vnd.c revision 1.273 1 1.273 ad /* $NetBSD: vnd.c,v 1.273 2020/01/17 19:31:30 ad Exp $ */
2 1.41 thorpej
3 1.41 thorpej /*-
4 1.177 ad * Copyright (c) 1996, 1997, 1998, 2008 The NetBSD Foundation, Inc.
5 1.41 thorpej * All rights reserved.
6 1.41 thorpej *
7 1.41 thorpej * This code is derived from software contributed to The NetBSD Foundation
8 1.41 thorpej * by Jason R. Thorpe.
9 1.41 thorpej *
10 1.41 thorpej * Redistribution and use in source and binary forms, with or without
11 1.41 thorpej * modification, are permitted provided that the following conditions
12 1.41 thorpej * are met:
13 1.41 thorpej * 1. Redistributions of source code must retain the above copyright
14 1.41 thorpej * notice, this list of conditions and the following disclaimer.
15 1.41 thorpej * 2. Redistributions in binary form must reproduce the above copyright
16 1.41 thorpej * notice, this list of conditions and the following disclaimer in the
17 1.41 thorpej * documentation and/or other materials provided with the distribution.
18 1.41 thorpej *
19 1.41 thorpej * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 1.41 thorpej * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 1.41 thorpej * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 1.48 jtc * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 1.48 jtc * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 1.41 thorpej * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 1.41 thorpej * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 1.41 thorpej * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 1.41 thorpej * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 1.41 thorpej * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 1.41 thorpej * POSSIBILITY OF SUCH DAMAGE.
30 1.41 thorpej */
31 1.11 cgd
32 1.1 brezak /*
33 1.215 rmind * Copyright (c) 1988 University of Utah.
34 1.1 brezak * Copyright (c) 1990, 1993
35 1.1 brezak * The Regents of the University of California. All rights reserved.
36 1.1 brezak *
37 1.1 brezak * This code is derived from software contributed to Berkeley by
38 1.1 brezak * the Systems Programming Group of the University of Utah Computer
39 1.1 brezak * Science Department.
40 1.1 brezak *
41 1.1 brezak * Redistribution and use in source and binary forms, with or without
42 1.1 brezak * modification, are permitted provided that the following conditions
43 1.1 brezak * are met:
44 1.1 brezak * 1. Redistributions of source code must retain the above copyright
45 1.1 brezak * notice, this list of conditions and the following disclaimer.
46 1.1 brezak * 2. Redistributions in binary form must reproduce the above copyright
47 1.1 brezak * notice, this list of conditions and the following disclaimer in the
48 1.1 brezak * documentation and/or other materials provided with the distribution.
49 1.102 agc * 3. Neither the name of the University nor the names of its contributors
50 1.102 agc * may be used to endorse or promote products derived from this software
51 1.102 agc * without specific prior written permission.
52 1.102 agc *
53 1.102 agc * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54 1.102 agc * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55 1.102 agc * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56 1.102 agc * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57 1.102 agc * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58 1.102 agc * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59 1.102 agc * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60 1.102 agc * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61 1.102 agc * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62 1.102 agc * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63 1.102 agc * SUCH DAMAGE.
64 1.102 agc *
65 1.102 agc * from: Utah $Hdr: vn.c 1.13 94/04/02$
66 1.102 agc *
67 1.102 agc * @(#)vn.c 8.9 (Berkeley) 5/14/95
68 1.102 agc */
69 1.102 agc
70 1.102 agc /*
71 1.1 brezak * Vnode disk driver.
72 1.1 brezak *
73 1.1 brezak * Block/character interface to a vnode. Allows one to treat a file
74 1.1 brezak * as a disk (e.g. build a filesystem in it, mount it, etc.).
75 1.1 brezak *
76 1.157 jmmv * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
77 1.157 jmmv * this uses them to avoid distorting the local buffer cache. If those
78 1.157 jmmv * block-level operations are not available, this falls back to the regular
79 1.157 jmmv * read and write calls. Using these may distort the cache in some cases
80 1.157 jmmv * but better have the driver working than preventing it to work on file
81 1.157 jmmv * systems where the block-level operations are not implemented for
82 1.157 jmmv * whatever reason.
83 1.1 brezak *
84 1.1 brezak * NOTE 2: There is a security issue involved with this driver.
85 1.1 brezak * Once mounted all access to the contents of the "mapped" file via
86 1.1 brezak * the special file is controlled by the permissions on the special
87 1.1 brezak * file, the protection of the mapped file is ignored (effectively,
88 1.1 brezak * by using root credentials in all transactions).
89 1.1 brezak *
90 1.1 brezak * NOTE 3: Doesn't interact with leases, should it?
91 1.1 brezak */
92 1.75 lukem
93 1.75 lukem #include <sys/cdefs.h>
94 1.273 ad __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.273 2020/01/17 19:31:30 ad Exp $");
95 1.55 thorpej
96 1.74 mrg #if defined(_KERNEL_OPT)
97 1.115 hubertf #include "opt_vnd.h"
98 1.211 mrg #include "opt_compat_netbsd.h"
99 1.74 mrg #endif
100 1.1 brezak
101 1.1 brezak #include <sys/param.h>
102 1.1 brezak #include <sys/systm.h>
103 1.1 brezak #include <sys/namei.h>
104 1.1 brezak #include <sys/proc.h>
105 1.112 bouyer #include <sys/kthread.h>
106 1.1 brezak #include <sys/errno.h>
107 1.1 brezak #include <sys/buf.h>
108 1.111 yamt #include <sys/bufq.h>
109 1.1 brezak #include <sys/malloc.h>
110 1.1 brezak #include <sys/ioctl.h>
111 1.16 cgd #include <sys/disklabel.h>
112 1.22 thorpej #include <sys/device.h>
113 1.22 thorpej #include <sys/disk.h>
114 1.22 thorpej #include <sys/stat.h>
115 1.1 brezak #include <sys/mount.h>
116 1.1 brezak #include <sys/vnode.h>
117 1.266 hannken #include <sys/fstrans.h>
118 1.1 brezak #include <sys/file.h>
119 1.1 brezak #include <sys/uio.h>
120 1.26 christos #include <sys/conf.h>
121 1.147 elad #include <sys/kauth.h>
122 1.257 pgoyette #include <sys/module.h>
123 1.271 pgoyette #include <sys/compat_stub.h>
124 1.181 ad
125 1.181 ad #include <net/zlib.h>
126 1.1 brezak
127 1.157 jmmv #include <miscfs/genfs/genfs.h>
128 1.1 brezak #include <miscfs/specfs/specdev.h>
129 1.1 brezak
130 1.204 dyoung #include <dev/dkvar.h>
131 1.41 thorpej #include <dev/vndvar.h>
132 1.41 thorpej
133 1.248 christos #include "ioconf.h"
134 1.248 christos
135 1.41 thorpej #if defined(VNDDEBUG) && !defined(DEBUG)
136 1.116 christos #define DEBUG
137 1.41 thorpej #endif
138 1.1 brezak
139 1.1 brezak #ifdef DEBUG
140 1.17 cgd int dovndcluster = 1;
141 1.116 christos #define VDB_FOLLOW 0x01
142 1.116 christos #define VDB_INIT 0x02
143 1.116 christos #define VDB_IO 0x04
144 1.116 christos #define VDB_LABEL 0x08
145 1.250 christos int vnddebug = 0;
146 1.1 brezak #endif
147 1.1 brezak
148 1.116 christos #define vndunit(x) DISKUNIT(x)
149 1.18 cgd
150 1.36 pk struct vndxfer {
151 1.129 yamt struct buf vx_buf;
152 1.129 yamt struct vnd_softc *vx_vnd;
153 1.18 cgd };
154 1.129 yamt #define VND_BUFTOXFER(bp) ((struct vndxfer *)(void *)bp)
155 1.1 brezak
156 1.116 christos #define VND_GETXFER(vnd) pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
157 1.116 christos #define VND_PUTXFER(vnd, vx) pool_put(&(vnd)->sc_vxpool, (vx))
158 1.59 thorpej
159 1.116 christos #define VNDLABELDEV(dev) \
160 1.116 christos (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
161 1.41 thorpej
162 1.220 hannken #define VND_MAXPENDING(vnd) ((vnd)->sc_maxactive * 4)
163 1.269 mlelstv #define VND_MAXPAGES(vnd) (1024 * 1024 / PAGE_SIZE)
164 1.220 hannken
165 1.1 brezak
166 1.108 thorpej static void vndclear(struct vnd_softc *, int);
167 1.199 dyoung static int vnddoclear(struct vnd_softc *, int, int, bool);
168 1.147 elad static int vndsetcred(struct vnd_softc *, kauth_cred_t);
169 1.108 thorpej static void vndthrottle(struct vnd_softc *, struct vnode *);
170 1.108 thorpej static void vndiodone(struct buf *);
171 1.95 drochner #if 0
172 1.108 thorpej static void vndshutdown(void);
173 1.95 drochner #endif
174 1.16 cgd
175 1.108 thorpej static void vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
176 1.131 cube static void vndgetdisklabel(dev_t, struct vnd_softc *);
177 1.41 thorpej
178 1.108 thorpej static int vndlock(struct vnd_softc *);
179 1.108 thorpej static void vndunlock(struct vnd_softc *);
180 1.115 hubertf #ifdef VND_COMPRESSION
181 1.115 hubertf static void compstrategy(struct buf *, off_t);
182 1.115 hubertf static void *vnd_alloc(void *, u_int, u_int);
183 1.115 hubertf static void vnd_free(void *, void *);
184 1.115 hubertf #endif /* VND_COMPRESSION */
185 1.86 gehenna
186 1.157 jmmv static void vndthread(void *);
187 1.162 thorpej static bool vnode_has_op(const struct vnode *, int);
188 1.157 jmmv static void handle_with_rdwr(struct vnd_softc *, const struct buf *,
189 1.157 jmmv struct buf *);
190 1.157 jmmv static void handle_with_strategy(struct vnd_softc *, const struct buf *,
191 1.157 jmmv struct buf *);
192 1.222 christos static void vnd_set_geometry(struct vnd_softc *);
193 1.112 bouyer
194 1.108 thorpej static dev_type_open(vndopen);
195 1.108 thorpej static dev_type_close(vndclose);
196 1.108 thorpej static dev_type_read(vndread);
197 1.108 thorpej static dev_type_write(vndwrite);
198 1.108 thorpej static dev_type_ioctl(vndioctl);
199 1.108 thorpej static dev_type_strategy(vndstrategy);
200 1.108 thorpej static dev_type_dump(vnddump);
201 1.108 thorpej static dev_type_size(vndsize);
202 1.86 gehenna
203 1.86 gehenna const struct bdevsw vnd_bdevsw = {
204 1.228 dholland .d_open = vndopen,
205 1.228 dholland .d_close = vndclose,
206 1.228 dholland .d_strategy = vndstrategy,
207 1.228 dholland .d_ioctl = vndioctl,
208 1.228 dholland .d_dump = vnddump,
209 1.228 dholland .d_psize = vndsize,
210 1.231 dholland .d_discard = nodiscard,
211 1.228 dholland .d_flag = D_DISK
212 1.86 gehenna };
213 1.86 gehenna
214 1.86 gehenna const struct cdevsw vnd_cdevsw = {
215 1.228 dholland .d_open = vndopen,
216 1.228 dholland .d_close = vndclose,
217 1.228 dholland .d_read = vndread,
218 1.228 dholland .d_write = vndwrite,
219 1.228 dholland .d_ioctl = vndioctl,
220 1.228 dholland .d_stop = nostop,
221 1.228 dholland .d_tty = notty,
222 1.228 dholland .d_poll = nopoll,
223 1.228 dholland .d_mmap = nommap,
224 1.228 dholland .d_kqfilter = nokqfilter,
225 1.232 dholland .d_discard = nodiscard,
226 1.228 dholland .d_flag = D_DISK
227 1.86 gehenna };
228 1.22 thorpej
229 1.176 cube static int vnd_match(device_t, cfdata_t, void *);
230 1.176 cube static void vnd_attach(device_t, device_t, void *);
231 1.176 cube static int vnd_detach(device_t, int);
232 1.131 cube
233 1.199 dyoung CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
234 1.199 dyoung vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
235 1.131 cube
236 1.131 cube static struct vnd_softc *vnd_spawn(int);
237 1.176 cube int vnd_destroy(device_t);
238 1.89 mrg
239 1.243 mlelstv static struct dkdriver vnddkdriver = {
240 1.243 mlelstv .d_strategy = vndstrategy,
241 1.243 mlelstv .d_minphys = minphys
242 1.243 mlelstv };
243 1.210 riz
244 1.1 brezak void
245 1.160 christos vndattach(int num)
246 1.1 brezak {
247 1.131 cube int error;
248 1.131 cube
249 1.131 cube error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
250 1.131 cube if (error)
251 1.246 prlw1 aprint_error("%s: unable to register cfattach, error = %d\n",
252 1.246 prlw1 vnd_cd.cd_name, error);
253 1.131 cube }
254 1.1 brezak
255 1.131 cube static int
256 1.176 cube vnd_match(device_t self, cfdata_t cfdata, void *aux)
257 1.131 cube {
258 1.176 cube
259 1.131 cube return 1;
260 1.76 mrg }
261 1.76 mrg
262 1.131 cube static void
263 1.176 cube vnd_attach(device_t parent, device_t self, void *aux)
264 1.76 mrg {
265 1.176 cube struct vnd_softc *sc = device_private(self);
266 1.131 cube
267 1.176 cube sc->sc_dev = self;
268 1.131 cube sc->sc_comp_offsets = NULL;
269 1.131 cube sc->sc_comp_buff = NULL;
270 1.131 cube sc->sc_comp_decombuf = NULL;
271 1.131 cube bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
272 1.210 riz disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
273 1.173 smb if (!pmf_device_register(self, NULL, NULL))
274 1.173 smb aprint_error_dev(self, "couldn't establish power handler\n");
275 1.131 cube }
276 1.82 hannken
277 1.131 cube static int
278 1.176 cube vnd_detach(device_t self, int flags)
279 1.131 cube {
280 1.199 dyoung int error;
281 1.176 cube struct vnd_softc *sc = device_private(self);
282 1.199 dyoung
283 1.199 dyoung if (sc->sc_flags & VNF_INITED) {
284 1.199 dyoung error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
285 1.199 dyoung if (error != 0)
286 1.199 dyoung return error;
287 1.199 dyoung }
288 1.132 cube
289 1.173 smb pmf_device_deregister(self);
290 1.132 cube bufq_free(sc->sc_tab);
291 1.170 ad disk_destroy(&sc->sc_dkdev);
292 1.132 cube
293 1.131 cube return 0;
294 1.131 cube }
295 1.89 mrg
296 1.131 cube static struct vnd_softc *
297 1.131 cube vnd_spawn(int unit)
298 1.131 cube {
299 1.201 cegger cfdata_t cf;
300 1.76 mrg
301 1.131 cube cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
302 1.131 cube cf->cf_name = vnd_cd.cd_name;
303 1.131 cube cf->cf_atname = vnd_cd.cd_name;
304 1.131 cube cf->cf_unit = unit;
305 1.131 cube cf->cf_fstate = FSTATE_STAR;
306 1.89 mrg
307 1.176 cube return device_private(config_attach_pseudo(cf));
308 1.1 brezak }
309 1.1 brezak
310 1.133 cube int
311 1.176 cube vnd_destroy(device_t dev)
312 1.133 cube {
313 1.133 cube int error;
314 1.176 cube cfdata_t cf;
315 1.133 cube
316 1.145 thorpej cf = device_cfdata(dev);
317 1.146 cube error = config_detach(dev, DETACH_QUIET);
318 1.133 cube if (error)
319 1.133 cube return error;
320 1.136 yamt free(cf, M_DEVBUF);
321 1.133 cube return 0;
322 1.133 cube }
323 1.133 cube
324 1.108 thorpej static int
325 1.160 christos vndopen(dev_t dev, int flags, int mode, struct lwp *l)
326 1.1 brezak {
327 1.17 cgd int unit = vndunit(dev);
328 1.22 thorpej struct vnd_softc *sc;
329 1.22 thorpej int error = 0, part, pmask;
330 1.41 thorpej struct disklabel *lp;
331 1.1 brezak
332 1.1 brezak #ifdef DEBUG
333 1.17 cgd if (vnddebug & VDB_FOLLOW)
334 1.191 cegger printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
335 1.1 brezak #endif
336 1.182 cegger sc = device_lookup_private(&vnd_cd, unit);
337 1.131 cube if (sc == NULL) {
338 1.131 cube sc = vnd_spawn(unit);
339 1.131 cube if (sc == NULL)
340 1.131 cube return ENOMEM;
341 1.242 mlelstv
342 1.242 mlelstv /* compatibility, keep disklabel after close */
343 1.242 mlelstv sc->sc_flags = VNF_KLABEL;
344 1.131 cube }
345 1.22 thorpej
346 1.24 christos if ((error = vndlock(sc)) != 0)
347 1.197 dyoung return error;
348 1.22 thorpej
349 1.247 mlelstv mutex_enter(&sc->sc_dkdev.dk_openlock);
350 1.247 mlelstv
351 1.202 dyoung if ((sc->sc_flags & VNF_CLEARING) != 0) {
352 1.202 dyoung error = ENXIO;
353 1.202 dyoung goto done;
354 1.202 dyoung }
355 1.199 dyoung
356 1.41 thorpej lp = sc->sc_dkdev.dk_label;
357 1.41 thorpej
358 1.22 thorpej part = DISKPART(dev);
359 1.22 thorpej pmask = (1 << part);
360 1.22 thorpej
361 1.221 mlelstv if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
362 1.221 mlelstv error = EBUSY;
363 1.221 mlelstv goto done;
364 1.221 mlelstv }
365 1.221 mlelstv
366 1.221 mlelstv if (sc->sc_flags & VNF_INITED) {
367 1.221 mlelstv if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
368 1.221 mlelstv /*
369 1.221 mlelstv * If any non-raw partition is open, but the disk
370 1.221 mlelstv * has been invalidated, disallow further opens.
371 1.221 mlelstv */
372 1.221 mlelstv if ((sc->sc_flags & VNF_VLABEL) == 0) {
373 1.221 mlelstv error = EIO;
374 1.221 mlelstv goto done;
375 1.221 mlelstv }
376 1.221 mlelstv } else {
377 1.221 mlelstv /*
378 1.221 mlelstv * Load the partition info if not already loaded.
379 1.221 mlelstv */
380 1.221 mlelstv if ((sc->sc_flags & VNF_VLABEL) == 0) {
381 1.221 mlelstv sc->sc_flags |= VNF_VLABEL;
382 1.221 mlelstv vndgetdisklabel(dev, sc);
383 1.221 mlelstv }
384 1.221 mlelstv }
385 1.221 mlelstv }
386 1.41 thorpej
387 1.41 thorpej /* Check that the partitions exists. */
388 1.41 thorpej if (part != RAW_PART) {
389 1.41 thorpej if (((sc->sc_flags & VNF_INITED) == 0) ||
390 1.41 thorpej ((part >= lp->d_npartitions) ||
391 1.41 thorpej (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
392 1.41 thorpej error = ENXIO;
393 1.41 thorpej goto done;
394 1.41 thorpej }
395 1.41 thorpej }
396 1.41 thorpej
397 1.22 thorpej /* Prevent our unit from being unconfigured while open. */
398 1.22 thorpej switch (mode) {
399 1.22 thorpej case S_IFCHR:
400 1.22 thorpej sc->sc_dkdev.dk_copenmask |= pmask;
401 1.22 thorpej break;
402 1.22 thorpej
403 1.22 thorpej case S_IFBLK:
404 1.22 thorpej sc->sc_dkdev.dk_bopenmask |= pmask;
405 1.22 thorpej break;
406 1.22 thorpej }
407 1.22 thorpej sc->sc_dkdev.dk_openmask =
408 1.22 thorpej sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
409 1.22 thorpej
410 1.41 thorpej done:
411 1.247 mlelstv mutex_exit(&sc->sc_dkdev.dk_openlock);
412 1.22 thorpej vndunlock(sc);
413 1.197 dyoung return error;
414 1.1 brezak }
415 1.1 brezak
416 1.108 thorpej static int
417 1.160 christos vndclose(dev_t dev, int flags, int mode, struct lwp *l)
418 1.4 deraadt {
419 1.22 thorpej int unit = vndunit(dev);
420 1.22 thorpej struct vnd_softc *sc;
421 1.22 thorpej int error = 0, part;
422 1.22 thorpej
423 1.4 deraadt #ifdef DEBUG
424 1.17 cgd if (vnddebug & VDB_FOLLOW)
425 1.191 cegger printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
426 1.4 deraadt #endif
427 1.182 cegger sc = device_lookup_private(&vnd_cd, unit);
428 1.131 cube if (sc == NULL)
429 1.131 cube return ENXIO;
430 1.22 thorpej
431 1.24 christos if ((error = vndlock(sc)) != 0)
432 1.197 dyoung return error;
433 1.22 thorpej
434 1.247 mlelstv mutex_enter(&sc->sc_dkdev.dk_openlock);
435 1.247 mlelstv
436 1.22 thorpej part = DISKPART(dev);
437 1.22 thorpej
438 1.22 thorpej /* ...that much closer to allowing unconfiguration... */
439 1.22 thorpej switch (mode) {
440 1.22 thorpej case S_IFCHR:
441 1.22 thorpej sc->sc_dkdev.dk_copenmask &= ~(1 << part);
442 1.22 thorpej break;
443 1.22 thorpej
444 1.22 thorpej case S_IFBLK:
445 1.22 thorpej sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
446 1.22 thorpej break;
447 1.22 thorpej }
448 1.22 thorpej sc->sc_dkdev.dk_openmask =
449 1.22 thorpej sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
450 1.22 thorpej
451 1.242 mlelstv /* are we last opener ? */
452 1.242 mlelstv if (sc->sc_dkdev.dk_openmask == 0) {
453 1.242 mlelstv if ((sc->sc_flags & VNF_KLABEL) == 0)
454 1.242 mlelstv sc->sc_flags &= ~VNF_VLABEL;
455 1.242 mlelstv }
456 1.242 mlelstv
457 1.247 mlelstv mutex_exit(&sc->sc_dkdev.dk_openlock);
458 1.247 mlelstv
459 1.22 thorpej vndunlock(sc);
460 1.134 cube
461 1.134 cube if ((sc->sc_flags & VNF_INITED) == 0) {
462 1.176 cube if ((error = vnd_destroy(sc->sc_dev)) != 0) {
463 1.176 cube aprint_error_dev(sc->sc_dev,
464 1.176 cube "unable to detach instance\n");
465 1.134 cube return error;
466 1.134 cube }
467 1.134 cube }
468 1.134 cube
469 1.197 dyoung return 0;
470 1.4 deraadt }
471 1.4 deraadt
472 1.1 brezak /*
473 1.135 cube * Queue the request, and wakeup the kernel thread to handle it.
474 1.1 brezak */
475 1.108 thorpej static void
476 1.108 thorpej vndstrategy(struct buf *bp)
477 1.1 brezak {
478 1.17 cgd int unit = vndunit(bp->b_dev);
479 1.131 cube struct vnd_softc *vnd =
480 1.182 cegger device_lookup_private(&vnd_cd, unit);
481 1.189 bouyer struct disklabel *lp;
482 1.127 yamt daddr_t blkno;
483 1.112 bouyer int s = splbio();
484 1.112 bouyer
485 1.189 bouyer if (vnd == NULL) {
486 1.189 bouyer bp->b_error = ENXIO;
487 1.189 bouyer goto done;
488 1.189 bouyer }
489 1.189 bouyer lp = vnd->sc_dkdev.dk_label;
490 1.189 bouyer
491 1.17 cgd if ((vnd->sc_flags & VNF_INITED) == 0) {
492 1.1 brezak bp->b_error = ENXIO;
493 1.169 ad goto done;
494 1.41 thorpej }
495 1.41 thorpej
496 1.41 thorpej /*
497 1.41 thorpej * The transfer must be a whole number of blocks.
498 1.41 thorpej */
499 1.41 thorpej if ((bp->b_bcount % lp->d_secsize) != 0) {
500 1.41 thorpej bp->b_error = EINVAL;
501 1.169 ad goto done;
502 1.1 brezak }
503 1.41 thorpej
504 1.41 thorpej /*
505 1.94 yamt * check if we're read-only.
506 1.94 yamt */
507 1.94 yamt if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
508 1.94 yamt bp->b_error = EACCES;
509 1.169 ad goto done;
510 1.165 yamt }
511 1.165 yamt
512 1.165 yamt /* If it's a nil transfer, wake up the top half now. */
513 1.165 yamt if (bp->b_bcount == 0) {
514 1.94 yamt goto done;
515 1.94 yamt }
516 1.94 yamt
517 1.41 thorpej /*
518 1.112 bouyer * Do bounds checking and adjust transfer. If there's an error,
519 1.112 bouyer * the bounds check will flag that for us.
520 1.41 thorpej */
521 1.139 yamt if (DISKPART(bp->b_dev) == RAW_PART) {
522 1.139 yamt if (bounds_check_with_mediasize(bp, DEV_BSIZE,
523 1.139 yamt vnd->sc_size) <= 0)
524 1.139 yamt goto done;
525 1.139 yamt } else {
526 1.112 bouyer if (bounds_check_with_label(&vnd->sc_dkdev,
527 1.112 bouyer bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
528 1.112 bouyer goto done;
529 1.1 brezak }
530 1.127 yamt
531 1.127 yamt /*
532 1.127 yamt * Put the block number in terms of the logical blocksize
533 1.127 yamt * of the "device".
534 1.127 yamt */
535 1.127 yamt
536 1.127 yamt blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
537 1.127 yamt
538 1.127 yamt /*
539 1.127 yamt * Translate the partition-relative block number to an absolute.
540 1.127 yamt */
541 1.127 yamt if (DISKPART(bp->b_dev) != RAW_PART) {
542 1.127 yamt struct partition *pp;
543 1.127 yamt
544 1.127 yamt pp = &vnd->sc_dkdev.dk_label->d_partitions[
545 1.127 yamt DISKPART(bp->b_dev)];
546 1.127 yamt blkno += pp->p_offset;
547 1.127 yamt }
548 1.127 yamt bp->b_rawblkno = blkno;
549 1.127 yamt
550 1.112 bouyer #ifdef DEBUG
551 1.112 bouyer if (vnddebug & VDB_FOLLOW)
552 1.112 bouyer printf("vndstrategy(%p): unit %d\n", bp, unit);
553 1.112 bouyer #endif
554 1.220 hannken if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
555 1.220 hannken KASSERT(vnd->sc_pending >= 0 &&
556 1.220 hannken vnd->sc_pending <= VND_MAXPENDING(vnd));
557 1.220 hannken while (vnd->sc_pending == VND_MAXPENDING(vnd))
558 1.220 hannken tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
559 1.220 hannken vnd->sc_pending++;
560 1.220 hannken }
561 1.192 yamt bufq_put(vnd->sc_tab, bp);
562 1.112 bouyer wakeup(&vnd->sc_tab);
563 1.112 bouyer splx(s);
564 1.112 bouyer return;
565 1.165 yamt
566 1.112 bouyer done:
567 1.165 yamt bp->b_resid = bp->b_bcount;
568 1.112 bouyer biodone(bp);
569 1.112 bouyer splx(s);
570 1.112 bouyer }
571 1.41 thorpej
572 1.184 cegger static bool
573 1.184 cegger vnode_has_strategy(struct vnd_softc *vnd)
574 1.184 cegger {
575 1.184 cegger return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
576 1.184 cegger vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
577 1.184 cegger }
578 1.184 cegger
579 1.265 mlelstv /* Verify that I/O requests cannot be smaller than the
580 1.265 mlelstv * smallest I/O size supported by the backend.
581 1.265 mlelstv */
582 1.255 mlelstv static bool
583 1.255 mlelstv vnode_has_large_blocks(struct vnd_softc *vnd)
584 1.255 mlelstv {
585 1.265 mlelstv u_int32_t vnd_secsize, iosize;
586 1.255 mlelstv
587 1.265 mlelstv iosize = vnd->sc_iosize;
588 1.255 mlelstv vnd_secsize = vnd->sc_geom.vng_secsize;
589 1.255 mlelstv
590 1.265 mlelstv return vnd_secsize % iosize != 0;
591 1.255 mlelstv }
592 1.255 mlelstv
593 1.185 cegger /* XXX this function needs a reliable check to detect
594 1.185 cegger * sparse files. Otherwise, bmap/strategy may be used
595 1.185 cegger * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
596 1.185 cegger * works on sparse files.
597 1.185 cegger */
598 1.185 cegger #if notyet
599 1.184 cegger static bool
600 1.184 cegger vnode_strategy_probe(struct vnd_softc *vnd)
601 1.184 cegger {
602 1.184 cegger int error;
603 1.184 cegger daddr_t nbn;
604 1.184 cegger
605 1.184 cegger if (!vnode_has_strategy(vnd))
606 1.184 cegger return false;
607 1.184 cegger
608 1.255 mlelstv if (vnode_has_large_blocks(vnd))
609 1.255 mlelstv return false;
610 1.255 mlelstv
611 1.184 cegger /* Convert the first logical block number to its
612 1.184 cegger * physical block number.
613 1.184 cegger */
614 1.184 cegger error = 0;
615 1.200 ad vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
616 1.184 cegger error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
617 1.209 hannken VOP_UNLOCK(vnd->sc_vp);
618 1.184 cegger
619 1.184 cegger /* Test if that worked. */
620 1.184 cegger if (error == 0 && (long)nbn == -1)
621 1.184 cegger return false;
622 1.184 cegger
623 1.184 cegger return true;
624 1.184 cegger }
625 1.185 cegger #endif
626 1.184 cegger
627 1.126 yamt static void
628 1.112 bouyer vndthread(void *arg)
629 1.112 bouyer {
630 1.112 bouyer struct vnd_softc *vnd = arg;
631 1.157 jmmv int s;
632 1.157 jmmv
633 1.184 cegger /* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
634 1.157 jmmv * directly access the backing vnode. If we can, use these two
635 1.157 jmmv * operations to avoid messing with the local buffer cache.
636 1.157 jmmv * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
637 1.157 jmmv * which are guaranteed to work with any file system. */
638 1.218 hannken if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
639 1.218 hannken ! vnode_has_strategy(vnd))
640 1.218 hannken vnd->sc_flags |= VNF_USE_VN_RDWR;
641 1.157 jmmv
642 1.255 mlelstv /* VOP_STRATEGY can only be used if the backing vnode allows
643 1.255 mlelstv * to access blocks as small as defined by the vnd geometry.
644 1.255 mlelstv */
645 1.255 mlelstv if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
646 1.255 mlelstv vnode_has_large_blocks(vnd))
647 1.255 mlelstv vnd->sc_flags |= VNF_USE_VN_RDWR;
648 1.255 mlelstv
649 1.157 jmmv #ifdef DEBUG
650 1.157 jmmv if (vnddebug & VDB_INIT)
651 1.157 jmmv printf("vndthread: vp %p, %s\n", vnd->sc_vp,
652 1.218 hannken (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
653 1.157 jmmv "using bmap/strategy operations" :
654 1.157 jmmv "using read/write operations");
655 1.157 jmmv #endif
656 1.36 pk
657 1.61 thorpej s = splbio();
658 1.112 bouyer vnd->sc_flags |= VNF_KTHREAD;
659 1.112 bouyer wakeup(&vnd->sc_kthread);
660 1.37 pk
661 1.112 bouyer /*
662 1.157 jmmv * Dequeue requests and serve them depending on the available
663 1.157 jmmv * vnode operations.
664 1.112 bouyer */
665 1.112 bouyer while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
666 1.129 yamt struct vndxfer *vnx;
667 1.129 yamt struct buf *obp;
668 1.129 yamt struct buf *bp;
669 1.129 yamt
670 1.192 yamt obp = bufq_get(vnd->sc_tab);
671 1.129 yamt if (obp == NULL) {
672 1.112 bouyer tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
673 1.112 bouyer continue;
674 1.112 bouyer };
675 1.220 hannken if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
676 1.220 hannken KASSERT(vnd->sc_pending > 0 &&
677 1.220 hannken vnd->sc_pending <= VND_MAXPENDING(vnd));
678 1.220 hannken if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
679 1.220 hannken wakeup(&vnd->sc_pending);
680 1.220 hannken }
681 1.112 bouyer splx(s);
682 1.5 cgd #ifdef DEBUG
683 1.112 bouyer if (vnddebug & VDB_FOLLOW)
684 1.183 cegger printf("vndthread(%p)\n", obp);
685 1.1 brezak #endif
686 1.18 cgd
687 1.112 bouyer if (vnd->sc_vp->v_mount == NULL) {
688 1.129 yamt obp->b_error = ENXIO;
689 1.112 bouyer goto done;
690 1.112 bouyer }
691 1.115 hubertf #ifdef VND_COMPRESSION
692 1.115 hubertf /* handle a compressed read */
693 1.226 martin if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
694 1.157 jmmv off_t bn;
695 1.244 prlw1
696 1.157 jmmv /* Convert to a byte offset within the file. */
697 1.157 jmmv bn = obp->b_rawblkno *
698 1.157 jmmv vnd->sc_dkdev.dk_label->d_secsize;
699 1.157 jmmv
700 1.129 yamt compstrategy(obp, bn);
701 1.115 hubertf goto done;
702 1.115 hubertf }
703 1.115 hubertf #endif /* VND_COMPRESSION */
704 1.244 prlw1
705 1.1 brezak /*
706 1.112 bouyer * Allocate a header for this transfer and link it to the
707 1.112 bouyer * buffer
708 1.1 brezak */
709 1.1 brezak s = splbio();
710 1.112 bouyer vnx = VND_GETXFER(vnd);
711 1.1 brezak splx(s);
712 1.129 yamt vnx->vx_vnd = vnd;
713 1.129 yamt
714 1.157 jmmv s = splbio();
715 1.157 jmmv while (vnd->sc_active >= vnd->sc_maxactive) {
716 1.157 jmmv tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
717 1.157 jmmv }
718 1.157 jmmv vnd->sc_active++;
719 1.157 jmmv splx(s);
720 1.157 jmmv
721 1.157 jmmv /* Instrumentation. */
722 1.157 jmmv disk_busy(&vnd->sc_dkdev);
723 1.157 jmmv
724 1.129 yamt bp = &vnx->vx_buf;
725 1.175 ad buf_init(bp);
726 1.175 ad bp->b_flags = (obp->b_flags & B_READ);
727 1.175 ad bp->b_oflags = obp->b_oflags;
728 1.175 ad bp->b_cflags = obp->b_cflags;
729 1.129 yamt bp->b_iodone = vndiodone;
730 1.129 yamt bp->b_private = obp;
731 1.138 yamt bp->b_vp = vnd->sc_vp;
732 1.217 rmind bp->b_objlock = bp->b_vp->v_interlock;
733 1.129 yamt bp->b_data = obp->b_data;
734 1.165 yamt bp->b_bcount = obp->b_bcount;
735 1.129 yamt BIO_COPYPRIO(bp, obp);
736 1.129 yamt
737 1.270 hannken /* Make sure the request succeeds while suspending this fs. */
738 1.270 hannken fstrans_start_lazy(vnd->sc_vp->v_mount);
739 1.270 hannken
740 1.157 jmmv /* Handle the request using the appropriate operations. */
741 1.218 hannken if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
742 1.157 jmmv handle_with_strategy(vnd, obp, bp);
743 1.157 jmmv else
744 1.157 jmmv handle_with_rdwr(vnd, obp, bp);
745 1.157 jmmv
746 1.270 hannken fstrans_done(vnd->sc_vp->v_mount);
747 1.270 hannken
748 1.157 jmmv s = splbio();
749 1.157 jmmv continue;
750 1.157 jmmv
751 1.157 jmmv done:
752 1.157 jmmv biodone(obp);
753 1.129 yamt s = splbio();
754 1.157 jmmv }
755 1.157 jmmv
756 1.157 jmmv vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
757 1.157 jmmv wakeup(&vnd->sc_kthread);
758 1.157 jmmv splx(s);
759 1.157 jmmv kthread_exit(0);
760 1.157 jmmv }
761 1.157 jmmv
762 1.157 jmmv /*
763 1.157 jmmv * Checks if the given vnode supports the requested operation.
764 1.157 jmmv * The operation is specified the offset returned by VOFFSET.
765 1.157 jmmv *
766 1.157 jmmv * XXX The test below used to determine this is quite fragile
767 1.157 jmmv * because it relies on the file system to use genfs to specify
768 1.157 jmmv * unimplemented operations. There might be another way to do
769 1.157 jmmv * it more cleanly.
770 1.157 jmmv */
771 1.162 thorpej static bool
772 1.157 jmmv vnode_has_op(const struct vnode *vp, int opoffset)
773 1.157 jmmv {
774 1.157 jmmv int (*defaultp)(void *);
775 1.157 jmmv int (*opp)(void *);
776 1.157 jmmv
777 1.157 jmmv defaultp = vp->v_op[VOFFSET(vop_default)];
778 1.157 jmmv opp = vp->v_op[opoffset];
779 1.157 jmmv
780 1.157 jmmv return opp != defaultp && opp != genfs_eopnotsupp &&
781 1.157 jmmv opp != genfs_badop && opp != genfs_nullop;
782 1.157 jmmv }
783 1.157 jmmv
784 1.157 jmmv /*
785 1.245 prlw1 * Handles the read/write request given in 'bp' using the vnode's VOP_READ
786 1.157 jmmv * and VOP_WRITE operations.
787 1.157 jmmv *
788 1.157 jmmv * 'obp' is a pointer to the original request fed to the vnd device.
789 1.157 jmmv */
790 1.157 jmmv static void
791 1.157 jmmv handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
792 1.157 jmmv {
793 1.162 thorpej bool doread;
794 1.157 jmmv off_t offset;
795 1.218 hannken size_t len, resid;
796 1.157 jmmv struct vnode *vp;
797 1.157 jmmv
798 1.157 jmmv doread = bp->b_flags & B_READ;
799 1.157 jmmv offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
800 1.218 hannken len = bp->b_bcount;
801 1.157 jmmv vp = vnd->sc_vp;
802 1.157 jmmv
803 1.157 jmmv #if defined(DEBUG)
804 1.157 jmmv if (vnddebug & VDB_IO)
805 1.157 jmmv printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
806 1.157 jmmv ", secsize %d, offset %" PRIu64
807 1.161 chs ", bcount %d\n",
808 1.157 jmmv vp, doread ? "read" : "write", obp->b_rawblkno,
809 1.157 jmmv vnd->sc_dkdev.dk_label->d_secsize, offset,
810 1.161 chs bp->b_bcount);
811 1.157 jmmv #endif
812 1.157 jmmv
813 1.157 jmmv /* Issue the read or write operation. */
814 1.157 jmmv bp->b_error =
815 1.157 jmmv vn_rdwr(doread ? UIO_READ : UIO_WRITE,
816 1.218 hannken vp, bp->b_data, len, offset, UIO_SYSSPACE,
817 1.269 mlelstv IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_DIRECT,
818 1.269 mlelstv vnd->sc_cred, &resid, NULL);
819 1.158 martin bp->b_resid = resid;
820 1.157 jmmv
821 1.269 mlelstv /*
822 1.269 mlelstv * Avoid caching too many pages, the vnd user
823 1.269 mlelstv * is usually a filesystem and caches itself.
824 1.269 mlelstv * We need some amount of caching to not hinder
825 1.269 mlelstv * read-ahead and write-behind operations.
826 1.269 mlelstv */
827 1.218 hannken mutex_enter(vp->v_interlock);
828 1.269 mlelstv if (vp->v_uobj.uo_npages > VND_MAXPAGES(vnd))
829 1.269 mlelstv (void) VOP_PUTPAGES(vp, 0, 0,
830 1.269 mlelstv PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE);
831 1.269 mlelstv else
832 1.269 mlelstv mutex_exit(vp->v_interlock);
833 1.218 hannken
834 1.157 jmmv /* We need to increase the number of outputs on the vnode if
835 1.159 jmmv * there was any write to it. */
836 1.175 ad if (!doread) {
837 1.217 rmind mutex_enter(vp->v_interlock);
838 1.175 ad vp->v_numoutput++;
839 1.217 rmind mutex_exit(vp->v_interlock);
840 1.175 ad }
841 1.157 jmmv
842 1.157 jmmv biodone(bp);
843 1.157 jmmv }
844 1.157 jmmv
845 1.157 jmmv /*
846 1.157 jmmv * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
847 1.157 jmmv * and VOP_STRATEGY operations.
848 1.157 jmmv *
849 1.157 jmmv * 'obp' is a pointer to the original request fed to the vnd device.
850 1.157 jmmv */
851 1.157 jmmv static void
852 1.157 jmmv handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
853 1.157 jmmv struct buf *bp)
854 1.157 jmmv {
855 1.157 jmmv int bsize, error, flags, skipped;
856 1.157 jmmv size_t resid, sz;
857 1.157 jmmv off_t bn, offset;
858 1.175 ad struct vnode *vp;
859 1.240 bouyer struct buf *nbp = NULL;
860 1.157 jmmv
861 1.157 jmmv flags = obp->b_flags;
862 1.157 jmmv
863 1.157 jmmv
864 1.157 jmmv /* convert to a byte offset within the file. */
865 1.157 jmmv bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
866 1.157 jmmv
867 1.157 jmmv bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
868 1.157 jmmv skipped = 0;
869 1.138 yamt
870 1.157 jmmv /*
871 1.157 jmmv * Break the request into bsize pieces and feed them
872 1.157 jmmv * sequentially using VOP_BMAP/VOP_STRATEGY.
873 1.157 jmmv * We do it this way to keep from flooding NFS servers if we
874 1.157 jmmv * are connected to an NFS file. This places the burden on
875 1.157 jmmv * the client rather than the server.
876 1.157 jmmv */
877 1.157 jmmv error = 0;
878 1.165 yamt bp->b_resid = bp->b_bcount;
879 1.240 bouyer for (offset = 0, resid = bp->b_resid; /* true */;
880 1.157 jmmv resid -= sz, offset += sz) {
881 1.157 jmmv daddr_t nbn;
882 1.157 jmmv int off, nra;
883 1.157 jmmv
884 1.157 jmmv nra = 0;
885 1.200 ad vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
886 1.157 jmmv error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
887 1.209 hannken VOP_UNLOCK(vnd->sc_vp);
888 1.51 pk
889 1.157 jmmv if (error == 0 && (long)nbn == -1)
890 1.157 jmmv error = EIO;
891 1.126 yamt
892 1.112 bouyer /*
893 1.157 jmmv * If there was an error or a hole in the file...punt.
894 1.157 jmmv * Note that we may have to wait for any operations
895 1.157 jmmv * that we have already fired off before releasing
896 1.157 jmmv * the buffer.
897 1.157 jmmv *
898 1.157 jmmv * XXX we could deal with holes here but it would be
899 1.157 jmmv * a hassle (in the write case).
900 1.112 bouyer */
901 1.157 jmmv if (error) {
902 1.157 jmmv skipped += resid;
903 1.157 jmmv break;
904 1.157 jmmv }
905 1.41 thorpej
906 1.157 jmmv #ifdef DEBUG
907 1.157 jmmv if (!dovndcluster)
908 1.112 bouyer nra = 0;
909 1.112 bouyer #endif
910 1.51 pk
911 1.157 jmmv off = bn % bsize;
912 1.157 jmmv sz = MIN(((off_t)1 + nra) * bsize - off, resid);
913 1.116 christos #ifdef DEBUG
914 1.157 jmmv if (vnddebug & VDB_IO)
915 1.157 jmmv printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
916 1.175 ad " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
917 1.175 ad nbn, sz);
918 1.112 bouyer #endif
919 1.51 pk
920 1.175 ad nbp = getiobuf(vp, true);
921 1.157 jmmv nestiobuf_setup(bp, nbp, offset, sz);
922 1.157 jmmv nbp->b_blkno = nbn + btodb(off);
923 1.51 pk
924 1.129 yamt #if 0 /* XXX #ifdef DEBUG */
925 1.157 jmmv if (vnddebug & VDB_IO)
926 1.157 jmmv printf("vndstart(%ld): bp %p vp %p blkno "
927 1.157 jmmv "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
928 1.157 jmmv (long) (vnd-vnd_softc), &nbp->vb_buf,
929 1.157 jmmv nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
930 1.157 jmmv nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
931 1.157 jmmv nbp->vb_buf.b_bcount);
932 1.157 jmmv #endif
933 1.240 bouyer if (resid == sz) {
934 1.240 bouyer break;
935 1.240 bouyer }
936 1.157 jmmv VOP_STRATEGY(vp, nbp);
937 1.157 jmmv bn += sz;
938 1.157 jmmv }
939 1.240 bouyer if (!(flags & B_READ)) {
940 1.240 bouyer struct vnode *w_vp;
941 1.240 bouyer /*
942 1.240 bouyer * this is the last nested buf, account for
943 1.240 bouyer * the parent buf write too.
944 1.240 bouyer * This has to be done last, so that
945 1.240 bouyer * fsync won't wait for this write which
946 1.241 bouyer * has no chance to complete before all nested bufs
947 1.240 bouyer * have been queued. But it has to be done
948 1.244 prlw1 * before the last VOP_STRATEGY()
949 1.240 bouyer * or the call to nestiobuf_done().
950 1.240 bouyer */
951 1.240 bouyer w_vp = bp->b_vp;
952 1.240 bouyer mutex_enter(w_vp->v_interlock);
953 1.240 bouyer w_vp->v_numoutput++;
954 1.240 bouyer mutex_exit(w_vp->v_interlock);
955 1.240 bouyer }
956 1.240 bouyer KASSERT(skipped != 0 || nbp != NULL);
957 1.244 prlw1 if (skipped)
958 1.240 bouyer nestiobuf_done(bp, skipped, error);
959 1.244 prlw1 else
960 1.240 bouyer VOP_STRATEGY(vp, nbp);
961 1.1 brezak }
962 1.1 brezak
963 1.126 yamt static void
964 1.129 yamt vndiodone(struct buf *bp)
965 1.126 yamt {
966 1.129 yamt struct vndxfer *vnx = VND_BUFTOXFER(bp);
967 1.129 yamt struct vnd_softc *vnd = vnx->vx_vnd;
968 1.129 yamt struct buf *obp = bp->b_private;
969 1.206 bouyer int s = splbio();
970 1.126 yamt
971 1.273 ad KERNEL_LOCK(1, NULL); /* XXXSMP */
972 1.129 yamt KASSERT(&vnx->vx_buf == bp);
973 1.129 yamt KASSERT(vnd->sc_active > 0);
974 1.126 yamt #ifdef DEBUG
975 1.126 yamt if (vnddebug & VDB_IO) {
976 1.126 yamt printf("vndiodone1: bp %p iodone: error %d\n",
977 1.169 ad bp, bp->b_error);
978 1.126 yamt }
979 1.126 yamt #endif
980 1.126 yamt disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
981 1.126 yamt (bp->b_flags & B_READ));
982 1.129 yamt vnd->sc_active--;
983 1.129 yamt if (vnd->sc_active == 0) {
984 1.129 yamt wakeup(&vnd->sc_tab);
985 1.1 brezak }
986 1.273 ad KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */
987 1.206 bouyer splx(s);
988 1.129 yamt obp->b_error = bp->b_error;
989 1.129 yamt obp->b_resid = bp->b_resid;
990 1.186 kardel buf_destroy(bp);
991 1.129 yamt VND_PUTXFER(vnd, vnx);
992 1.129 yamt biodone(obp);
993 1.20 mycroft }
994 1.20 mycroft
995 1.22 thorpej /* ARGSUSED */
996 1.108 thorpej static int
997 1.160 christos vndread(dev_t dev, struct uio *uio, int flags)
998 1.20 mycroft {
999 1.22 thorpej int unit = vndunit(dev);
1000 1.22 thorpej struct vnd_softc *sc;
1001 1.20 mycroft
1002 1.20 mycroft #ifdef DEBUG
1003 1.20 mycroft if (vnddebug & VDB_FOLLOW)
1004 1.191 cegger printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
1005 1.20 mycroft #endif
1006 1.22 thorpej
1007 1.182 cegger sc = device_lookup_private(&vnd_cd, unit);
1008 1.131 cube if (sc == NULL)
1009 1.131 cube return ENXIO;
1010 1.22 thorpej
1011 1.22 thorpej if ((sc->sc_flags & VNF_INITED) == 0)
1012 1.197 dyoung return ENXIO;
1013 1.22 thorpej
1014 1.197 dyoung return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
1015 1.20 mycroft }
1016 1.20 mycroft
1017 1.22 thorpej /* ARGSUSED */
1018 1.108 thorpej static int
1019 1.160 christos vndwrite(dev_t dev, struct uio *uio, int flags)
1020 1.20 mycroft {
1021 1.22 thorpej int unit = vndunit(dev);
1022 1.22 thorpej struct vnd_softc *sc;
1023 1.20 mycroft
1024 1.20 mycroft #ifdef DEBUG
1025 1.20 mycroft if (vnddebug & VDB_FOLLOW)
1026 1.191 cegger printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
1027 1.20 mycroft #endif
1028 1.22 thorpej
1029 1.182 cegger sc = device_lookup_private(&vnd_cd, unit);
1030 1.131 cube if (sc == NULL)
1031 1.131 cube return ENXIO;
1032 1.22 thorpej
1033 1.22 thorpej if ((sc->sc_flags & VNF_INITED) == 0)
1034 1.197 dyoung return ENXIO;
1035 1.22 thorpej
1036 1.197 dyoung return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
1037 1.1 brezak }
1038 1.1 brezak
1039 1.120 christos static int
1040 1.124 christos vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
1041 1.120 christos {
1042 1.219 hannken int error;
1043 1.120 christos struct vnd_softc *vnd;
1044 1.120 christos
1045 1.120 christos if (*un == -1)
1046 1.120 christos *un = unit;
1047 1.120 christos if (*un < 0)
1048 1.120 christos return EINVAL;
1049 1.120 christos
1050 1.182 cegger vnd = device_lookup_private(&vnd_cd, *un);
1051 1.131 cube if (vnd == NULL)
1052 1.225 christos return -1;
1053 1.120 christos
1054 1.120 christos if ((vnd->sc_flags & VNF_INITED) == 0)
1055 1.122 christos return -1;
1056 1.120 christos
1057 1.219 hannken vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1058 1.219 hannken error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1059 1.219 hannken VOP_UNLOCK(vnd->sc_vp);
1060 1.219 hannken return error;
1061 1.120 christos }
1062 1.120 christos
1063 1.199 dyoung static int
1064 1.199 dyoung vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1065 1.199 dyoung {
1066 1.199 dyoung int error;
1067 1.199 dyoung
1068 1.199 dyoung if ((error = vndlock(vnd)) != 0)
1069 1.199 dyoung return error;
1070 1.199 dyoung
1071 1.199 dyoung /*
1072 1.199 dyoung * Don't unconfigure if any other partitions are open
1073 1.199 dyoung * or if both the character and block flavors of this
1074 1.199 dyoung * partition are open.
1075 1.199 dyoung */
1076 1.204 dyoung if (DK_BUSY(vnd, pmask) && !force) {
1077 1.199 dyoung vndunlock(vnd);
1078 1.199 dyoung return EBUSY;
1079 1.199 dyoung }
1080 1.199 dyoung
1081 1.221 mlelstv /* Delete all of our wedges */
1082 1.221 mlelstv dkwedge_delall(&vnd->sc_dkdev);
1083 1.221 mlelstv
1084 1.199 dyoung /*
1085 1.199 dyoung * XXX vndclear() might call vndclose() implicitly;
1086 1.199 dyoung * release lock to avoid recursion
1087 1.199 dyoung *
1088 1.199 dyoung * Set VNF_CLEARING to prevent vndopen() from
1089 1.199 dyoung * sneaking in after we vndunlock().
1090 1.199 dyoung */
1091 1.199 dyoung vnd->sc_flags |= VNF_CLEARING;
1092 1.199 dyoung vndunlock(vnd);
1093 1.199 dyoung vndclear(vnd, minor);
1094 1.199 dyoung #ifdef DEBUG
1095 1.199 dyoung if (vnddebug & VDB_INIT)
1096 1.249 christos printf("%s: CLRed\n", __func__);
1097 1.199 dyoung #endif
1098 1.199 dyoung
1099 1.199 dyoung /* Destroy the xfer and buffer pools. */
1100 1.199 dyoung pool_destroy(&vnd->sc_vxpool);
1101 1.199 dyoung
1102 1.199 dyoung /* Detach the disk. */
1103 1.199 dyoung disk_detach(&vnd->sc_dkdev);
1104 1.199 dyoung
1105 1.199 dyoung return 0;
1106 1.199 dyoung }
1107 1.199 dyoung
1108 1.249 christos static int
1109 1.249 christos vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
1110 1.249 christos {
1111 1.249 christos int error;
1112 1.249 christos
1113 1.249 christos KASSERT(l);
1114 1.251 christos
1115 1.252 christos /* the first member is always int vnd_unit in all the versions */
1116 1.249 christos if (*(int *)data >= vnd_cd.cd_ndevs)
1117 1.249 christos return ENXIO;
1118 1.249 christos
1119 1.249 christos switch (error = vnd_cget(l, unit, (int *)data, va)) {
1120 1.249 christos case -1:
1121 1.249 christos /* unused is not an error */
1122 1.253 christos memset(va, 0, sizeof(*va));
1123 1.249 christos /*FALLTHROUGH*/
1124 1.249 christos case 0:
1125 1.249 christos return 0;
1126 1.249 christos default:
1127 1.249 christos return error;
1128 1.249 christos }
1129 1.249 christos }
1130 1.249 christos
1131 1.1 brezak /* ARGSUSED */
1132 1.108 thorpej static int
1133 1.163 christos vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1134 1.1 brezak {
1135 1.199 dyoung bool force;
1136 1.17 cgd int unit = vndunit(dev);
1137 1.65 augustss struct vnd_softc *vnd;
1138 1.17 cgd struct vnd_ioctl *vio;
1139 1.1 brezak struct vattr vattr;
1140 1.214 dholland struct pathbuf *pb;
1141 1.1 brezak struct nameidata nd;
1142 1.32 mycroft int error, part, pmask;
1143 1.223 christos uint64_t geomsize;
1144 1.94 yamt int fflags;
1145 1.70 fvdl #ifdef __HAVE_OLD_DISKLABEL
1146 1.70 fvdl struct disklabel newlabel;
1147 1.70 fvdl #endif
1148 1.1 brezak
1149 1.1 brezak #ifdef DEBUG
1150 1.17 cgd if (vnddebug & VDB_FOLLOW)
1151 1.191 cegger printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1152 1.148 ad dev, cmd, data, flag, l->l_proc, unit);
1153 1.1 brezak #endif
1154 1.249 christos /* Do the get's first; they don't need initialization or verification */
1155 1.249 christos switch (cmd) {
1156 1.271 pgoyette case VNDIOCGET:
1157 1.249 christos if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1158 1.249 christos return error;
1159 1.249 christos
1160 1.271 pgoyette struct vnd_user *vnu = data;
1161 1.249 christos vnu->vnu_dev = vattr.va_fsid;
1162 1.249 christos vnu->vnu_ino = vattr.va_fileid;
1163 1.249 christos return 0;
1164 1.249 christos
1165 1.271 pgoyette default:
1166 1.271 pgoyette /* First check for COMPAT_50 hook */
1167 1.272 pgoyette MODULE_HOOK_CALL(compat_vndioctl_50_hook,
1168 1.271 pgoyette (cmd, l, data, unit, &vattr, vndioctl_get),
1169 1.271 pgoyette enosys(), error);
1170 1.249 christos
1171 1.271 pgoyette /*
1172 1.271 pgoyette * If not present, then COMPAT_30 hook also not
1173 1.271 pgoyette * present, so just continue with checks for the
1174 1.271 pgoyette * "write" commands
1175 1.271 pgoyette */
1176 1.271 pgoyette if (error == ENOSYS) {
1177 1.271 pgoyette error = 0;
1178 1.271 pgoyette break;
1179 1.271 pgoyette }
1180 1.249 christos
1181 1.271 pgoyette /* If not already handled, try the COMPAT_30 hook */
1182 1.271 pgoyette if (error == EPASSTHROUGH)
1183 1.272 pgoyette MODULE_HOOK_CALL(compat_vndioctl_30_hook,
1184 1.271 pgoyette (cmd, l, data, unit, &vattr, vndioctl_get),
1185 1.271 pgoyette enosys(), error);
1186 1.271 pgoyette
1187 1.271 pgoyette /* If no COMPAT_30 module, or not handled, check writes */
1188 1.271 pgoyette if (error == ENOSYS || error == EPASSTHROUGH) {
1189 1.271 pgoyette error = 0;
1190 1.271 pgoyette break;
1191 1.271 pgoyette }
1192 1.271 pgoyette return error;
1193 1.249 christos }
1194 1.249 christos
1195 1.249 christos vnd = device_lookup_private(&vnd_cd, unit);
1196 1.249 christos if (vnd == NULL)
1197 1.131 cube return ENXIO;
1198 1.17 cgd vio = (struct vnd_ioctl *)data;
1199 1.43 thorpej
1200 1.43 thorpej /* Must be open for writes for these commands... */
1201 1.43 thorpej switch (cmd) {
1202 1.271 pgoyette case VNDIOCSET50:
1203 1.271 pgoyette case VNDIOCCLR50:
1204 1.271 pgoyette if (!compat_vndioctl_50_hook.hooked)
1205 1.271 pgoyette return EINVAL;
1206 1.271 pgoyette /* FALLTHROUGH */
1207 1.43 thorpej case VNDIOCSET:
1208 1.43 thorpej case VNDIOCCLR:
1209 1.43 thorpej case DIOCSDINFO:
1210 1.43 thorpej case DIOCWDINFO:
1211 1.70 fvdl #ifdef __HAVE_OLD_DISKLABEL
1212 1.70 fvdl case ODIOCSDINFO:
1213 1.70 fvdl case ODIOCWDINFO:
1214 1.70 fvdl #endif
1215 1.99 thorpej case DIOCKLABEL:
1216 1.43 thorpej case DIOCWLABEL:
1217 1.43 thorpej if ((flag & FWRITE) == 0)
1218 1.197 dyoung return EBADF;
1219 1.43 thorpej }
1220 1.43 thorpej
1221 1.43 thorpej /* Must be initialized for these... */
1222 1.1 brezak switch (cmd) {
1223 1.43 thorpej case VNDIOCCLR:
1224 1.212 mrg case VNDIOCCLR50:
1225 1.44 kleink case DIOCGDINFO:
1226 1.43 thorpej case DIOCSDINFO:
1227 1.43 thorpej case DIOCWDINFO:
1228 1.256 christos case DIOCGPARTINFO:
1229 1.99 thorpej case DIOCKLABEL:
1230 1.43 thorpej case DIOCWLABEL:
1231 1.47 thorpej case DIOCGDEFLABEL:
1232 1.194 christos case DIOCCACHESYNC:
1233 1.70 fvdl #ifdef __HAVE_OLD_DISKLABEL
1234 1.70 fvdl case ODIOCGDINFO:
1235 1.70 fvdl case ODIOCSDINFO:
1236 1.70 fvdl case ODIOCWDINFO:
1237 1.70 fvdl case ODIOCGDEFLABEL:
1238 1.70 fvdl #endif
1239 1.43 thorpej if ((vnd->sc_flags & VNF_INITED) == 0)
1240 1.197 dyoung return ENXIO;
1241 1.43 thorpej }
1242 1.1 brezak
1243 1.238 christos error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1244 1.238 christos if (error != EPASSTHROUGH)
1245 1.238 christos return error;
1246 1.238 christos
1247 1.238 christos
1248 1.43 thorpej switch (cmd) {
1249 1.212 mrg case VNDIOCSET50:
1250 1.17 cgd case VNDIOCSET:
1251 1.17 cgd if (vnd->sc_flags & VNF_INITED)
1252 1.197 dyoung return EBUSY;
1253 1.22 thorpej
1254 1.24 christos if ((error = vndlock(vnd)) != 0)
1255 1.197 dyoung return error;
1256 1.22 thorpej
1257 1.94 yamt fflags = FREAD;
1258 1.94 yamt if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1259 1.94 yamt fflags |= FWRITE;
1260 1.267 mlelstv if ((vio->vnd_flags & VNDIOF_FILEIO) != 0)
1261 1.267 mlelstv vnd->sc_flags |= VNF_USE_VN_RDWR;
1262 1.214 dholland error = pathbuf_copyin(vio->vnd_file, &pb);
1263 1.214 dholland if (error) {
1264 1.214 dholland goto unlock_and_exit;
1265 1.214 dholland }
1266 1.214 dholland NDINIT(&nd, LOOKUP, FOLLOW, pb);
1267 1.214 dholland if ((error = vn_open(&nd, fflags, 0)) != 0) {
1268 1.214 dholland pathbuf_destroy(pb);
1269 1.83 enami goto unlock_and_exit;
1270 1.214 dholland }
1271 1.143 dogcow KASSERT(l);
1272 1.171 pooka error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1273 1.94 yamt if (!error && nd.ni_vp->v_type != VREG)
1274 1.94 yamt error = EOPNOTSUPP;
1275 1.205 dsl if (!error && vattr.va_bytes < vattr.va_size)
1276 1.218 hannken /* File is definitely sparse, use vn_rdwr() */
1277 1.218 hannken vnd->sc_flags |= VNF_USE_VN_RDWR;
1278 1.115 hubertf if (error) {
1279 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1280 1.115 hubertf goto close_and_exit;
1281 1.115 hubertf }
1282 1.115 hubertf
1283 1.116 christos /* If using a compressed file, initialize its info */
1284 1.115 hubertf /* (or abort with an error if kernel has no compression) */
1285 1.267 mlelstv if (vio->vnd_flags & VNDIOF_COMP) {
1286 1.115 hubertf #ifdef VND_COMPRESSION
1287 1.116 christos struct vnd_comp_header *ch;
1288 1.116 christos int i;
1289 1.261 riastrad uint32_t comp_size;
1290 1.261 riastrad uint32_t comp_maxsize;
1291 1.244 prlw1
1292 1.116 christos /* allocate space for compresed file header */
1293 1.116 christos ch = malloc(sizeof(struct vnd_comp_header),
1294 1.261 riastrad M_TEMP, M_WAITOK);
1295 1.244 prlw1
1296 1.116 christos /* read compressed file header */
1297 1.163 christos error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1298 1.261 riastrad sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1299 1.261 riastrad IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1300 1.198 dyoung if (error) {
1301 1.116 christos free(ch, M_TEMP);
1302 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1303 1.116 christos goto close_and_exit;
1304 1.116 christos }
1305 1.244 prlw1
1306 1.261 riastrad if (be32toh(ch->block_size) == 0 ||
1307 1.261 riastrad be32toh(ch->num_blocks) > UINT32_MAX - 1) {
1308 1.260 riastrad free(ch, M_TEMP);
1309 1.260 riastrad VOP_UNLOCK(nd.ni_vp);
1310 1.260 riastrad goto close_and_exit;
1311 1.260 riastrad }
1312 1.260 riastrad
1313 1.116 christos /* save some header info */
1314 1.261 riastrad vnd->sc_comp_blksz = be32toh(ch->block_size);
1315 1.116 christos /* note last offset is the file byte size */
1316 1.261 riastrad vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1;
1317 1.116 christos free(ch, M_TEMP);
1318 1.254 christos if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) {
1319 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1320 1.116 christos error = EINVAL;
1321 1.116 christos goto close_and_exit;
1322 1.116 christos }
1323 1.260 riastrad KASSERT(0 < vnd->sc_comp_blksz);
1324 1.260 riastrad KASSERT(0 < vnd->sc_comp_numoffs);
1325 1.262 riastrad /*
1326 1.262 riastrad * @#^@!$& gcc -Wtype-limits refuses to let me
1327 1.262 riastrad * write SIZE_MAX/sizeof(uint64_t) < numoffs,
1328 1.262 riastrad * because the range of the type on amd64 makes
1329 1.262 riastrad * the comparisons always false.
1330 1.262 riastrad */
1331 1.262 riastrad #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT)
1332 1.262 riastrad if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) {
1333 1.262 riastrad VOP_UNLOCK(nd.ni_vp);
1334 1.262 riastrad error = EINVAL;
1335 1.262 riastrad goto close_and_exit;
1336 1.262 riastrad }
1337 1.262 riastrad #endif
1338 1.262 riastrad if ((vattr.va_size < sizeof(struct vnd_comp_header)) ||
1339 1.260 riastrad (vattr.va_size - sizeof(struct vnd_comp_header) <
1340 1.260 riastrad sizeof(uint64_t)*vnd->sc_comp_numoffs) ||
1341 1.260 riastrad (UQUAD_MAX/vnd->sc_comp_blksz <
1342 1.260 riastrad vnd->sc_comp_numoffs - 1)) {
1343 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1344 1.116 christos error = EINVAL;
1345 1.116 christos goto close_and_exit;
1346 1.116 christos }
1347 1.244 prlw1
1348 1.116 christos /* set decompressed file size */
1349 1.260 riastrad KASSERT(vnd->sc_comp_numoffs - 1 <=
1350 1.260 riastrad UQUAD_MAX/vnd->sc_comp_blksz);
1351 1.116 christos vattr.va_size =
1352 1.152 elad ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1353 1.152 elad (u_quad_t)vnd->sc_comp_blksz;
1354 1.244 prlw1
1355 1.116 christos /* allocate space for all the compressed offsets */
1356 1.260 riastrad __CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t));
1357 1.116 christos vnd->sc_comp_offsets =
1358 1.261 riastrad malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs,
1359 1.261 riastrad M_DEVBUF, M_WAITOK);
1360 1.244 prlw1
1361 1.116 christos /* read in the offsets */
1362 1.116 christos error = vn_rdwr(UIO_READ, nd.ni_vp,
1363 1.261 riastrad (void *)vnd->sc_comp_offsets,
1364 1.261 riastrad sizeof(uint64_t) * vnd->sc_comp_numoffs,
1365 1.261 riastrad sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1366 1.148 ad IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1367 1.198 dyoung if (error) {
1368 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1369 1.116 christos goto close_and_exit;
1370 1.116 christos }
1371 1.116 christos /*
1372 1.116 christos * find largest block size (used for allocation limit).
1373 1.116 christos * Also convert offset to native byte order.
1374 1.116 christos */
1375 1.116 christos comp_maxsize = 0;
1376 1.116 christos for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1377 1.116 christos vnd->sc_comp_offsets[i] =
1378 1.261 riastrad be64toh(vnd->sc_comp_offsets[i]);
1379 1.261 riastrad comp_size =
1380 1.261 riastrad be64toh(vnd->sc_comp_offsets[i + 1])
1381 1.261 riastrad - vnd->sc_comp_offsets[i];
1382 1.116 christos if (comp_size > comp_maxsize)
1383 1.116 christos comp_maxsize = comp_size;
1384 1.116 christos }
1385 1.116 christos vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1386 1.261 riastrad be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs
1387 1.261 riastrad - 1]);
1388 1.244 prlw1
1389 1.116 christos /* create compressed data buffer */
1390 1.116 christos vnd->sc_comp_buff = malloc(comp_maxsize,
1391 1.261 riastrad M_DEVBUF, M_WAITOK);
1392 1.244 prlw1
1393 1.116 christos /* create decompressed buffer */
1394 1.116 christos vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1395 1.261 riastrad M_DEVBUF, M_WAITOK);
1396 1.116 christos vnd->sc_comp_buffblk = -1;
1397 1.244 prlw1
1398 1.116 christos /* Initialize decompress stream */
1399 1.196 cegger memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1400 1.116 christos vnd->sc_comp_stream.zalloc = vnd_alloc;
1401 1.116 christos vnd->sc_comp_stream.zfree = vnd_free;
1402 1.116 christos error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1403 1.198 dyoung if (error) {
1404 1.198 dyoung if (vnd->sc_comp_stream.msg)
1405 1.116 christos printf("vnd%d: compressed file, %s\n",
1406 1.261 riastrad unit, vnd->sc_comp_stream.msg);
1407 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1408 1.116 christos error = EINVAL;
1409 1.116 christos goto close_and_exit;
1410 1.116 christos }
1411 1.244 prlw1
1412 1.116 christos vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1413 1.115 hubertf #else /* !VND_COMPRESSION */
1414 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1415 1.115 hubertf error = EOPNOTSUPP;
1416 1.83 enami goto close_and_exit;
1417 1.115 hubertf #endif /* VND_COMPRESSION */
1418 1.116 christos }
1419 1.244 prlw1
1420 1.209 hannken VOP_UNLOCK(nd.ni_vp);
1421 1.17 cgd vnd->sc_vp = nd.ni_vp;
1422 1.17 cgd vnd->sc_size = btodb(vattr.va_size); /* note truncation */
1423 1.41 thorpej
1424 1.265 mlelstv /* get smallest I/O size for underlying device, fall back to
1425 1.265 mlelstv * fundamental I/O size of underlying filesystem
1426 1.265 mlelstv */
1427 1.265 mlelstv error = bdev_ioctl(vattr.va_fsid, DIOCGSECTORSIZE, &vnd->sc_iosize, FKIOCTL, l);
1428 1.265 mlelstv if (error)
1429 1.265 mlelstv vnd->sc_iosize = vnd->sc_vp->v_mount->mnt_stat.f_frsize;
1430 1.265 mlelstv
1431 1.41 thorpej /*
1432 1.41 thorpej * Use pseudo-geometry specified. If none was provided,
1433 1.41 thorpej * use "standard" Adaptec fictitious geometry.
1434 1.41 thorpej */
1435 1.41 thorpej if (vio->vnd_flags & VNDIOF_HASGEOM) {
1436 1.41 thorpej
1437 1.72 thorpej memcpy(&vnd->sc_geom, &vio->vnd_geom,
1438 1.41 thorpej sizeof(vio->vnd_geom));
1439 1.41 thorpej
1440 1.41 thorpej /*
1441 1.41 thorpej * Sanity-check the sector size.
1442 1.41 thorpej */
1443 1.254 christos if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) ||
1444 1.254 christos vnd->sc_geom.vng_ntracks == 0 ||
1445 1.254 christos vnd->sc_geom.vng_nsectors == 0) {
1446 1.83 enami error = EINVAL;
1447 1.83 enami goto close_and_exit;
1448 1.41 thorpej }
1449 1.41 thorpej
1450 1.41 thorpej /*
1451 1.268 mlelstv * Compute missing cylinder count from size
1452 1.268 mlelstv */
1453 1.268 mlelstv if (vnd->sc_geom.vng_ncylinders == 0)
1454 1.268 mlelstv vnd->sc_geom.vng_ncylinders = vnd->sc_size /
1455 1.268 mlelstv (vnd->sc_geom.vng_ntracks *
1456 1.268 mlelstv vnd->sc_geom.vng_nsectors);
1457 1.268 mlelstv
1458 1.268 mlelstv /*
1459 1.41 thorpej * Compute the size (in DEV_BSIZE blocks) specified
1460 1.41 thorpej * by the geometry.
1461 1.41 thorpej */
1462 1.235 mlelstv geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1463 1.41 thorpej vnd->sc_geom.vng_ntracks *
1464 1.235 mlelstv vnd->sc_geom.vng_ncylinders *
1465 1.41 thorpej (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1466 1.41 thorpej
1467 1.41 thorpej /*
1468 1.41 thorpej * Sanity-check the size against the specified
1469 1.41 thorpej * geometry.
1470 1.41 thorpej */
1471 1.41 thorpej if (vnd->sc_size < geomsize) {
1472 1.83 enami error = EINVAL;
1473 1.83 enami goto close_and_exit;
1474 1.41 thorpej }
1475 1.118 drochner } else if (vnd->sc_size >= (32 * 64)) {
1476 1.41 thorpej /*
1477 1.41 thorpej * Size must be at least 2048 DEV_BSIZE blocks
1478 1.41 thorpej * (1M) in order to use this geometry.
1479 1.41 thorpej */
1480 1.41 thorpej vnd->sc_geom.vng_secsize = DEV_BSIZE;
1481 1.41 thorpej vnd->sc_geom.vng_nsectors = 32;
1482 1.41 thorpej vnd->sc_geom.vng_ntracks = 64;
1483 1.41 thorpej vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1484 1.118 drochner } else {
1485 1.118 drochner vnd->sc_geom.vng_secsize = DEV_BSIZE;
1486 1.118 drochner vnd->sc_geom.vng_nsectors = 1;
1487 1.118 drochner vnd->sc_geom.vng_ntracks = 1;
1488 1.118 drochner vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1489 1.41 thorpej }
1490 1.41 thorpej
1491 1.222 christos vnd_set_geometry(vnd);
1492 1.174 riz
1493 1.94 yamt if (vio->vnd_flags & VNDIOF_READONLY) {
1494 1.94 yamt vnd->sc_flags |= VNF_READONLY;
1495 1.94 yamt }
1496 1.94 yamt
1497 1.148 ad if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1498 1.83 enami goto close_and_exit;
1499 1.112 bouyer
1500 1.17 cgd vndthrottle(vnd, vnd->sc_vp);
1501 1.205 dsl vio->vnd_osize = dbtob(vnd->sc_size);
1502 1.212 mrg if (cmd != VNDIOCSET50)
1503 1.205 dsl vio->vnd_size = dbtob(vnd->sc_size);
1504 1.17 cgd vnd->sc_flags |= VNF_INITED;
1505 1.112 bouyer
1506 1.112 bouyer /* create the kernel thread, wait for it to be up */
1507 1.168 ad error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1508 1.216 joerg &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1509 1.112 bouyer if (error)
1510 1.112 bouyer goto close_and_exit;
1511 1.112 bouyer while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1512 1.112 bouyer tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1513 1.112 bouyer }
1514 1.17 cgd #ifdef DEBUG
1515 1.17 cgd if (vnddebug & VDB_INIT)
1516 1.45 fair printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1517 1.41 thorpej vnd->sc_vp, (unsigned long) vnd->sc_size,
1518 1.41 thorpej vnd->sc_geom.vng_secsize,
1519 1.41 thorpej vnd->sc_geom.vng_nsectors,
1520 1.41 thorpej vnd->sc_geom.vng_ntracks,
1521 1.41 thorpej vnd->sc_geom.vng_ncylinders);
1522 1.1 brezak #endif
1523 1.22 thorpej
1524 1.23 thorpej /* Attach the disk. */
1525 1.170 ad disk_attach(&vnd->sc_dkdev);
1526 1.23 thorpej
1527 1.59 thorpej /* Initialize the xfer and buffer pools. */
1528 1.59 thorpej pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1529 1.166 ad 0, 0, "vndxpl", NULL, IPL_BIO);
1530 1.59 thorpej
1531 1.22 thorpej vndunlock(vnd);
1532 1.22 thorpej
1533 1.214 dholland pathbuf_destroy(pb);
1534 1.221 mlelstv
1535 1.221 mlelstv /* Discover wedges on this disk */
1536 1.221 mlelstv dkwedge_discover(&vnd->sc_dkdev);
1537 1.221 mlelstv
1538 1.1 brezak break;
1539 1.83 enami
1540 1.83 enami close_and_exit:
1541 1.177 ad (void) vn_close(nd.ni_vp, fflags, l->l_cred);
1542 1.214 dholland pathbuf_destroy(pb);
1543 1.83 enami unlock_and_exit:
1544 1.115 hubertf #ifdef VND_COMPRESSION
1545 1.116 christos /* free any allocated memory (for compressed file) */
1546 1.198 dyoung if (vnd->sc_comp_offsets) {
1547 1.116 christos free(vnd->sc_comp_offsets, M_DEVBUF);
1548 1.116 christos vnd->sc_comp_offsets = NULL;
1549 1.116 christos }
1550 1.198 dyoung if (vnd->sc_comp_buff) {
1551 1.116 christos free(vnd->sc_comp_buff, M_DEVBUF);
1552 1.116 christos vnd->sc_comp_buff = NULL;
1553 1.116 christos }
1554 1.198 dyoung if (vnd->sc_comp_decombuf) {
1555 1.116 christos free(vnd->sc_comp_decombuf, M_DEVBUF);
1556 1.116 christos vnd->sc_comp_decombuf = NULL;
1557 1.116 christos }
1558 1.115 hubertf #endif /* VND_COMPRESSION */
1559 1.83 enami vndunlock(vnd);
1560 1.197 dyoung return error;
1561 1.1 brezak
1562 1.212 mrg case VNDIOCCLR50:
1563 1.17 cgd case VNDIOCCLR:
1564 1.22 thorpej part = DISKPART(dev);
1565 1.22 thorpej pmask = (1 << part);
1566 1.199 dyoung force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1567 1.22 thorpej
1568 1.199 dyoung if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1569 1.199 dyoung return error;
1570 1.23 thorpej
1571 1.1 brezak break;
1572 1.80 atatat
1573 1.1 brezak
1574 1.41 thorpej case DIOCWDINFO:
1575 1.41 thorpej case DIOCSDINFO:
1576 1.70 fvdl #ifdef __HAVE_OLD_DISKLABEL
1577 1.70 fvdl case ODIOCWDINFO:
1578 1.70 fvdl case ODIOCSDINFO:
1579 1.70 fvdl #endif
1580 1.70 fvdl {
1581 1.70 fvdl struct disklabel *lp;
1582 1.70 fvdl
1583 1.41 thorpej if ((error = vndlock(vnd)) != 0)
1584 1.197 dyoung return error;
1585 1.41 thorpej
1586 1.41 thorpej vnd->sc_flags |= VNF_LABELLING;
1587 1.41 thorpej
1588 1.70 fvdl #ifdef __HAVE_OLD_DISKLABEL
1589 1.70 fvdl if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1590 1.70 fvdl memset(&newlabel, 0, sizeof newlabel);
1591 1.70 fvdl memcpy(&newlabel, data, sizeof (struct olddisklabel));
1592 1.70 fvdl lp = &newlabel;
1593 1.70 fvdl } else
1594 1.70 fvdl #endif
1595 1.70 fvdl lp = (struct disklabel *)data;
1596 1.70 fvdl
1597 1.41 thorpej error = setdisklabel(vnd->sc_dkdev.dk_label,
1598 1.70 fvdl lp, 0, vnd->sc_dkdev.dk_cpulabel);
1599 1.41 thorpej if (error == 0) {
1600 1.70 fvdl if (cmd == DIOCWDINFO
1601 1.70 fvdl #ifdef __HAVE_OLD_DISKLABEL
1602 1.70 fvdl || cmd == ODIOCWDINFO
1603 1.70 fvdl #endif
1604 1.70 fvdl )
1605 1.41 thorpej error = writedisklabel(VNDLABELDEV(dev),
1606 1.41 thorpej vndstrategy, vnd->sc_dkdev.dk_label,
1607 1.41 thorpej vnd->sc_dkdev.dk_cpulabel);
1608 1.41 thorpej }
1609 1.41 thorpej
1610 1.41 thorpej vnd->sc_flags &= ~VNF_LABELLING;
1611 1.41 thorpej
1612 1.41 thorpej vndunlock(vnd);
1613 1.41 thorpej
1614 1.41 thorpej if (error)
1615 1.197 dyoung return error;
1616 1.41 thorpej break;
1617 1.70 fvdl }
1618 1.41 thorpej
1619 1.99 thorpej case DIOCKLABEL:
1620 1.99 thorpej if (*(int *)data != 0)
1621 1.99 thorpej vnd->sc_flags |= VNF_KLABEL;
1622 1.99 thorpej else
1623 1.99 thorpej vnd->sc_flags &= ~VNF_KLABEL;
1624 1.99 thorpej break;
1625 1.99 thorpej
1626 1.41 thorpej case DIOCWLABEL:
1627 1.41 thorpej if (*(int *)data != 0)
1628 1.41 thorpej vnd->sc_flags |= VNF_WLABEL;
1629 1.41 thorpej else
1630 1.41 thorpej vnd->sc_flags &= ~VNF_WLABEL;
1631 1.41 thorpej break;
1632 1.22 thorpej
1633 1.47 thorpej case DIOCGDEFLABEL:
1634 1.47 thorpej vndgetdefaultlabel(vnd, (struct disklabel *)data);
1635 1.47 thorpej break;
1636 1.70 fvdl
1637 1.70 fvdl #ifdef __HAVE_OLD_DISKLABEL
1638 1.70 fvdl case ODIOCGDEFLABEL:
1639 1.70 fvdl vndgetdefaultlabel(vnd, &newlabel);
1640 1.70 fvdl if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1641 1.71 fvdl return ENOTTY;
1642 1.70 fvdl memcpy(data, &newlabel, sizeof (struct olddisklabel));
1643 1.70 fvdl break;
1644 1.70 fvdl #endif
1645 1.195 apb
1646 1.194 christos case DIOCCACHESYNC:
1647 1.194 christos vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1648 1.194 christos error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1649 1.195 apb FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1650 1.209 hannken VOP_UNLOCK(vnd->sc_vp);
1651 1.194 christos return error;
1652 1.47 thorpej
1653 1.1 brezak default:
1654 1.197 dyoung return ENOTTY;
1655 1.1 brezak }
1656 1.22 thorpej
1657 1.197 dyoung return 0;
1658 1.1 brezak }
1659 1.1 brezak
1660 1.1 brezak /*
1661 1.1 brezak * Duplicate the current processes' credentials. Since we are called only
1662 1.1 brezak * as the result of a SET ioctl and only root can do that, any future access
1663 1.1 brezak * to this "disk" is essentially as root. Note that credentials may change
1664 1.1 brezak * if some other uid can write directly to the mapped file (NFS).
1665 1.1 brezak */
1666 1.108 thorpej static int
1667 1.147 elad vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1668 1.1 brezak {
1669 1.1 brezak struct uio auio;
1670 1.1 brezak struct iovec aiov;
1671 1.5 cgd char *tmpbuf;
1672 1.5 cgd int error;
1673 1.1 brezak
1674 1.147 elad vnd->sc_cred = kauth_cred_dup(cred);
1675 1.5 cgd tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1676 1.5 cgd
1677 1.1 brezak /* XXX: Horrible kludge to establish credentials for NFS */
1678 1.1 brezak aiov.iov_base = tmpbuf;
1679 1.264 riastrad aiov.iov_len = uimin(DEV_BSIZE, dbtob(vnd->sc_size));
1680 1.1 brezak auio.uio_iov = &aiov;
1681 1.1 brezak auio.uio_iovcnt = 1;
1682 1.1 brezak auio.uio_offset = 0;
1683 1.1 brezak auio.uio_rw = UIO_READ;
1684 1.1 brezak auio.uio_resid = aiov.iov_len;
1685 1.141 yamt UIO_SETUP_SYSSPACE(&auio);
1686 1.56 fvdl vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1687 1.17 cgd error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1688 1.54 bad if (error == 0) {
1689 1.54 bad /*
1690 1.54 bad * Because vnd does all IO directly through the vnode
1691 1.56 fvdl * we need to flush (at least) the buffer from the above
1692 1.54 bad * VOP_READ from the buffer cache to prevent cache
1693 1.54 bad * incoherencies. Also, be careful to write dirty
1694 1.54 bad * buffers back to stable storage.
1695 1.54 bad */
1696 1.54 bad error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1697 1.124 christos curlwp, 0, 0);
1698 1.54 bad }
1699 1.209 hannken VOP_UNLOCK(vnd->sc_vp);
1700 1.5 cgd
1701 1.5 cgd free(tmpbuf, M_TEMP);
1702 1.197 dyoung return error;
1703 1.1 brezak }
1704 1.1 brezak
1705 1.1 brezak /*
1706 1.1 brezak * Set maxactive based on FS type
1707 1.1 brezak */
1708 1.108 thorpej static void
1709 1.108 thorpej vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1710 1.1 brezak {
1711 1.5 cgd
1712 1.208 pooka if (vp->v_tag == VT_NFS)
1713 1.17 cgd vnd->sc_maxactive = 2;
1714 1.1 brezak else
1715 1.17 cgd vnd->sc_maxactive = 8;
1716 1.1 brezak
1717 1.17 cgd if (vnd->sc_maxactive < 1)
1718 1.17 cgd vnd->sc_maxactive = 1;
1719 1.1 brezak }
1720 1.1 brezak
1721 1.95 drochner #if 0
1722 1.108 thorpej static void
1723 1.108 thorpej vndshutdown(void)
1724 1.1 brezak {
1725 1.65 augustss struct vnd_softc *vnd;
1726 1.1 brezak
1727 1.17 cgd for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1728 1.17 cgd if (vnd->sc_flags & VNF_INITED)
1729 1.17 cgd vndclear(vnd);
1730 1.1 brezak }
1731 1.95 drochner #endif
1732 1.1 brezak
1733 1.108 thorpej static void
1734 1.108 thorpej vndclear(struct vnd_softc *vnd, int myminor)
1735 1.1 brezak {
1736 1.65 augustss struct vnode *vp = vnd->sc_vp;
1737 1.94 yamt int fflags = FREAD;
1738 1.95 drochner int bmaj, cmaj, i, mn;
1739 1.113 yamt int s;
1740 1.1 brezak
1741 1.1 brezak #ifdef DEBUG
1742 1.17 cgd if (vnddebug & VDB_FOLLOW)
1743 1.30 christos printf("vndclear(%p): vp %p\n", vnd, vp);
1744 1.1 brezak #endif
1745 1.95 drochner /* locate the major number */
1746 1.95 drochner bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1747 1.95 drochner cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1748 1.95 drochner
1749 1.95 drochner /* Nuke the vnodes for any open instances */
1750 1.95 drochner for (i = 0; i < MAXPARTITIONS; i++) {
1751 1.176 cube mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1752 1.95 drochner vdevgone(bmaj, mn, mn, VBLK);
1753 1.95 drochner if (mn != myminor) /* XXX avoid to kill own vnode */
1754 1.95 drochner vdevgone(cmaj, mn, mn, VCHR);
1755 1.95 drochner }
1756 1.95 drochner
1757 1.94 yamt if ((vnd->sc_flags & VNF_READONLY) == 0)
1758 1.94 yamt fflags |= FWRITE;
1759 1.112 bouyer
1760 1.113 yamt s = splbio();
1761 1.123 yamt bufq_drain(vnd->sc_tab);
1762 1.113 yamt splx(s);
1763 1.113 yamt
1764 1.112 bouyer vnd->sc_flags |= VNF_VUNCONF;
1765 1.112 bouyer wakeup(&vnd->sc_tab);
1766 1.112 bouyer while (vnd->sc_flags & VNF_KTHREAD)
1767 1.112 bouyer tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1768 1.112 bouyer
1769 1.115 hubertf #ifdef VND_COMPRESSION
1770 1.116 christos /* free the compressed file buffers */
1771 1.198 dyoung if (vnd->sc_flags & VNF_COMP) {
1772 1.198 dyoung if (vnd->sc_comp_offsets) {
1773 1.116 christos free(vnd->sc_comp_offsets, M_DEVBUF);
1774 1.116 christos vnd->sc_comp_offsets = NULL;
1775 1.116 christos }
1776 1.198 dyoung if (vnd->sc_comp_buff) {
1777 1.116 christos free(vnd->sc_comp_buff, M_DEVBUF);
1778 1.116 christos vnd->sc_comp_buff = NULL;
1779 1.116 christos }
1780 1.198 dyoung if (vnd->sc_comp_decombuf) {
1781 1.116 christos free(vnd->sc_comp_decombuf, M_DEVBUF);
1782 1.116 christos vnd->sc_comp_decombuf = NULL;
1783 1.116 christos }
1784 1.116 christos }
1785 1.115 hubertf #endif /* VND_COMPRESSION */
1786 1.112 bouyer vnd->sc_flags &=
1787 1.242 mlelstv ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1788 1.203 dyoung | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1789 1.197 dyoung if (vp == NULL)
1790 1.112 bouyer panic("vndclear: null vp");
1791 1.177 ad (void) vn_close(vp, fflags, vnd->sc_cred);
1792 1.147 elad kauth_cred_free(vnd->sc_cred);
1793 1.197 dyoung vnd->sc_vp = NULL;
1794 1.197 dyoung vnd->sc_cred = NULL;
1795 1.17 cgd vnd->sc_size = 0;
1796 1.1 brezak }
1797 1.1 brezak
1798 1.108 thorpej static int
1799 1.108 thorpej vndsize(dev_t dev)
1800 1.1 brezak {
1801 1.41 thorpej struct vnd_softc *sc;
1802 1.41 thorpej struct disklabel *lp;
1803 1.41 thorpej int part, unit, omask;
1804 1.41 thorpej int size;
1805 1.41 thorpej
1806 1.41 thorpej unit = vndunit(dev);
1807 1.182 cegger sc = device_lookup_private(&vnd_cd, unit);
1808 1.131 cube if (sc == NULL)
1809 1.131 cube return -1;
1810 1.41 thorpej
1811 1.41 thorpej if ((sc->sc_flags & VNF_INITED) == 0)
1812 1.197 dyoung return -1;
1813 1.41 thorpej
1814 1.41 thorpej part = DISKPART(dev);
1815 1.41 thorpej omask = sc->sc_dkdev.dk_openmask & (1 << part);
1816 1.41 thorpej lp = sc->sc_dkdev.dk_label;
1817 1.41 thorpej
1818 1.124 christos if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp)) /* XXX */
1819 1.197 dyoung return -1;
1820 1.41 thorpej
1821 1.41 thorpej if (lp->d_partitions[part].p_fstype != FS_SWAP)
1822 1.41 thorpej size = -1;
1823 1.41 thorpej else
1824 1.41 thorpej size = lp->d_partitions[part].p_size *
1825 1.41 thorpej (lp->d_secsize / DEV_BSIZE);
1826 1.41 thorpej
1827 1.124 christos if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp)) /* XXX */
1828 1.197 dyoung return -1;
1829 1.1 brezak
1830 1.197 dyoung return size;
1831 1.1 brezak }
1832 1.1 brezak
1833 1.108 thorpej static int
1834 1.163 christos vnddump(dev_t dev, daddr_t blkno, void *va,
1835 1.160 christos size_t size)
1836 1.1 brezak {
1837 1.16 cgd
1838 1.19 cgd /* Not implemented. */
1839 1.19 cgd return ENXIO;
1840 1.41 thorpej }
1841 1.41 thorpej
1842 1.108 thorpej static void
1843 1.108 thorpej vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1844 1.41 thorpej {
1845 1.41 thorpej struct vndgeom *vng = &sc->sc_geom;
1846 1.41 thorpej struct partition *pp;
1847 1.235 mlelstv unsigned spb;
1848 1.41 thorpej
1849 1.72 thorpej memset(lp, 0, sizeof(*lp));
1850 1.41 thorpej
1851 1.235 mlelstv spb = vng->vng_secsize / DEV_BSIZE;
1852 1.235 mlelstv if (sc->sc_size / spb > UINT32_MAX)
1853 1.233 mlelstv lp->d_secperunit = UINT32_MAX;
1854 1.233 mlelstv else
1855 1.235 mlelstv lp->d_secperunit = sc->sc_size / spb;
1856 1.41 thorpej lp->d_secsize = vng->vng_secsize;
1857 1.41 thorpej lp->d_nsectors = vng->vng_nsectors;
1858 1.41 thorpej lp->d_ntracks = vng->vng_ntracks;
1859 1.41 thorpej lp->d_ncylinders = vng->vng_ncylinders;
1860 1.41 thorpej lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1861 1.41 thorpej
1862 1.41 thorpej strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1863 1.239 christos lp->d_type = DKTYPE_VND;
1864 1.41 thorpej strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1865 1.41 thorpej lp->d_rpm = 3600;
1866 1.41 thorpej lp->d_interleave = 1;
1867 1.41 thorpej lp->d_flags = 0;
1868 1.41 thorpej
1869 1.41 thorpej pp = &lp->d_partitions[RAW_PART];
1870 1.41 thorpej pp->p_offset = 0;
1871 1.41 thorpej pp->p_size = lp->d_secperunit;
1872 1.41 thorpej pp->p_fstype = FS_UNUSED;
1873 1.41 thorpej lp->d_npartitions = RAW_PART + 1;
1874 1.41 thorpej
1875 1.41 thorpej lp->d_magic = DISKMAGIC;
1876 1.41 thorpej lp->d_magic2 = DISKMAGIC;
1877 1.41 thorpej lp->d_checksum = dkcksum(lp);
1878 1.47 thorpej }
1879 1.47 thorpej
1880 1.47 thorpej /*
1881 1.47 thorpej * Read the disklabel from a vnd. If one is not present, create a fake one.
1882 1.47 thorpej */
1883 1.108 thorpej static void
1884 1.131 cube vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1885 1.47 thorpej {
1886 1.97 dsl const char *errstring;
1887 1.47 thorpej struct disklabel *lp = sc->sc_dkdev.dk_label;
1888 1.47 thorpej struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1889 1.47 thorpej int i;
1890 1.47 thorpej
1891 1.72 thorpej memset(clp, 0, sizeof(*clp));
1892 1.47 thorpej
1893 1.47 thorpej vndgetdefaultlabel(sc, lp);
1894 1.41 thorpej
1895 1.41 thorpej /*
1896 1.41 thorpej * Call the generic disklabel extraction routine.
1897 1.41 thorpej */
1898 1.41 thorpej errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1899 1.41 thorpej if (errstring) {
1900 1.41 thorpej /*
1901 1.41 thorpej * Lack of disklabel is common, but we print the warning
1902 1.41 thorpej * anyway, since it might contain other useful information.
1903 1.41 thorpej */
1904 1.176 cube aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1905 1.41 thorpej
1906 1.41 thorpej /*
1907 1.41 thorpej * For historical reasons, if there's no disklabel
1908 1.41 thorpej * present, all partitions must be FS_BSDFFS and
1909 1.41 thorpej * occupy the entire disk.
1910 1.41 thorpej */
1911 1.41 thorpej for (i = 0; i < MAXPARTITIONS; i++) {
1912 1.52 enami /*
1913 1.52 enami * Don't wipe out port specific hack (such as
1914 1.52 enami * dos partition hack of i386 port).
1915 1.52 enami */
1916 1.96 dsl if (lp->d_partitions[i].p_size != 0)
1917 1.52 enami continue;
1918 1.52 enami
1919 1.41 thorpej lp->d_partitions[i].p_size = lp->d_secperunit;
1920 1.41 thorpej lp->d_partitions[i].p_offset = 0;
1921 1.41 thorpej lp->d_partitions[i].p_fstype = FS_BSDFFS;
1922 1.41 thorpej }
1923 1.41 thorpej
1924 1.41 thorpej strncpy(lp->d_packname, "default label",
1925 1.41 thorpej sizeof(lp->d_packname));
1926 1.47 thorpej
1927 1.96 dsl lp->d_npartitions = MAXPARTITIONS;
1928 1.47 thorpej lp->d_checksum = dkcksum(lp);
1929 1.41 thorpej }
1930 1.1 brezak }
1931 1.22 thorpej
1932 1.22 thorpej /*
1933 1.22 thorpej * Wait interruptibly for an exclusive lock.
1934 1.22 thorpej *
1935 1.22 thorpej * XXX
1936 1.22 thorpej * Several drivers do this; it should be abstracted and made MP-safe.
1937 1.22 thorpej */
1938 1.22 thorpej static int
1939 1.108 thorpej vndlock(struct vnd_softc *sc)
1940 1.22 thorpej {
1941 1.22 thorpej int error;
1942 1.22 thorpej
1943 1.22 thorpej while ((sc->sc_flags & VNF_LOCKED) != 0) {
1944 1.22 thorpej sc->sc_flags |= VNF_WANTED;
1945 1.22 thorpej if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1946 1.197 dyoung return error;
1947 1.22 thorpej }
1948 1.22 thorpej sc->sc_flags |= VNF_LOCKED;
1949 1.197 dyoung return 0;
1950 1.22 thorpej }
1951 1.22 thorpej
1952 1.22 thorpej /*
1953 1.22 thorpej * Unlock and wake up any waiters.
1954 1.22 thorpej */
1955 1.22 thorpej static void
1956 1.108 thorpej vndunlock(struct vnd_softc *sc)
1957 1.22 thorpej {
1958 1.22 thorpej
1959 1.22 thorpej sc->sc_flags &= ~VNF_LOCKED;
1960 1.22 thorpej if ((sc->sc_flags & VNF_WANTED) != 0) {
1961 1.22 thorpej sc->sc_flags &= ~VNF_WANTED;
1962 1.22 thorpej wakeup(sc);
1963 1.22 thorpej }
1964 1.22 thorpej }
1965 1.115 hubertf
1966 1.115 hubertf #ifdef VND_COMPRESSION
1967 1.115 hubertf /* compressed file read */
1968 1.115 hubertf static void
1969 1.116 christos compstrategy(struct buf *bp, off_t bn)
1970 1.116 christos {
1971 1.116 christos int error;
1972 1.116 christos int unit = vndunit(bp->b_dev);
1973 1.131 cube struct vnd_softc *vnd =
1974 1.182 cegger device_lookup_private(&vnd_cd, unit);
1975 1.116 christos u_int32_t comp_block;
1976 1.116 christos struct uio auio;
1977 1.163 christos char *addr;
1978 1.116 christos int s;
1979 1.116 christos
1980 1.116 christos /* set up constants for data move */
1981 1.116 christos auio.uio_rw = UIO_READ;
1982 1.141 yamt UIO_SETUP_SYSSPACE(&auio);
1983 1.116 christos
1984 1.116 christos /* read, and transfer the data */
1985 1.116 christos addr = bp->b_data;
1986 1.165 yamt bp->b_resid = bp->b_bcount;
1987 1.116 christos s = splbio();
1988 1.116 christos while (bp->b_resid > 0) {
1989 1.116 christos unsigned length;
1990 1.116 christos size_t length_in_buffer;
1991 1.116 christos u_int32_t offset_in_buffer;
1992 1.116 christos struct iovec aiov;
1993 1.116 christos
1994 1.116 christos /* calculate the compressed block number */
1995 1.116 christos comp_block = bn / (off_t)vnd->sc_comp_blksz;
1996 1.116 christos
1997 1.116 christos /* check for good block number */
1998 1.116 christos if (comp_block >= vnd->sc_comp_numoffs) {
1999 1.116 christos bp->b_error = EINVAL;
2000 1.116 christos splx(s);
2001 1.116 christos return;
2002 1.116 christos }
2003 1.116 christos
2004 1.116 christos /* read in the compressed block, if not in buffer */
2005 1.116 christos if (comp_block != vnd->sc_comp_buffblk) {
2006 1.116 christos length = vnd->sc_comp_offsets[comp_block + 1] -
2007 1.116 christos vnd->sc_comp_offsets[comp_block];
2008 1.116 christos vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
2009 1.116 christos error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
2010 1.116 christos length, vnd->sc_comp_offsets[comp_block],
2011 1.187 ad UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
2012 1.187 ad NULL, NULL);
2013 1.116 christos if (error) {
2014 1.116 christos bp->b_error = error;
2015 1.209 hannken VOP_UNLOCK(vnd->sc_vp);
2016 1.116 christos splx(s);
2017 1.116 christos return;
2018 1.116 christos }
2019 1.116 christos /* uncompress the buffer */
2020 1.116 christos vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
2021 1.116 christos vnd->sc_comp_stream.avail_in = length;
2022 1.116 christos vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
2023 1.116 christos vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
2024 1.116 christos inflateReset(&vnd->sc_comp_stream);
2025 1.116 christos error = inflate(&vnd->sc_comp_stream, Z_FINISH);
2026 1.116 christos if (error != Z_STREAM_END) {
2027 1.116 christos if (vnd->sc_comp_stream.msg)
2028 1.176 cube aprint_normal_dev(vnd->sc_dev,
2029 1.176 cube "compressed file, %s\n",
2030 1.116 christos vnd->sc_comp_stream.msg);
2031 1.116 christos bp->b_error = EBADMSG;
2032 1.209 hannken VOP_UNLOCK(vnd->sc_vp);
2033 1.116 christos splx(s);
2034 1.116 christos return;
2035 1.116 christos }
2036 1.116 christos vnd->sc_comp_buffblk = comp_block;
2037 1.209 hannken VOP_UNLOCK(vnd->sc_vp);
2038 1.116 christos }
2039 1.116 christos
2040 1.116 christos /* transfer the usable uncompressed data */
2041 1.116 christos offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
2042 1.116 christos length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
2043 1.116 christos if (length_in_buffer > bp->b_resid)
2044 1.116 christos length_in_buffer = bp->b_resid;
2045 1.116 christos auio.uio_iov = &aiov;
2046 1.116 christos auio.uio_iovcnt = 1;
2047 1.116 christos aiov.iov_base = addr;
2048 1.116 christos aiov.iov_len = length_in_buffer;
2049 1.116 christos auio.uio_resid = aiov.iov_len;
2050 1.116 christos auio.uio_offset = 0;
2051 1.116 christos error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
2052 1.116 christos length_in_buffer, &auio);
2053 1.116 christos if (error) {
2054 1.116 christos bp->b_error = error;
2055 1.116 christos splx(s);
2056 1.116 christos return;
2057 1.116 christos }
2058 1.116 christos
2059 1.116 christos bn += length_in_buffer;
2060 1.116 christos addr += length_in_buffer;
2061 1.116 christos bp->b_resid -= length_in_buffer;
2062 1.116 christos }
2063 1.116 christos splx(s);
2064 1.115 hubertf }
2065 1.115 hubertf
2066 1.115 hubertf /* compression memory allocation routines */
2067 1.115 hubertf static void *
2068 1.160 christos vnd_alloc(void *aux, u_int items, u_int siz)
2069 1.116 christos {
2070 1.117 christos return malloc(items * siz, M_TEMP, M_NOWAIT);
2071 1.115 hubertf }
2072 1.115 hubertf
2073 1.115 hubertf static void
2074 1.160 christos vnd_free(void *aux, void *ptr)
2075 1.115 hubertf {
2076 1.116 christos free(ptr, M_TEMP);
2077 1.115 hubertf }
2078 1.115 hubertf #endif /* VND_COMPRESSION */
2079 1.174 riz
2080 1.174 riz static void
2081 1.222 christos vnd_set_geometry(struct vnd_softc *vnd)
2082 1.174 riz {
2083 1.222 christos struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2084 1.174 riz
2085 1.222 christos memset(dg, 0, sizeof(*dg));
2086 1.174 riz
2087 1.224 christos dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2088 1.222 christos vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2089 1.222 christos dg->dg_secsize = vnd->sc_geom.vng_secsize;
2090 1.222 christos dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2091 1.222 christos dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2092 1.222 christos dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2093 1.174 riz
2094 1.229 prlw1 #ifdef DEBUG
2095 1.229 prlw1 if (vnddebug & VDB_LABEL) {
2096 1.229 prlw1 printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2097 1.229 prlw1 printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2098 1.229 prlw1 }
2099 1.229 prlw1 #endif
2100 1.222 christos disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2101 1.174 riz }
2102 1.193 haad
2103 1.230 pooka #ifdef VND_COMPRESSION
2104 1.230 pooka #define VND_DEPENDS "zlib"
2105 1.230 pooka #else
2106 1.230 pooka #define VND_DEPENDS NULL
2107 1.230 pooka #endif
2108 1.230 pooka
2109 1.230 pooka MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2110 1.257 pgoyette
2111 1.257 pgoyette #ifdef _MODULE
2112 1.257 pgoyette int vnd_bmajor = -1, vnd_cmajor = -1;
2113 1.257 pgoyette
2114 1.193 haad CFDRIVER_DECL(vnd, DV_DISK, NULL);
2115 1.257 pgoyette #endif
2116 1.193 haad
2117 1.193 haad static int
2118 1.193 haad vnd_modcmd(modcmd_t cmd, void *arg)
2119 1.193 haad {
2120 1.257 pgoyette int error = 0;
2121 1.244 prlw1
2122 1.193 haad switch (cmd) {
2123 1.193 haad case MODULE_CMD_INIT:
2124 1.257 pgoyette #ifdef _MODULE
2125 1.193 haad error = config_cfdriver_attach(&vnd_cd);
2126 1.193 haad if (error)
2127 1.193 haad break;
2128 1.193 haad
2129 1.193 haad error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2130 1.193 haad if (error) {
2131 1.193 haad config_cfdriver_detach(&vnd_cd);
2132 1.259 pgoyette #ifdef DIAGNOSTIC
2133 1.257 pgoyette aprint_error("%s: unable to register cfattach for \n"
2134 1.257 pgoyette "%s, error %d", __func__, vnd_cd.cd_name, error);
2135 1.259 pgoyette #endif
2136 1.193 haad break;
2137 1.193 haad }
2138 1.244 prlw1
2139 1.257 pgoyette /*
2140 1.257 pgoyette * Attach the {b,c}devsw's
2141 1.257 pgoyette */
2142 1.257 pgoyette error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2143 1.257 pgoyette &vnd_cdevsw, &vnd_cmajor);
2144 1.257 pgoyette /*
2145 1.257 pgoyette * If devsw_attach fails, remove from autoconf database
2146 1.257 pgoyette */
2147 1.193 haad if (error) {
2148 1.193 haad config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2149 1.193 haad config_cfdriver_detach(&vnd_cd);
2150 1.259 pgoyette #ifdef DIAGNOSTIC
2151 1.257 pgoyette aprint_error("%s: unable to attach %s devsw, "
2152 1.257 pgoyette "error %d", __func__, vnd_cd.cd_name, error);
2153 1.259 pgoyette #endif
2154 1.193 haad break;
2155 1.193 haad }
2156 1.257 pgoyette #endif
2157 1.193 haad break;
2158 1.193 haad
2159 1.193 haad case MODULE_CMD_FINI:
2160 1.257 pgoyette #ifdef _MODULE
2161 1.257 pgoyette /*
2162 1.257 pgoyette * Remove {b,c}devsw's
2163 1.257 pgoyette */
2164 1.257 pgoyette devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2165 1.257 pgoyette
2166 1.257 pgoyette /*
2167 1.257 pgoyette * Now remove device from autoconf database
2168 1.257 pgoyette */
2169 1.193 haad error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2170 1.257 pgoyette if (error) {
2171 1.258 pgoyette (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2172 1.257 pgoyette &vnd_cdevsw, &vnd_cmajor);
2173 1.259 pgoyette #ifdef DIAGNOSTIC
2174 1.257 pgoyette aprint_error("%s: failed to detach %s cfattach, "
2175 1.257 pgoyette "error %d\n", __func__, vnd_cd.cd_name, error);
2176 1.259 pgoyette #endif
2177 1.257 pgoyette break;
2178 1.257 pgoyette }
2179 1.257 pgoyette error = config_cfdriver_detach(&vnd_cd);
2180 1.257 pgoyette if (error) {
2181 1.258 pgoyette (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2182 1.258 pgoyette (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2183 1.257 pgoyette &vnd_cdevsw, &vnd_cmajor);
2184 1.259 pgoyette #ifdef DIAGNOSTIC
2185 1.257 pgoyette aprint_error("%s: failed to detach %s cfdriver, "
2186 1.257 pgoyette "error %d\n", __func__, vnd_cd.cd_name, error);
2187 1.257 pgoyette break;
2188 1.259 pgoyette #endif
2189 1.257 pgoyette }
2190 1.257 pgoyette #endif
2191 1.193 haad break;
2192 1.193 haad
2193 1.193 haad case MODULE_CMD_STAT:
2194 1.193 haad return ENOTTY;
2195 1.193 haad
2196 1.193 haad default:
2197 1.193 haad return ENOTTY;
2198 1.193 haad }
2199 1.193 haad
2200 1.193 haad return error;
2201 1.193 haad }
2202