ld.c revision 1.80 1 1.80 christos /* $NetBSD: ld.c,v 1.80 2014/12/31 19:52:05 christos Exp $ */
2 1.1 ad
3 1.1 ad /*-
4 1.1 ad * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc.
5 1.1 ad * All rights reserved.
6 1.1 ad *
7 1.1 ad * This code is derived from software contributed to The NetBSD Foundation
8 1.1 ad * by Andrew Doran and Charles M. Hannum.
9 1.1 ad *
10 1.1 ad * Redistribution and use in source and binary forms, with or without
11 1.1 ad * modification, are permitted provided that the following conditions
12 1.1 ad * are met:
13 1.1 ad * 1. Redistributions of source code must retain the above copyright
14 1.1 ad * notice, this list of conditions and the following disclaimer.
15 1.1 ad * 2. Redistributions in binary form must reproduce the above copyright
16 1.1 ad * notice, this list of conditions and the following disclaimer in the
17 1.1 ad * documentation and/or other materials provided with the distribution.
18 1.1 ad *
19 1.1 ad * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 1.1 ad * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 1.1 ad * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 1.1 ad * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 1.1 ad * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 1.1 ad * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 1.1 ad * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 1.1 ad * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 1.1 ad * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 1.1 ad * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 1.1 ad * POSSIBILITY OF SUCH DAMAGE.
30 1.1 ad */
31 1.1 ad
32 1.1 ad /*
33 1.1 ad * Disk driver for use by RAID controllers.
34 1.1 ad */
35 1.12 lukem
36 1.12 lukem #include <sys/cdefs.h>
37 1.80 christos __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.80 2014/12/31 19:52:05 christos Exp $");
38 1.1 ad
39 1.1 ad #include <sys/param.h>
40 1.1 ad #include <sys/systm.h>
41 1.1 ad #include <sys/kernel.h>
42 1.1 ad #include <sys/device.h>
43 1.1 ad #include <sys/queue.h>
44 1.1 ad #include <sys/proc.h>
45 1.1 ad #include <sys/buf.h>
46 1.33 yamt #include <sys/bufq.h>
47 1.1 ad #include <sys/endian.h>
48 1.1 ad #include <sys/disklabel.h>
49 1.1 ad #include <sys/disk.h>
50 1.1 ad #include <sys/dkio.h>
51 1.1 ad #include <sys/stat.h>
52 1.1 ad #include <sys/conf.h>
53 1.1 ad #include <sys/fcntl.h>
54 1.2 ad #include <sys/vnode.h>
55 1.1 ad #include <sys/syslog.h>
56 1.44 ad #include <sys/mutex.h>
57 1.1 ad #include <sys/rnd.h>
58 1.1 ad
59 1.1 ad #include <dev/ldvar.h>
60 1.1 ad
61 1.43 riz #include <prop/proplib.h>
62 1.43 riz
63 1.1 ad static void ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
64 1.1 ad static void ldgetdisklabel(struct ld_softc *);
65 1.1 ad static void ldminphys(struct buf *bp);
66 1.67 jmcneill static bool ld_suspend(device_t, const pmf_qual_t *);
67 1.55 jmcneill static bool ld_shutdown(device_t, int);
68 1.44 ad static void ldstart(struct ld_softc *, struct buf *);
69 1.71 christos static void ld_set_geometry(struct ld_softc *);
70 1.65 cegger static void ld_config_interrupts (device_t);
71 1.66 dyoung static int ldlastclose(device_t);
72 1.1 ad
73 1.1 ad extern struct cfdriver ld_cd;
74 1.1 ad
75 1.29 thorpej static dev_type_open(ldopen);
76 1.29 thorpej static dev_type_close(ldclose);
77 1.29 thorpej static dev_type_read(ldread);
78 1.29 thorpej static dev_type_write(ldwrite);
79 1.29 thorpej static dev_type_ioctl(ldioctl);
80 1.29 thorpej static dev_type_strategy(ldstrategy);
81 1.29 thorpej static dev_type_dump(lddump);
82 1.29 thorpej static dev_type_size(ldsize);
83 1.16 gehenna
84 1.16 gehenna const struct bdevsw ld_bdevsw = {
85 1.72 dholland .d_open = ldopen,
86 1.72 dholland .d_close = ldclose,
87 1.72 dholland .d_strategy = ldstrategy,
88 1.72 dholland .d_ioctl = ldioctl,
89 1.72 dholland .d_dump = lddump,
90 1.72 dholland .d_psize = ldsize,
91 1.73 dholland .d_discard = nodiscard,
92 1.72 dholland .d_flag = D_DISK
93 1.16 gehenna };
94 1.16 gehenna
95 1.16 gehenna const struct cdevsw ld_cdevsw = {
96 1.72 dholland .d_open = ldopen,
97 1.72 dholland .d_close = ldclose,
98 1.72 dholland .d_read = ldread,
99 1.72 dholland .d_write = ldwrite,
100 1.72 dholland .d_ioctl = ldioctl,
101 1.72 dholland .d_stop = nostop,
102 1.72 dholland .d_tty = notty,
103 1.72 dholland .d_poll = nopoll,
104 1.72 dholland .d_mmap = nommap,
105 1.72 dholland .d_kqfilter = nokqfilter,
106 1.74 dholland .d_discard = nodiscard,
107 1.72 dholland .d_flag = D_DISK
108 1.16 gehenna };
109 1.16 gehenna
110 1.30 thorpej static struct dkdriver lddkdriver = { ldstrategy, ldminphys };
111 1.1 ad
112 1.1 ad void
113 1.1 ad ldattach(struct ld_softc *sc)
114 1.1 ad {
115 1.37 christos char tbuf[9];
116 1.1 ad
117 1.53 ad mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
118 1.44 ad
119 1.7 ad if ((sc->sc_flags & LDF_ENABLED) == 0) {
120 1.63 tron aprint_normal_dev(sc->sc_dv, "disabled\n");
121 1.7 ad return;
122 1.7 ad }
123 1.7 ad
124 1.1 ad /* Initialise and attach the disk structure. */
125 1.63 tron disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
126 1.1 ad disk_attach(&sc->sc_dk);
127 1.1 ad
128 1.1 ad if (sc->sc_maxxfer > MAXPHYS)
129 1.1 ad sc->sc_maxxfer = MAXPHYS;
130 1.9 ad
131 1.19 thorpej /* Build synthetic geometry if necessary. */
132 1.19 thorpej if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
133 1.19 thorpej sc->sc_ncylinders == 0) {
134 1.28 dbj uint64_t ncyl;
135 1.28 dbj
136 1.19 thorpej if (sc->sc_secperunit <= 528 * 2048) /* 528MB */
137 1.19 thorpej sc->sc_nheads = 16;
138 1.19 thorpej else if (sc->sc_secperunit <= 1024 * 2048) /* 1GB */
139 1.19 thorpej sc->sc_nheads = 32;
140 1.19 thorpej else if (sc->sc_secperunit <= 21504 * 2048) /* 21GB */
141 1.19 thorpej sc->sc_nheads = 64;
142 1.19 thorpej else if (sc->sc_secperunit <= 43008 * 2048) /* 42GB */
143 1.19 thorpej sc->sc_nheads = 128;
144 1.19 thorpej else
145 1.19 thorpej sc->sc_nheads = 255;
146 1.19 thorpej
147 1.19 thorpej sc->sc_nsectors = 63;
148 1.28 dbj sc->sc_ncylinders = INT_MAX;
149 1.35 perry ncyl = sc->sc_secperunit /
150 1.19 thorpej (sc->sc_nheads * sc->sc_nsectors);
151 1.28 dbj if (ncyl < INT_MAX)
152 1.28 dbj sc->sc_ncylinders = (int)ncyl;
153 1.19 thorpej }
154 1.1 ad
155 1.37 christos format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
156 1.1 ad sc->sc_secsize);
157 1.63 tron aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
158 1.63 tron "%d bytes/sect x %"PRIu64" sectors\n",
159 1.57 cegger tbuf, sc->sc_ncylinders, sc->sc_nheads,
160 1.1 ad sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
161 1.68 kiyohara sc->sc_disksize512 = sc->sc_secperunit * sc->sc_secsize / DEV_BSIZE;
162 1.1 ad
163 1.71 christos ld_set_geometry(sc);
164 1.43 riz
165 1.1 ad /* Attach the device into the rnd source list. */
166 1.63 tron rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
167 1.75 tls RND_TYPE_DISK, RND_FLAG_DEFAULT);
168 1.1 ad
169 1.55 jmcneill /* Register with PMF */
170 1.67 jmcneill if (!pmf_device_register1(sc->sc_dv, ld_suspend, NULL, ld_shutdown))
171 1.63 tron aprint_error_dev(sc->sc_dv,
172 1.55 jmcneill "couldn't establish power handler\n");
173 1.55 jmcneill
174 1.38 yamt bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
175 1.30 thorpej
176 1.30 thorpej /* Discover wedges on this disk. */
177 1.63 tron config_interrupts(sc->sc_dv, ld_config_interrupts);
178 1.1 ad }
179 1.1 ad
180 1.3 ad int
181 1.37 christos ldadjqparam(struct ld_softc *sc, int xmax)
182 1.3 ad {
183 1.24 thorpej int s;
184 1.7 ad
185 1.7 ad s = splbio();
186 1.37 christos sc->sc_maxqueuecnt = xmax;
187 1.7 ad splx(s);
188 1.7 ad
189 1.24 thorpej return (0);
190 1.7 ad }
191 1.7 ad
192 1.7 ad int
193 1.7 ad ldbegindetach(struct ld_softc *sc, int flags)
194 1.7 ad {
195 1.24 thorpej int s, rv = 0;
196 1.7 ad
197 1.7 ad if ((sc->sc_flags & LDF_ENABLED) == 0)
198 1.7 ad return (0);
199 1.3 ad
200 1.66 dyoung rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
201 1.66 dyoung
202 1.66 dyoung if (rv != 0)
203 1.66 dyoung return rv;
204 1.3 ad
205 1.3 ad s = splbio();
206 1.24 thorpej sc->sc_maxqueuecnt = 0;
207 1.7 ad sc->sc_flags |= LDF_DETACH;
208 1.24 thorpej while (sc->sc_queuecnt > 0) {
209 1.24 thorpej sc->sc_flags |= LDF_DRAIN;
210 1.24 thorpej rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
211 1.24 thorpej if (rv)
212 1.24 thorpej break;
213 1.24 thorpej }
214 1.3 ad splx(s);
215 1.7 ad
216 1.7 ad return (rv);
217 1.3 ad }
218 1.3 ad
219 1.2 ad void
220 1.7 ad ldenddetach(struct ld_softc *sc)
221 1.2 ad {
222 1.13 drochner int s, bmaj, cmaj, i, mn;
223 1.2 ad
224 1.7 ad if ((sc->sc_flags & LDF_ENABLED) == 0)
225 1.7 ad return;
226 1.7 ad
227 1.2 ad /* Wait for commands queued with the hardware to complete. */
228 1.2 ad if (sc->sc_queuecnt != 0)
229 1.7 ad if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
230 1.63 tron printf("%s: not drained\n", device_xname(sc->sc_dv));
231 1.2 ad
232 1.2 ad /* Locate the major numbers. */
233 1.16 gehenna bmaj = bdevsw_lookup_major(&ld_bdevsw);
234 1.16 gehenna cmaj = cdevsw_lookup_major(&ld_cdevsw);
235 1.2 ad
236 1.2 ad /* Kill off any queued buffers. */
237 1.2 ad s = splbio();
238 1.38 yamt bufq_drain(sc->sc_bufq);
239 1.36 yamt splx(s);
240 1.36 yamt
241 1.38 yamt bufq_free(sc->sc_bufq);
242 1.2 ad
243 1.2 ad /* Nuke the vnodes for any open instances. */
244 1.13 drochner for (i = 0; i < MAXPARTITIONS; i++) {
245 1.63 tron mn = DISKMINOR(device_unit(sc->sc_dv), i);
246 1.13 drochner vdevgone(bmaj, mn, mn, VBLK);
247 1.13 drochner vdevgone(cmaj, mn, mn, VCHR);
248 1.13 drochner }
249 1.13 drochner
250 1.30 thorpej /* Delete all of our wedges. */
251 1.30 thorpej dkwedge_delall(&sc->sc_dk);
252 1.30 thorpej
253 1.2 ad /* Detach from the disk list. */
254 1.2 ad disk_detach(&sc->sc_dk);
255 1.50 ad disk_destroy(&sc->sc_dk);
256 1.2 ad
257 1.2 ad /* Unhook the entropy source. */
258 1.2 ad rnd_detach_source(&sc->sc_rnd_source);
259 1.2 ad
260 1.56 jmcneill /* Deregister with PMF */
261 1.63 tron pmf_device_deregister(sc->sc_dv);
262 1.56 jmcneill
263 1.24 thorpej /*
264 1.24 thorpej * XXX We can't really flush the cache here, beceause the
265 1.24 thorpej * XXX device may already be non-existent from the controller's
266 1.24 thorpej * XXX perspective.
267 1.24 thorpej */
268 1.24 thorpej #if 0
269 1.2 ad /* Flush the device's cache. */
270 1.2 ad if (sc->sc_flush != NULL)
271 1.62 simonb if ((*sc->sc_flush)(sc, 0) != 0)
272 1.70 chs aprint_error_dev(sc->sc_dv, "unable to flush cache\n");
273 1.24 thorpej #endif
274 1.61 ws mutex_destroy(&sc->sc_mutex);
275 1.2 ad }
276 1.2 ad
277 1.8 lukem /* ARGSUSED */
278 1.55 jmcneill static bool
279 1.67 jmcneill ld_suspend(device_t dev, const pmf_qual_t *qual)
280 1.67 jmcneill {
281 1.67 jmcneill return ld_shutdown(dev, 0);
282 1.67 jmcneill }
283 1.67 jmcneill
284 1.67 jmcneill /* ARGSUSED */
285 1.67 jmcneill static bool
286 1.55 jmcneill ld_shutdown(device_t dev, int flags)
287 1.1 ad {
288 1.55 jmcneill struct ld_softc *sc = device_private(dev);
289 1.1 ad
290 1.62 simonb if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
291 1.55 jmcneill printf("%s: unable to flush cache\n", device_xname(dev));
292 1.55 jmcneill return false;
293 1.1 ad }
294 1.55 jmcneill
295 1.55 jmcneill return true;
296 1.1 ad }
297 1.1 ad
298 1.8 lukem /* ARGSUSED */
299 1.29 thorpej static int
300 1.42 christos ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
301 1.1 ad {
302 1.1 ad struct ld_softc *sc;
303 1.30 thorpej int error, unit, part;
304 1.1 ad
305 1.1 ad unit = DISKUNIT(dev);
306 1.59 tsutsui if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
307 1.1 ad return (ENXIO);
308 1.1 ad if ((sc->sc_flags & LDF_ENABLED) == 0)
309 1.1 ad return (ENODEV);
310 1.1 ad part = DISKPART(dev);
311 1.30 thorpej
312 1.48 ad mutex_enter(&sc->sc_dk.dk_openlock);
313 1.1 ad
314 1.22 thorpej if (sc->sc_dk.dk_openmask == 0) {
315 1.22 thorpej /* Load the partition info if not already loaded. */
316 1.22 thorpej if ((sc->sc_flags & LDF_VLABEL) == 0)
317 1.22 thorpej ldgetdisklabel(sc);
318 1.22 thorpej }
319 1.1 ad
320 1.1 ad /* Check that the partition exists. */
321 1.1 ad if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
322 1.1 ad sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
323 1.30 thorpej error = ENXIO;
324 1.30 thorpej goto bad1;
325 1.1 ad }
326 1.1 ad
327 1.1 ad /* Ensure only one open at a time. */
328 1.1 ad switch (fmt) {
329 1.1 ad case S_IFCHR:
330 1.1 ad sc->sc_dk.dk_copenmask |= (1 << part);
331 1.1 ad break;
332 1.1 ad case S_IFBLK:
333 1.1 ad sc->sc_dk.dk_bopenmask |= (1 << part);
334 1.1 ad break;
335 1.1 ad }
336 1.1 ad sc->sc_dk.dk_openmask =
337 1.1 ad sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
338 1.1 ad
339 1.48 ad error = 0;
340 1.30 thorpej bad1:
341 1.48 ad mutex_exit(&sc->sc_dk.dk_openlock);
342 1.30 thorpej return (error);
343 1.1 ad }
344 1.1 ad
345 1.66 dyoung static int
346 1.66 dyoung ldlastclose(device_t self)
347 1.66 dyoung {
348 1.66 dyoung struct ld_softc *sc = device_private(self);
349 1.66 dyoung
350 1.66 dyoung if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
351 1.66 dyoung aprint_error_dev(self, "unable to flush cache\n");
352 1.66 dyoung if ((sc->sc_flags & LDF_KLABEL) == 0)
353 1.66 dyoung sc->sc_flags &= ~LDF_VLABEL;
354 1.66 dyoung
355 1.66 dyoung return 0;
356 1.66 dyoung }
357 1.66 dyoung
358 1.8 lukem /* ARGSUSED */
359 1.29 thorpej static int
360 1.42 christos ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
361 1.1 ad {
362 1.1 ad struct ld_softc *sc;
363 1.48 ad int part, unit;
364 1.1 ad
365 1.1 ad unit = DISKUNIT(dev);
366 1.1 ad part = DISKPART(dev);
367 1.59 tsutsui sc = device_lookup_private(&ld_cd, unit);
368 1.30 thorpej
369 1.48 ad mutex_enter(&sc->sc_dk.dk_openlock);
370 1.1 ad
371 1.1 ad switch (fmt) {
372 1.1 ad case S_IFCHR:
373 1.1 ad sc->sc_dk.dk_copenmask &= ~(1 << part);
374 1.1 ad break;
375 1.1 ad case S_IFBLK:
376 1.1 ad sc->sc_dk.dk_bopenmask &= ~(1 << part);
377 1.1 ad break;
378 1.1 ad }
379 1.1 ad sc->sc_dk.dk_openmask =
380 1.1 ad sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
381 1.1 ad
382 1.66 dyoung if (sc->sc_dk.dk_openmask == 0)
383 1.66 dyoung ldlastclose(sc->sc_dv);
384 1.1 ad
385 1.48 ad mutex_exit(&sc->sc_dk.dk_openlock);
386 1.1 ad return (0);
387 1.1 ad }
388 1.1 ad
389 1.8 lukem /* ARGSUSED */
390 1.29 thorpej static int
391 1.42 christos ldread(dev_t dev, struct uio *uio, int ioflag)
392 1.1 ad {
393 1.1 ad
394 1.1 ad return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
395 1.1 ad }
396 1.1 ad
397 1.8 lukem /* ARGSUSED */
398 1.29 thorpej static int
399 1.42 christos ldwrite(dev_t dev, struct uio *uio, int ioflag)
400 1.1 ad {
401 1.1 ad
402 1.1 ad return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
403 1.1 ad }
404 1.1 ad
405 1.8 lukem /* ARGSUSED */
406 1.29 thorpej static int
407 1.46 christos ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
408 1.1 ad {
409 1.1 ad struct ld_softc *sc;
410 1.80 christos int unit, error;
411 1.4 fvdl #ifdef __HAVE_OLD_DISKLABEL
412 1.6 itojun struct disklabel newlabel;
413 1.4 fvdl #endif
414 1.6 itojun struct disklabel *lp;
415 1.1 ad
416 1.1 ad unit = DISKUNIT(dev);
417 1.59 tsutsui sc = device_lookup_private(&ld_cd, unit);
418 1.1 ad
419 1.80 christos error = disk_ioctl(&sc->sc_dk, dev, cmd, addr, flag, l);
420 1.43 riz if (error != EPASSTHROUGH)
421 1.43 riz return (error);
422 1.43 riz
423 1.47 tron error = 0;
424 1.1 ad switch (cmd) {
425 1.1 ad case DIOCWDINFO:
426 1.1 ad case DIOCSDINFO:
427 1.4 fvdl #ifdef __HAVE_OLD_DISKLABEL
428 1.4 fvdl case ODIOCWDINFO:
429 1.4 fvdl case ODIOCSDINFO:
430 1.4 fvdl
431 1.4 fvdl if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
432 1.4 fvdl memset(&newlabel, 0, sizeof newlabel);
433 1.4 fvdl memcpy(&newlabel, addr, sizeof (struct olddisklabel));
434 1.4 fvdl lp = &newlabel;
435 1.4 fvdl } else
436 1.4 fvdl #endif
437 1.4 fvdl lp = (struct disklabel *)addr;
438 1.4 fvdl
439 1.1 ad if ((flag & FWRITE) == 0)
440 1.1 ad return (EBADF);
441 1.1 ad
442 1.48 ad mutex_enter(&sc->sc_dk.dk_openlock);
443 1.1 ad sc->sc_flags |= LDF_LABELLING;
444 1.1 ad
445 1.1 ad error = setdisklabel(sc->sc_dk.dk_label,
446 1.4 fvdl lp, /*sc->sc_dk.dk_openmask : */0,
447 1.1 ad sc->sc_dk.dk_cpulabel);
448 1.4 fvdl if (error == 0 && (cmd == DIOCWDINFO
449 1.4 fvdl #ifdef __HAVE_OLD_DISKLABEL
450 1.4 fvdl || cmd == ODIOCWDINFO
451 1.4 fvdl #endif
452 1.4 fvdl ))
453 1.1 ad error = writedisklabel(
454 1.35 perry MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
455 1.35 perry ldstrategy, sc->sc_dk.dk_label,
456 1.1 ad sc->sc_dk.dk_cpulabel);
457 1.1 ad
458 1.1 ad sc->sc_flags &= ~LDF_LABELLING;
459 1.48 ad mutex_exit(&sc->sc_dk.dk_openlock);
460 1.1 ad break;
461 1.1 ad
462 1.22 thorpej case DIOCKLABEL:
463 1.22 thorpej if ((flag & FWRITE) == 0)
464 1.22 thorpej return (EBADF);
465 1.22 thorpej if (*(int *)addr)
466 1.22 thorpej sc->sc_flags |= LDF_KLABEL;
467 1.22 thorpej else
468 1.22 thorpej sc->sc_flags &= ~LDF_KLABEL;
469 1.22 thorpej break;
470 1.22 thorpej
471 1.1 ad case DIOCWLABEL:
472 1.1 ad if ((flag & FWRITE) == 0)
473 1.1 ad return (EBADF);
474 1.1 ad if (*(int *)addr)
475 1.1 ad sc->sc_flags |= LDF_WLABEL;
476 1.1 ad else
477 1.1 ad sc->sc_flags &= ~LDF_WLABEL;
478 1.1 ad break;
479 1.1 ad
480 1.1 ad case DIOCGDEFLABEL:
481 1.1 ad ldgetdefaultlabel(sc, (struct disklabel *)addr);
482 1.1 ad break;
483 1.4 fvdl
484 1.4 fvdl #ifdef __HAVE_OLD_DISKLABEL
485 1.4 fvdl case ODIOCGDEFLABEL:
486 1.4 fvdl ldgetdefaultlabel(sc, &newlabel);
487 1.4 fvdl if (newlabel.d_npartitions > OLDMAXPARTITIONS)
488 1.5 fvdl return ENOTTY;
489 1.4 fvdl memcpy(addr, &newlabel, sizeof (struct olddisklabel));
490 1.4 fvdl break;
491 1.4 fvdl #endif
492 1.1 ad
493 1.32 thorpej case DIOCCACHESYNC:
494 1.32 thorpej /*
495 1.32 thorpej * XXX Do we really need to care about having a writable
496 1.32 thorpej * file descriptor here?
497 1.32 thorpej */
498 1.32 thorpej if ((flag & FWRITE) == 0)
499 1.32 thorpej error = EBADF;
500 1.32 thorpej else if (sc->sc_flush)
501 1.62 simonb error = (*sc->sc_flush)(sc, 0);
502 1.32 thorpej else
503 1.32 thorpej error = 0; /* XXX Error out instead? */
504 1.32 thorpej break;
505 1.32 thorpej
506 1.51 xtraeme case DIOCGSTRATEGY:
507 1.51 xtraeme {
508 1.51 xtraeme struct disk_strategy *dks = (void *)addr;
509 1.51 xtraeme
510 1.52 xtraeme mutex_enter(&sc->sc_mutex);
511 1.51 xtraeme strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
512 1.51 xtraeme sizeof(dks->dks_name));
513 1.52 xtraeme mutex_exit(&sc->sc_mutex);
514 1.51 xtraeme dks->dks_paramlen = 0;
515 1.51 xtraeme
516 1.51 xtraeme return 0;
517 1.51 xtraeme }
518 1.51 xtraeme case DIOCSSTRATEGY:
519 1.51 xtraeme {
520 1.51 xtraeme struct disk_strategy *dks = (void *)addr;
521 1.76 matt struct bufq_state *new_bufq, *old_bufq;
522 1.30 thorpej
523 1.51 xtraeme if ((flag & FWRITE) == 0)
524 1.51 xtraeme return EPERM;
525 1.51 xtraeme
526 1.51 xtraeme if (dks->dks_param != NULL)
527 1.51 xtraeme return EINVAL;
528 1.51 xtraeme
529 1.51 xtraeme dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
530 1.76 matt error = bufq_alloc(&new_bufq, dks->dks_name,
531 1.51 xtraeme BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
532 1.51 xtraeme if (error)
533 1.51 xtraeme return error;
534 1.51 xtraeme
535 1.52 xtraeme mutex_enter(&sc->sc_mutex);
536 1.76 matt old_bufq = sc->sc_bufq;
537 1.76 matt bufq_move(new_bufq, old_bufq);
538 1.76 matt sc->sc_bufq = new_bufq;
539 1.52 xtraeme mutex_exit(&sc->sc_mutex);
540 1.76 matt bufq_free(old_bufq);
541 1.51 xtraeme
542 1.51 xtraeme return 0;
543 1.51 xtraeme }
544 1.1 ad default:
545 1.1 ad error = ENOTTY;
546 1.1 ad break;
547 1.1 ad }
548 1.1 ad
549 1.1 ad return (error);
550 1.1 ad }
551 1.1 ad
552 1.29 thorpej static void
553 1.1 ad ldstrategy(struct buf *bp)
554 1.1 ad {
555 1.1 ad struct ld_softc *sc;
556 1.23 thorpej struct disklabel *lp;
557 1.23 thorpej daddr_t blkno;
558 1.23 thorpej int s, part;
559 1.1 ad
560 1.59 tsutsui sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
561 1.23 thorpej part = DISKPART(bp->b_dev);
562 1.1 ad
563 1.7 ad if ((sc->sc_flags & LDF_DETACH) != 0) {
564 1.2 ad bp->b_error = EIO;
565 1.49 ad goto done;
566 1.2 ad }
567 1.2 ad
568 1.1 ad lp = sc->sc_dk.dk_label;
569 1.1 ad
570 1.1 ad /*
571 1.1 ad * The transfer must be a whole number of blocks and the offset must
572 1.1 ad * not be negative.
573 1.1 ad */
574 1.1 ad if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
575 1.23 thorpej bp->b_error = EINVAL;
576 1.49 ad goto done;
577 1.1 ad }
578 1.1 ad
579 1.23 thorpej /* If it's a null transfer, return immediately. */
580 1.23 thorpej if (bp->b_bcount == 0)
581 1.23 thorpej goto done;
582 1.1 ad
583 1.1 ad /*
584 1.1 ad * Do bounds checking and adjust the transfer. If error, process.
585 1.1 ad * If past the end of partition, just return.
586 1.1 ad */
587 1.68 kiyohara if (part == RAW_PART) {
588 1.68 kiyohara if (bounds_check_with_mediasize(bp, DEV_BSIZE,
589 1.68 kiyohara sc->sc_disksize512) <= 0)
590 1.68 kiyohara goto done;
591 1.68 kiyohara } else {
592 1.68 kiyohara if (bounds_check_with_label(&sc->sc_dk, bp,
593 1.68 kiyohara (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0)
594 1.68 kiyohara goto done;
595 1.1 ad }
596 1.1 ad
597 1.1 ad /*
598 1.23 thorpej * Convert the block number to absolute and put it in terms
599 1.23 thorpej * of the device's logical block size.
600 1.1 ad */
601 1.23 thorpej if (lp->d_secsize == DEV_BSIZE)
602 1.23 thorpej blkno = bp->b_blkno;
603 1.23 thorpej else if (lp->d_secsize > DEV_BSIZE)
604 1.23 thorpej blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
605 1.1 ad else
606 1.23 thorpej blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
607 1.1 ad
608 1.11 thorpej if (part != RAW_PART)
609 1.23 thorpej blkno += lp->d_partitions[part].p_offset;
610 1.23 thorpej
611 1.23 thorpej bp->b_rawblkno = blkno;
612 1.1 ad
613 1.1 ad s = splbio();
614 1.44 ad ldstart(sc, bp);
615 1.1 ad splx(s);
616 1.23 thorpej return;
617 1.23 thorpej
618 1.23 thorpej done:
619 1.23 thorpej bp->b_resid = bp->b_bcount;
620 1.23 thorpej biodone(bp);
621 1.23 thorpej }
622 1.23 thorpej
623 1.23 thorpej static void
624 1.44 ad ldstart(struct ld_softc *sc, struct buf *bp)
625 1.23 thorpej {
626 1.23 thorpej int error;
627 1.1 ad
628 1.44 ad mutex_enter(&sc->sc_mutex);
629 1.44 ad
630 1.44 ad if (bp != NULL)
631 1.64 yamt bufq_put(sc->sc_bufq, bp);
632 1.44 ad
633 1.23 thorpej while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
634 1.23 thorpej /* See if there is work to do. */
635 1.64 yamt if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
636 1.23 thorpej break;
637 1.23 thorpej
638 1.23 thorpej disk_busy(&sc->sc_dk);
639 1.23 thorpej sc->sc_queuecnt++;
640 1.23 thorpej
641 1.23 thorpej if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
642 1.23 thorpej /*
643 1.23 thorpej * The back-end is running the job; remove it from
644 1.23 thorpej * the queue.
645 1.23 thorpej */
646 1.64 yamt (void) bufq_get(sc->sc_bufq);
647 1.23 thorpej } else {
648 1.23 thorpej disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
649 1.23 thorpej sc->sc_queuecnt--;
650 1.23 thorpej if (error == EAGAIN) {
651 1.23 thorpej /*
652 1.23 thorpej * Temporary resource shortage in the
653 1.23 thorpej * back-end; just defer the job until
654 1.23 thorpej * later.
655 1.23 thorpej *
656 1.23 thorpej * XXX We might consider a watchdog timer
657 1.23 thorpej * XXX to make sure we are kicked into action.
658 1.23 thorpej */
659 1.23 thorpej break;
660 1.23 thorpej } else {
661 1.64 yamt (void) bufq_get(sc->sc_bufq);
662 1.23 thorpej bp->b_error = error;
663 1.23 thorpej bp->b_resid = bp->b_bcount;
664 1.44 ad mutex_exit(&sc->sc_mutex);
665 1.23 thorpej biodone(bp);
666 1.44 ad mutex_enter(&sc->sc_mutex);
667 1.23 thorpej }
668 1.23 thorpej }
669 1.1 ad }
670 1.44 ad
671 1.44 ad mutex_exit(&sc->sc_mutex);
672 1.1 ad }
673 1.1 ad
674 1.1 ad void
675 1.1 ad lddone(struct ld_softc *sc, struct buf *bp)
676 1.1 ad {
677 1.1 ad
678 1.49 ad if (bp->b_error != 0) {
679 1.1 ad diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
680 1.1 ad printf("\n");
681 1.1 ad }
682 1.1 ad
683 1.18 mrg disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
684 1.18 mrg (bp->b_flags & B_READ));
685 1.1 ad rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
686 1.1 ad biodone(bp);
687 1.1 ad
688 1.44 ad mutex_enter(&sc->sc_mutex);
689 1.7 ad if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
690 1.24 thorpej if ((sc->sc_flags & LDF_DRAIN) != 0) {
691 1.24 thorpej sc->sc_flags &= ~LDF_DRAIN;
692 1.7 ad wakeup(&sc->sc_queuecnt);
693 1.24 thorpej }
694 1.44 ad mutex_exit(&sc->sc_mutex);
695 1.44 ad ldstart(sc, NULL);
696 1.44 ad } else
697 1.44 ad mutex_exit(&sc->sc_mutex);
698 1.1 ad }
699 1.1 ad
700 1.29 thorpej static int
701 1.1 ad ldsize(dev_t dev)
702 1.1 ad {
703 1.1 ad struct ld_softc *sc;
704 1.1 ad int part, unit, omask, size;
705 1.1 ad
706 1.1 ad unit = DISKUNIT(dev);
707 1.59 tsutsui if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
708 1.1 ad return (ENODEV);
709 1.1 ad if ((sc->sc_flags & LDF_ENABLED) == 0)
710 1.1 ad return (ENODEV);
711 1.1 ad part = DISKPART(dev);
712 1.1 ad
713 1.1 ad omask = sc->sc_dk.dk_openmask & (1 << part);
714 1.1 ad
715 1.1 ad if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
716 1.1 ad return (-1);
717 1.1 ad else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
718 1.1 ad size = -1;
719 1.1 ad else
720 1.1 ad size = sc->sc_dk.dk_label->d_partitions[part].p_size *
721 1.1 ad (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
722 1.1 ad if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
723 1.1 ad return (-1);
724 1.1 ad
725 1.1 ad return (size);
726 1.1 ad }
727 1.1 ad
728 1.1 ad /*
729 1.1 ad * Load the label information from the specified device.
730 1.1 ad */
731 1.1 ad static void
732 1.1 ad ldgetdisklabel(struct ld_softc *sc)
733 1.1 ad {
734 1.1 ad const char *errstring;
735 1.1 ad
736 1.1 ad ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
737 1.1 ad
738 1.1 ad /* Call the generic disklabel extraction routine. */
739 1.63 tron errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
740 1.40 thorpej RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
741 1.1 ad if (errstring != NULL)
742 1.63 tron printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
743 1.22 thorpej
744 1.22 thorpej /* In-core label now valid. */
745 1.22 thorpej sc->sc_flags |= LDF_VLABEL;
746 1.1 ad }
747 1.1 ad
748 1.1 ad /*
749 1.1 ad * Construct a ficticious label.
750 1.1 ad */
751 1.1 ad static void
752 1.1 ad ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
753 1.1 ad {
754 1.1 ad
755 1.1 ad memset(lp, 0, sizeof(struct disklabel));
756 1.1 ad
757 1.1 ad lp->d_secsize = sc->sc_secsize;
758 1.1 ad lp->d_ntracks = sc->sc_nheads;
759 1.1 ad lp->d_nsectors = sc->sc_nsectors;
760 1.1 ad lp->d_ncylinders = sc->sc_ncylinders;
761 1.1 ad lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
762 1.1 ad lp->d_type = DTYPE_LD;
763 1.21 itojun strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
764 1.21 itojun strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
765 1.77 mlelstv if (sc->sc_secperunit > UINT32_MAX)
766 1.77 mlelstv lp->d_secperunit = UINT32_MAX;
767 1.77 mlelstv else
768 1.77 mlelstv lp->d_secperunit = sc->sc_secperunit;
769 1.1 ad lp->d_rpm = 7200;
770 1.1 ad lp->d_interleave = 1;
771 1.1 ad lp->d_flags = 0;
772 1.1 ad
773 1.1 ad lp->d_partitions[RAW_PART].p_offset = 0;
774 1.77 mlelstv lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
775 1.1 ad lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
776 1.1 ad lp->d_npartitions = RAW_PART + 1;
777 1.1 ad
778 1.1 ad lp->d_magic = DISKMAGIC;
779 1.1 ad lp->d_magic2 = DISKMAGIC;
780 1.1 ad lp->d_checksum = dkcksum(lp);
781 1.1 ad }
782 1.1 ad
783 1.1 ad /*
784 1.1 ad * Take a dump.
785 1.1 ad */
786 1.29 thorpej static int
787 1.46 christos lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
788 1.1 ad {
789 1.46 christos char *va = vav;
790 1.1 ad struct ld_softc *sc;
791 1.1 ad struct disklabel *lp;
792 1.1 ad int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
793 1.1 ad static int dumping;
794 1.1 ad
795 1.1 ad unit = DISKUNIT(dev);
796 1.59 tsutsui if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
797 1.1 ad return (ENXIO);
798 1.1 ad if ((sc->sc_flags & LDF_ENABLED) == 0)
799 1.1 ad return (ENODEV);
800 1.3 ad if (sc->sc_dump == NULL)
801 1.3 ad return (ENXIO);
802 1.3 ad
803 1.3 ad /* Check if recursive dump; if so, punt. */
804 1.3 ad if (dumping)
805 1.3 ad return (EFAULT);
806 1.3 ad dumping = 1;
807 1.1 ad
808 1.1 ad /* Convert to disk sectors. Request must be a multiple of size. */
809 1.3 ad part = DISKPART(dev);
810 1.1 ad lp = sc->sc_dk.dk_label;
811 1.1 ad if ((size % lp->d_secsize) != 0)
812 1.1 ad return (EFAULT);
813 1.1 ad towrt = size / lp->d_secsize;
814 1.1 ad blkno = dbtob(blkno) / lp->d_secsize; /* blkno in DEV_BSIZE units */
815 1.1 ad
816 1.1 ad nsects = lp->d_partitions[part].p_size;
817 1.1 ad sectoff = lp->d_partitions[part].p_offset;
818 1.1 ad
819 1.1 ad /* Check transfer bounds against partition size. */
820 1.1 ad if ((blkno < 0) || ((blkno + towrt) > nsects))
821 1.1 ad return (EINVAL);
822 1.1 ad
823 1.1 ad /* Offset block number to start of partition. */
824 1.1 ad blkno += sectoff;
825 1.1 ad
826 1.1 ad /* Start dumping and return when done. */
827 1.3 ad maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
828 1.1 ad while (towrt > 0) {
829 1.3 ad nblk = min(maxblkcnt, towrt);
830 1.1 ad
831 1.1 ad if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
832 1.1 ad return (rv);
833 1.1 ad
834 1.1 ad towrt -= nblk;
835 1.1 ad blkno += nblk;
836 1.1 ad va += nblk * sc->sc_secsize;
837 1.1 ad }
838 1.1 ad
839 1.1 ad dumping = 0;
840 1.1 ad return (0);
841 1.1 ad }
842 1.1 ad
843 1.1 ad /*
844 1.1 ad * Adjust the size of a transfer.
845 1.1 ad */
846 1.1 ad static void
847 1.1 ad ldminphys(struct buf *bp)
848 1.1 ad {
849 1.1 ad struct ld_softc *sc;
850 1.1 ad
851 1.59 tsutsui sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
852 1.1 ad
853 1.1 ad if (bp->b_bcount > sc->sc_maxxfer)
854 1.1 ad bp->b_bcount = sc->sc_maxxfer;
855 1.1 ad minphys(bp);
856 1.1 ad }
857 1.43 riz
858 1.43 riz static void
859 1.71 christos ld_set_geometry(struct ld_softc *ld)
860 1.43 riz {
861 1.71 christos struct disk_geom *dg = &ld->sc_dk.dk_geom;
862 1.43 riz
863 1.71 christos memset(dg, 0, sizeof(*dg));
864 1.43 riz
865 1.71 christos dg->dg_secperunit = ld->sc_secperunit;
866 1.71 christos dg->dg_secsize = ld->sc_secsize;
867 1.71 christos dg->dg_nsectors = ld->sc_nsectors;
868 1.71 christos dg->dg_ntracks = ld->sc_nheads;
869 1.71 christos dg->dg_ncylinders = ld->sc_ncylinders;
870 1.43 riz
871 1.71 christos disk_set_info(ld->sc_dv, &ld->sc_dk, NULL);
872 1.43 riz }
873 1.45 riz
874 1.45 riz static void
875 1.65 cegger ld_config_interrupts(device_t d)
876 1.45 riz {
877 1.60 cube struct ld_softc *sc = device_private(d);
878 1.45 riz dkwedge_discover(&sc->sc_dk);
879 1.45 riz }
880