buff = NULL;
1234 }
1235 if(vnd->sc_comp_decombuf) {
1236 free(vnd->sc_comp_decombuf, M_DEVBUF);
1237 vnd->sc_comp_decombuf = NULL;
1238 }
1239 #endif /* VND_COMPRESSION */
1240 vndunlock(vnd);
1241 return (error);
1242
1243 case VNDIOCCLR:
1244 if ((error = vndlock(vnd)) != 0)
1245 return (error);
1246
1247 /*
1248 * Don't unconfigure if any other partitions are open
1249 * or if both the character and block flavors of this
1250 * partition are open.
1251 */
1252 part = DISKPART(dev);
1253 pmask = (1 << part);
1254 if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1255 ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1256 (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1257 !(vio->vnd_flags & VNDIOF_FORCE)) {
1258 vndunlock(vnd);
1259 return (EBUSY);
1260 }
1261
1262 /*
1263 * XXX vndclear() might call vndclose() implicitely;
1264 * release lock to avoid recursion
1265 */
1266 vndunlock(vnd);
1267 vndclear(vnd, minor(dev));
1268 #ifdef DEBUG
1269 if (vnddebug & VDB_INIT)
1270 printf("vndioctl: CLRed\n");
1271 #endif
1272
1273 /* Destroy the xfer and buffer pools. */
1274 pool_destroy(&vnd->sc_vxpool);
1275
1276 /* Detatch the disk. */
1277 pseudo_disk_detach(&vnd->sc_dkdev);
1278 break;
1279
1280 #ifdef COMPAT_30
1281 case VNDIOOCGET: {
1282 struct vnd_ouser *vnu;
1283 struct vattr va;
1284 vnu = (struct vnd_ouser *)data;
1285 KASSERT(l);
1286 switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1287 case 0:
1288 vnu->vnu_dev = va.va_fsid;
1289 vnu->vnu_ino = va.va_fileid;
1290 break;
1291 case -1:
1292 /* unused is not an error */
1293 vnu->vnu_dev = 0;
1294 vnu->vnu_ino = 0;
1295 break;
1296 default:
1297 return error;
1298 }
1299 break;
1300 }
1301 #endif
1302 case VNDIOCGET: {
1303 struct vnd_user *vnu;
1304 struct vattr va;
1305 vnu = (struct vnd_user *)data;
1306 KASSERT(l);
1307 switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1308 case 0:
1309 vnu->vnu_dev = va.va_fsid;
1310 vnu->vnu_ino = va.va_fileid;
1311 break;
1312 case -1:
1313 /* unused is not an error */
1314 vnu->vnu_dev = 0;
1315 vnu->vnu_ino = 0;
1316 break;
1317 default:
1318 return error;
1319 }
1320 break;
1321 }
1322
1323 case DIOCGDINFO:
1324 *(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1325 break;
1326
1327 #ifdef __HAVE_OLD_DISKLABEL
1328 case ODIOCGDINFO:
1329 newlabel = *(vnd->sc_dkdev.dk_label);
1330 if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1331 return ENOTTY;
1332 memcpy(data, &newlabel, sizeof (struct olddisklabel));
1333 break;
1334 #endif
1335
1336 case DIOCGPART:
1337 ((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1338 ((struct partinfo *)data)->part =
1339 &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1340 break;
1341
1342 case DIOCWDINFO:
1343 case DIOCSDINFO:
1344 #ifdef __HAVE_OLD_DISKLABEL
1345 case ODIOCWDINFO:
1346 case ODIOCSDINFO:
1347 #endif
1348 {
1349 struct disklabel *lp;
1350
1351 if ((error = vndlock(vnd)) != 0)
1352 return (error);
1353
1354 vnd->sc_flags |= VNF_LABELLING;
1355
1356 #ifdef __HAVE_OLD_DISKLABEL
1357 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1358 memset(&newlabel, 0, sizeof newlabel);
1359 memcpy(&newlabel, data, sizeof (struct olddisklabel));
1360 lp = &newlabel;
1361 } else
1362 #endif
1363 lp = (struct disklabel *)data;
1364
1365 error = setdisklabel(vnd->sc_dkdev.dk_label,
1366 lp, 0, vnd->sc_dkdev.dk_cpulabel);
1367 if (error == 0) {
1368 if (cmd == DIOCWDINFO
1369 #ifdef __HAVE_OLD_DISKLABEL
1370 || cmd == ODIOCWDINFO
1371 #endif
1372 )
1373 error = writedisklabel(VNDLABELDEV(dev),
1374 vndstrategy, vnd->sc_dkdev.dk_label,
1375 vnd->sc_dkdev.dk_cpulabel);
1376 }
1377
1378 vnd->sc_flags &= ~VNF_LABELLING;
1379
1380 vndunlock(vnd);
1381
1382 if (error)
1383 return (error);
1384 break;
1385 }
1386
1387 case DIOCKLABEL:
1388 if (*(int *)data != 0)
1389 vnd->sc_flags |= VNF_KLABEL;
1390 else
1391 vnd->sc_flags &= ~VNF_KLABEL;
1392 break;
1393
1394 case DIOCWLABEL:
1395 if (*(int *)data != 0)
1396 vnd->sc_flags |= VNF_WLABEL;
1397 else
1398 vnd->sc_flags &= ~VNF_WLABEL;
1399 break;
1400
1401 case DIOCGDEFLABEL:
1402 vndgetdefaultlabel(vnd, (struct disklabel *)data);
1403 break;
1404
1405 #ifdef __HAVE_OLD_DISKLABEL
1406 case ODIOCGDEFLABEL:
1407 vndgetdefaultlabel(vnd, &newlabel);
1408 if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1409 return ENOTTY;
1410 memcpy(data, &newlabel, sizeof (struct olddisklabel));
1411 break;
1412 #endif
1413
1414 default:
1415 return (ENOTTY);
1416 }
1417
1418 return (0);
1419 }
1420
1421 /*
1422 * Duplicate the current processes' credentials. Since we are called only
1423 * as the result of a SET ioctl and only root can do that, any future access
1424 * to this "disk" is essentially as root. Note that credentials may change
1425 * if some other uid can write directly to the mapped file (NFS).
1426 */
1427 static int
1428 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1429 {
1430 struct uio auio;
1431 struct iovec aiov;
1432 char *tmpbuf;
1433 int error;
1434
1435 vnd->sc_cred = kauth_cred_dup(cred);
1436 tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1437
1438 /* XXX: Horrible kludge to establish credentials for NFS */
1439 aiov.iov_base = tmpbuf;
1440 aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1441 auio.uio_iov = &aiov;
1442 auio.uio_iovcnt = 1;
1443 auio.uio_offset = 0;
1444 auio.uio_rw = UIO_READ;
1445 auio.uio_resid = aiov.iov_len;
1446 UIO_SETUP_SYSSPACE(&auio);
1447 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1448 error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1449 if (error == 0) {
1450 /*
1451 * Because vnd does all IO directly through the vnode
1452 * we need to flush (at least) the buffer from the above
1453 * VOP_READ from the buffer cache to prevent cache
1454 * incoherencies. Also, be careful to write dirty
1455 * buffers back to stable storage.
1456 */
1457 error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1458 curlwp, 0, 0);
1459 }
1460 VOP_UNLOCK(vnd->sc_vp, 0);
1461
1462 free(tmpbuf, M_TEMP);
1463 return (error);
1464 }
1465
1466 /*
1467 * Set maxactive based on FS type
1468 */
1469 static void
1470 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1471 {
1472 #ifdef NFS
1473 extern int (**nfsv2_vnodeop_p)(void *);
1474
1475 if (vp->v_op == nfsv2_vnodeop_p)
1476 vnd->sc_maxactive = 2;
1477 else
1478 #endif
1479 vnd->sc_maxactive = 8;
1480
1481 if (vnd->sc_maxactive < 1)
1482 vnd->sc_maxactive = 1;
1483 }
1484
1485 #if 0
1486 static void
1487 vndshutdown(void)
1488 {
1489 struct vnd_softc *vnd;
1490
1491 for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1492 if (vnd->sc_flags & VNF_INITED)
1493 vndclear(vnd);
1494 }
1495 #endif
1496
1497 static void
1498 vndclear(struct vnd_softc *vnd, int myminor)
1499 {
1500 struct vnode *vp = vnd->sc_vp;
1501 struct lwp *l = curlwp;
1502 int fflags = FREAD;
1503 int bmaj, cmaj, i, mn;
1504 int s;
1505
1506 #ifdef DEBUG
1507 if (vnddebug & VDB_FOLLOW)
1508 printf("vndclear(%p): vp %p\n", vnd, vp);
1509 #endif
1510 /* locate the major number */
1511 bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1512 cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1513
1514 /* Nuke the vnodes for any open instances */
1515 for (i = 0; i < MAXPARTITIONS; i++) {
1516 mn = DISKMINOR(device_unit(&vnd->sc_dev), i);
1517 vdevgone(bmaj, mn, mn, VBLK);
1518 if (mn != myminor) /* XXX avoid to kill own vnode */
1519 vdevgone(cmaj, mn, mn, VCHR);
1520 }
1521
1522 if ((vnd->sc_flags & VNF_READONLY) == 0)
1523 fflags |= FWRITE;
1524
1525 s = splbio();
1526 bufq_drain(vnd->sc_tab);
1527 splx(s);
1528
1529 vnd->sc_flags |= VNF_VUNCONF;
1530 wakeup(&vnd->sc_tab);
1531 while (vnd->sc_flags & VNF_KTHREAD)
1532 tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1533
1534 #ifdef VND_COMPRESSION
1535 /* free the compressed file buffers */
1536 if(vnd->sc_flags & VNF_COMP) {
1537 if(vnd->sc_comp_offsets) {
1538 free(vnd->sc_comp_offsets, M_DEVBUF);
1539 vnd->sc_comp_offsets = NULL;
1540 }
1541 if(vnd->sc_comp_buff) {
1542 free(vnd->sc_comp_buff, M_DEVBUF);
1543 vnd->sc_comp_buff = NULL;
1544 }
1545 if(vnd->sc_comp_decombuf) {
1546 free(vnd->sc_comp_decombuf, M_DEVBUF);
1547 vnd->sc_comp_decombuf = NULL;
1548 }
1549 }
1550 #endif /* VND_COMPRESSION */
1551 vnd->sc_flags &=
1552 ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1553 | VNF_VUNCONF | VNF_COMP);
1554 if (vp == (struct vnode *)0)
1555 panic("vndclear: null vp");
1556 (void) vn_close(vp, fflags, vnd->sc_cred, l);
1557 kauth_cred_free(vnd->sc_cred);
1558 vnd->sc_vp = (struct vnode *)0;
1559 vnd->sc_cred = (kauth_cred_t)0;
1560 vnd->sc_size = 0;
1561 }
1562
1563 static int
1564 vndsize(dev_t dev)
1565 {
1566 struct vnd_softc *sc;
1567 struct disklabel *lp;
1568 int part, unit, omask;
1569 int size;
1570
1571 unit = vndunit(dev);
1572 sc = (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1573 if (sc == NULL)
1574 return -1;
1575
1576 if ((sc->sc_flags & VNF_INITED) == 0)
1577 return (-1);
1578
1579 part = DISKPART(dev);
1580 omask = sc->sc_dkdev.dk_openmask & (1 << part);
1581 lp = sc->sc_dkdev.dk_label;
1582
1583 if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp)) /* XXX */
1584 return (-1);
1585
1586 if (lp->d_partitions[part].p_fstype != FS_SWAP)
1587 size = -1;
1588 else
1589 size = lp->d_partitions[part].p_size *
1590 (lp->d_secsize / DEV_BSIZE);
1591
1592 if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp)) /* XXX */
1593 return (-1);
1594
1595 return (size);
1596 }
1597
1598 static int
1599 vnddump(dev_t dev, daddr_t blkno, caddr_t va,
1600 size_t size)
1601 {
1602
1603 /* Not implemented. */
1604 return ENXIO;
1605 }
1606
1607 static void
1608 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1609 {
1610 struct vndgeom *vng = &sc->sc_geom;
1611 struct partition *pp;
1612
1613 memset(lp, 0, sizeof(*lp));
1614
1615 lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1616 lp->d_secsize = vng->vng_secsize;
1617 lp->d_nsectors = vng->vng_nsectors;
1618 lp->d_ntracks = vng->vng_ntracks;
1619 lp->d_ncylinders = vng->vng_ncylinders;
1620 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1621
1622 strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1623 lp->d_type = DTYPE_VND;
1624 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1625 lp->d_rpm = 3600;
1626 lp->d_interleave = 1;
1627 lp->d_flags = 0;
1628
1629 pp = &lp->d_partitions[RAW_PART];
1630 pp->p_offset = 0;
1631 pp->p_size = lp->d_secperunit;
1632 pp->p_fstype = FS_UNUSED;
1633 lp->d_npartitions = RAW_PART + 1;
1634
1635 lp->d_magic = DISKMAGIC;
1636 lp->d_magic2 = DISKMAGIC;
1637 lp->d_checksum = dkcksum(lp);
1638 }
1639
1640 /*
1641 * Read the disklabel from a vnd. If one is not present, create a fake one.
1642 */
1643 static void
1644 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1645 {
1646 const char *errstring;
1647 struct disklabel *lp = sc->sc_dkdev.dk_label;
1648 struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1649 int i;
1650
1651 memset(clp, 0, sizeof(*clp));
1652
1653 vndgetdefaultlabel(sc, lp);
1654
1655 /*
1656 * Call the generic disklabel extraction routine.
1657 */
1658 errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1659 if (errstring) {
1660 /*
1661 * Lack of disklabel is common, but we print the warning
1662 * anyway, since it might contain other useful information.
1663 */
1664 printf("%s: %s\n", sc->sc_dev.dv_xname, errstring);
1665
1666 /*
1667 * For historical reasons, if there's no disklabel
1668 * present, all partitions must be FS_BSDFFS and
1669 * occupy the entire disk.
1670 */
1671 for (i = 0; i < MAXPARTITIONS; i++) {
1672 /*
1673 * Don't wipe out port specific hack (such as
1674 * dos partition hack of i386 port).
1675 */
1676 if (lp->d_partitions[i].p_size != 0)
1677 continue;
1678
1679 lp->d_partitions[i].p_size = lp->d_secperunit;
1680 lp->d_partitions[i].p_offset = 0;
1681 lp->d_partitions[i].p_fstype = FS_BSDFFS;
1682 }
1683
1684 strncpy(lp->d_packname, "default label",
1685 sizeof(lp->d_packname));
1686
1687 lp->d_npartitions = MAXPARTITIONS;
1688 lp->d_checksum = dkcksum(lp);
1689 }
1690
1691 /* In-core label now valid. */
1692 sc->sc_flags |= VNF_VLABEL;
1693 }
1694
1695 /*
1696 * Wait interruptibly for an exclusive lock.
1697 *
1698 * XXX
1699 * Several drivers do this; it should be abstracted and made MP-safe.
1700 */
1701 static int
1702 vndlock(struct vnd_softc *sc)
1703 {
1704 int error;
1705
1706 while ((sc->sc_flags & VNF_LOCKED) != 0) {
1707 sc->sc_flags |= VNF_WANTED;
1708 if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1709 return (error);
1710 }
1711 sc->sc_flags |= VNF_LOCKED;
1712 return (0);
1713 }
1714
1715 /*
1716 * Unlock and wake up any waiters.
1717 */
1718 static void
1719 vndunlock(struct vnd_softc *sc)
1720 {
1721
1722 sc->sc_flags &= ~VNF_LOCKED;
1723 if ((sc->sc_flags & VNF_WANTED) != 0) {
1724 sc->sc_flags &= ~VNF_WANTED;
1725 wakeup(sc);
1726 }
1727 }
1728
1729 #ifdef VND_COMPRESSION
1730 /* compressed file read */
1731 static void
1732 compstrategy(struct buf *bp, off_t bn)
1733 {
1734 int error;
1735 int unit = vndunit(bp->b_dev);
1736 struct vnd_softc *vnd =
1737 (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1738 u_int32_t comp_block;
1739 struct uio auio;
1740 caddr_t addr;
1741 int s;
1742
1743 /* set up constants for data move */
1744 auio.uio_rw = UIO_READ;
1745 UIO_SETUP_SYSSPACE(&auio);
1746
1747 /* read, and transfer the data */
1748 addr = bp->b_data;
1749 bp->b_resid = bp->b_bcount;
1750 s = splbio();
1751 while (bp->b_resid > 0) {
1752 unsigned length;
1753 size_t length_in_buffer;
1754 u_int32_t offset_in_buffer;
1755 struct iovec aiov;
1756
1757 /* calculate the compressed block number */
1758 comp_block = bn / (off_t)vnd->sc_comp_blksz;
1759
1760 /* check for good block number */
1761 if (comp_block >= vnd->sc_comp_numoffs) {
1762 bp->b_error = EINVAL;
1763 bp->b_flags |= B_ERROR;
1764 splx(s);
1765 return;
1766 }
1767
1768 /* read in the compressed block, if not in buffer */
1769 if (comp_block != vnd->sc_comp_buffblk) {
1770 length = vnd->sc_comp_offsets[comp_block + 1] -
1771 vnd->sc_comp_offsets[comp_block];
1772 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1773 error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1774 length, vnd->sc_comp_offsets[comp_block],
1775 UIO_SYSSPACE, IO_UNIT, vnd->sc_cred, NULL, NULL);
1776 if (error) {
1777 bp->b_error = error;
1778 bp->b_flags |= B_ERROR;
1779 VOP_UNLOCK(vnd->sc_vp, 0);
1780 splx(s);
1781 return;
1782 }
1783 /* uncompress the buffer */
1784 vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1785 vnd->sc_comp_stream.avail_in = length;
1786 vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1787 vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1788 inflateReset(&vnd->sc_comp_stream);
1789 error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1790 if (error != Z_STREAM_END) {
1791 if (vnd->sc_comp_stream.msg)
1792 printf("%s: compressed file, %s\n",
1793 vnd->sc_dev.dv_xname,
1794 vnd->sc_comp_stream.msg);
1795 bp->b_error = EBADMSG;
1796 bp->b_flags |= B_ERROR;
1797 VOP_UNLOCK(vnd->sc_vp, 0);
1798 splx(s);
1799 return;
1800 }
1801 vnd->sc_comp_buffblk = comp_block;
1802 VOP_UNLOCK(vnd->sc_vp, 0);
1803 }
1804
1805 /* transfer the usable uncompressed data */
1806 offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1807 length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1808 if (length_in_buffer > bp->b_resid)
1809 length_in_buffer = bp->b_resid;
1810 auio.uio_iov = &aiov;
1811 auio.uio_iovcnt = 1;
1812 aiov.iov_base = addr;
1813 aiov.iov_len = length_in_buffer;
1814 auio.uio_resid = aiov.iov_len;
1815 auio.uio_offset = 0;
1816 error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1817 length_in_buffer, &auio);
1818 if (error) {
1819 bp->b_error = error;
1820 bp->b_flags |= B_ERROR;
1821 splx(s);
1822 return;
1823 }
1824
1825 bn += length_in_buffer;
1826 addr += length_in_buffer;
1827 bp->b_resid -= length_in_buffer;
1828 }
1829 splx(s);
1830 }
1831
1832 /* compression memory allocation routines */
1833 static void *
1834 vnd_alloc(void *aux, u_int items, u_int siz)
1835 {
1836 return malloc(items * siz, M_TEMP, M_NOWAIT);
1837 }
1838
1839 static void
1840 vnd_free(void *aux, void *ptr)
1841 {
1842 free(ptr, M_TEMP);
1843 }
1844 #endif /* VND_COMPRESSION */
1845
1846 static void
1847 vnd_set_properties(struct vnd_softc *vnd)
1848 {
1849 prop_dictionary_t disk_info, odisk_info, geom;
1850
1851 disk_info = prop_dictionary_create();
1852
1853 geom = prop_dictionary_create();
1854
1855 prop_dictionary_set_uint64(geom, "sectors-per-unit",
1856 vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1857 vnd->sc_geom.vng_ncylinders);
1858
1859 prop_dictionary_set_uint32(geom, "sector-size",
1860 vnd->sc_geom.vng_secsize);
1861
1862 prop_dictionary_set_uint16(geom, "sectors-per-track",
1863 vnd->sc_geom.vng_nsectors);
1864
1865 prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1866 vnd->sc_geom.vng_ntracks);
1867
1868 prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1869 vnd->sc_geom.vng_ncylinders);
1870
1871 prop_dictionary_set(disk_info, "geometry", geom);
1872 prop_object_release(geom);
1873
1874 prop_dictionary_set(device_properties(&vnd->sc_dev),
1875 "disk-info", disk_info);
1876
1877 /*
1878 * Don't release disk_info here; we keep a reference to it.
1879 * disk_detach() will release it when we go away.
1880 */
1881
1882 odisk_info = vnd->sc_dkdev.dk_info;
1883 vnd->sc_dkdev.dk_info = disk_info;
1884 if (odisk_info)
1885 prop_object_release(odisk_info);
1886 }
Cache object: 02902177f4cd61497d500bcac115f36e
|