xref: /netbsd-src/sys/dev/ld.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: ld.c,v 1.54 2008/01/04 21:17:48 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran and Charles M. Hannum.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Disk driver for use by RAID controllers.
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.54 2008/01/04 21:17:48 ad Exp $");
45 
46 #include "rnd.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/device.h>
52 #include <sys/queue.h>
53 #include <sys/proc.h>
54 #include <sys/buf.h>
55 #include <sys/bufq.h>
56 #include <sys/endian.h>
57 #include <sys/disklabel.h>
58 #include <sys/disk.h>
59 #include <sys/dkio.h>
60 #include <sys/stat.h>
61 #include <sys/conf.h>
62 #include <sys/fcntl.h>
63 #include <sys/vnode.h>
64 #include <sys/syslog.h>
65 #include <sys/mutex.h>
66 #if NRND > 0
67 #include <sys/rnd.h>
68 #endif
69 
70 #include <dev/ldvar.h>
71 
72 #include <prop/proplib.h>
73 
74 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
75 static void	ldgetdisklabel(struct ld_softc *);
76 static void	ldminphys(struct buf *bp);
77 static void	ldshutdown(void *);
78 static void	ldstart(struct ld_softc *, struct buf *);
79 static void	ld_set_properties(struct ld_softc *);
80 static void	ld_config_interrupts (struct device *);
81 
82 extern struct	cfdriver ld_cd;
83 
84 static dev_type_open(ldopen);
85 static dev_type_close(ldclose);
86 static dev_type_read(ldread);
87 static dev_type_write(ldwrite);
88 static dev_type_ioctl(ldioctl);
89 static dev_type_strategy(ldstrategy);
90 static dev_type_dump(lddump);
91 static dev_type_size(ldsize);
92 
93 const struct bdevsw ld_bdevsw = {
94 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
95 };
96 
97 const struct cdevsw ld_cdevsw = {
98 	ldopen, ldclose, ldread, ldwrite, ldioctl,
99 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
100 };
101 
102 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
103 static void	*ld_sdh;
104 
105 void
106 ldattach(struct ld_softc *sc)
107 {
108 	char tbuf[9];
109 
110 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
111 
112 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
113 		aprint_normal("%s: disabled\n", sc->sc_dv.dv_xname);
114 		return;
115 	}
116 
117 	/* Initialise and attach the disk structure. */
118 	disk_init(&sc->sc_dk, sc->sc_dv.dv_xname, &lddkdriver);
119 	disk_attach(&sc->sc_dk);
120 
121 	if (sc->sc_maxxfer > MAXPHYS)
122 		sc->sc_maxxfer = MAXPHYS;
123 
124 	/* Build synthetic geometry if necessary. */
125 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
126 	    sc->sc_ncylinders == 0) {
127 		uint64_t ncyl;
128 
129 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
130 			sc->sc_nheads = 16;
131 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
132 			sc->sc_nheads = 32;
133 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
134 			sc->sc_nheads = 64;
135 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
136 			sc->sc_nheads = 128;
137 		else
138 			sc->sc_nheads = 255;
139 
140 		sc->sc_nsectors = 63;
141 		sc->sc_ncylinders = INT_MAX;
142 		ncyl = sc->sc_secperunit /
143 		    (sc->sc_nheads * sc->sc_nsectors);
144 		if (ncyl < INT_MAX)
145 			sc->sc_ncylinders = (int)ncyl;
146 	}
147 
148 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
149 	    sc->sc_secsize);
150 	aprint_normal("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %"PRIu64" sectors\n",
151 	    sc->sc_dv.dv_xname, tbuf, sc->sc_ncylinders, sc->sc_nheads,
152 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
153 
154 	ld_set_properties(sc);
155 
156 #if NRND > 0
157 	/* Attach the device into the rnd source list. */
158 	rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
159 	    RND_TYPE_DISK, 0);
160 #endif
161 
162 	/* Set the `shutdownhook'. */
163 	if (ld_sdh == NULL)
164 		ld_sdh = shutdownhook_establish(ldshutdown, NULL);
165 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
166 
167 	/* Discover wedges on this disk. */
168 	config_interrupts(&sc->sc_dv, ld_config_interrupts);
169 }
170 
171 int
172 ldadjqparam(struct ld_softc *sc, int xmax)
173 {
174 	int s;
175 
176 	s = splbio();
177 	sc->sc_maxqueuecnt = xmax;
178 	splx(s);
179 
180 	return (0);
181 }
182 
183 int
184 ldbegindetach(struct ld_softc *sc, int flags)
185 {
186 	int s, rv = 0;
187 
188 	if ((sc->sc_flags & LDF_ENABLED) == 0)
189 		return (0);
190 
191 	if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
192 		return (EBUSY);
193 
194 	s = splbio();
195 	sc->sc_maxqueuecnt = 0;
196 	sc->sc_flags |= LDF_DETACH;
197 	while (sc->sc_queuecnt > 0) {
198 		sc->sc_flags |= LDF_DRAIN;
199 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
200 		if (rv)
201 			break;
202 	}
203 	splx(s);
204 
205 	return (rv);
206 }
207 
208 void
209 ldenddetach(struct ld_softc *sc)
210 {
211 	int s, bmaj, cmaj, i, mn;
212 
213 	if ((sc->sc_flags & LDF_ENABLED) == 0)
214 		return;
215 
216 	/* Wait for commands queued with the hardware to complete. */
217 	if (sc->sc_queuecnt != 0)
218 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
219 			printf("%s: not drained\n", sc->sc_dv.dv_xname);
220 
221 	/* Locate the major numbers. */
222 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
223 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
224 
225 	/* Kill off any queued buffers. */
226 	s = splbio();
227 	bufq_drain(sc->sc_bufq);
228 	splx(s);
229 
230 	bufq_free(sc->sc_bufq);
231 
232 	/* Nuke the vnodes for any open instances. */
233 	for (i = 0; i < MAXPARTITIONS; i++) {
234 		mn = DISKMINOR(device_unit(&sc->sc_dv), i);
235 		vdevgone(bmaj, mn, mn, VBLK);
236 		vdevgone(cmaj, mn, mn, VCHR);
237 	}
238 
239 	/* Delete all of our wedges. */
240 	dkwedge_delall(&sc->sc_dk);
241 
242 	/* Detach from the disk list. */
243 	disk_detach(&sc->sc_dk);
244 	disk_destroy(&sc->sc_dk);
245 
246 #if NRND > 0
247 	/* Unhook the entropy source. */
248 	rnd_detach_source(&sc->sc_rnd_source);
249 #endif
250 
251 	/*
252 	 * XXX We can't really flush the cache here, beceause the
253 	 * XXX device may already be non-existent from the controller's
254 	 * XXX perspective.
255 	 */
256 #if 0
257 	/* Flush the device's cache. */
258 	if (sc->sc_flush != NULL)
259 		if ((*sc->sc_flush)(sc) != 0)
260 			printf("%s: unable to flush cache\n",
261 			    sc->sc_dv.dv_xname);
262 #endif
263 }
264 
265 /* ARGSUSED */
266 static void
267 ldshutdown(void *cookie)
268 {
269 	struct ld_softc *sc;
270 	int i;
271 
272 	for (i = 0; i < ld_cd.cd_ndevs; i++) {
273 		if ((sc = device_lookup(&ld_cd, i)) == NULL)
274 			continue;
275 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
276 			printf("%s: unable to flush cache\n",
277 			    sc->sc_dv.dv_xname);
278 	}
279 }
280 
281 /* ARGSUSED */
282 static int
283 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
284 {
285 	struct ld_softc *sc;
286 	int error, unit, part;
287 
288 	unit = DISKUNIT(dev);
289 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
290 		return (ENXIO);
291 	if ((sc->sc_flags & LDF_ENABLED) == 0)
292 		return (ENODEV);
293 	part = DISKPART(dev);
294 
295 	mutex_enter(&sc->sc_dk.dk_openlock);
296 
297 	if (sc->sc_dk.dk_openmask == 0) {
298 		/* Load the partition info if not already loaded. */
299 		if ((sc->sc_flags & LDF_VLABEL) == 0)
300 			ldgetdisklabel(sc);
301 	}
302 
303 	/* Check that the partition exists. */
304 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
305 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
306 		error = ENXIO;
307 		goto bad1;
308 	}
309 
310 	/* Ensure only one open at a time. */
311 	switch (fmt) {
312 	case S_IFCHR:
313 		sc->sc_dk.dk_copenmask |= (1 << part);
314 		break;
315 	case S_IFBLK:
316 		sc->sc_dk.dk_bopenmask |= (1 << part);
317 		break;
318 	}
319 	sc->sc_dk.dk_openmask =
320 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
321 
322 	error = 0;
323  bad1:
324 	mutex_exit(&sc->sc_dk.dk_openlock);
325 	return (error);
326 }
327 
328 /* ARGSUSED */
329 static int
330 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
331 {
332 	struct ld_softc *sc;
333 	int part, unit;
334 
335 	unit = DISKUNIT(dev);
336 	part = DISKPART(dev);
337 	sc = device_lookup(&ld_cd, unit);
338 
339 	mutex_enter(&sc->sc_dk.dk_openlock);
340 
341 	switch (fmt) {
342 	case S_IFCHR:
343 		sc->sc_dk.dk_copenmask &= ~(1 << part);
344 		break;
345 	case S_IFBLK:
346 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
347 		break;
348 	}
349 	sc->sc_dk.dk_openmask =
350 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
351 
352 	if (sc->sc_dk.dk_openmask == 0) {
353 		if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
354 			printf("%s: unable to flush cache\n",
355 			    sc->sc_dv.dv_xname);
356 		if ((sc->sc_flags & LDF_KLABEL) == 0)
357 			sc->sc_flags &= ~LDF_VLABEL;
358 	}
359 
360 	mutex_exit(&sc->sc_dk.dk_openlock);
361 	return (0);
362 }
363 
364 /* ARGSUSED */
365 static int
366 ldread(dev_t dev, struct uio *uio, int ioflag)
367 {
368 
369 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
370 }
371 
372 /* ARGSUSED */
373 static int
374 ldwrite(dev_t dev, struct uio *uio, int ioflag)
375 {
376 
377 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
378 }
379 
380 /* ARGSUSED */
381 static int
382 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
383 {
384 	struct ld_softc *sc;
385 	int part, unit, error;
386 #ifdef __HAVE_OLD_DISKLABEL
387 	struct disklabel newlabel;
388 #endif
389 	struct disklabel *lp;
390 
391 	unit = DISKUNIT(dev);
392 	part = DISKPART(dev);
393 	sc = device_lookup(&ld_cd, unit);
394 
395 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
396 	if (error != EPASSTHROUGH)
397 		return (error);
398 
399 	error = 0;
400 	switch (cmd) {
401 	case DIOCGDINFO:
402 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
403 		return (0);
404 
405 #ifdef __HAVE_OLD_DISKLABEL
406 	case ODIOCGDINFO:
407 		newlabel = *(sc->sc_dk.dk_label);
408 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
409 			return ENOTTY;
410 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
411 		return (0);
412 #endif
413 
414 	case DIOCGPART:
415 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
416 		((struct partinfo *)addr)->part =
417 		    &sc->sc_dk.dk_label->d_partitions[part];
418 		break;
419 
420 	case DIOCWDINFO:
421 	case DIOCSDINFO:
422 #ifdef __HAVE_OLD_DISKLABEL
423 	case ODIOCWDINFO:
424 	case ODIOCSDINFO:
425 
426 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
427 			memset(&newlabel, 0, sizeof newlabel);
428 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
429 			lp = &newlabel;
430 		} else
431 #endif
432 		lp = (struct disklabel *)addr;
433 
434 		if ((flag & FWRITE) == 0)
435 			return (EBADF);
436 
437 		mutex_enter(&sc->sc_dk.dk_openlock);
438 		sc->sc_flags |= LDF_LABELLING;
439 
440 		error = setdisklabel(sc->sc_dk.dk_label,
441 		    lp, /*sc->sc_dk.dk_openmask : */0,
442 		    sc->sc_dk.dk_cpulabel);
443 		if (error == 0 && (cmd == DIOCWDINFO
444 #ifdef __HAVE_OLD_DISKLABEL
445 		    || cmd == ODIOCWDINFO
446 #endif
447 		    ))
448 			error = writedisklabel(
449 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
450 			    ldstrategy, sc->sc_dk.dk_label,
451 			    sc->sc_dk.dk_cpulabel);
452 
453 		sc->sc_flags &= ~LDF_LABELLING;
454 		mutex_exit(&sc->sc_dk.dk_openlock);
455 		break;
456 
457 	case DIOCKLABEL:
458 		if ((flag & FWRITE) == 0)
459 			return (EBADF);
460 		if (*(int *)addr)
461 			sc->sc_flags |= LDF_KLABEL;
462 		else
463 			sc->sc_flags &= ~LDF_KLABEL;
464 		break;
465 
466 	case DIOCWLABEL:
467 		if ((flag & FWRITE) == 0)
468 			return (EBADF);
469 		if (*(int *)addr)
470 			sc->sc_flags |= LDF_WLABEL;
471 		else
472 			sc->sc_flags &= ~LDF_WLABEL;
473 		break;
474 
475 	case DIOCGDEFLABEL:
476 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
477 		break;
478 
479 #ifdef __HAVE_OLD_DISKLABEL
480 	case ODIOCGDEFLABEL:
481 		ldgetdefaultlabel(sc, &newlabel);
482 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
483 			return ENOTTY;
484 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
485 		break;
486 #endif
487 
488 	case DIOCCACHESYNC:
489 		/*
490 		 * XXX Do we really need to care about having a writable
491 		 * file descriptor here?
492 		 */
493 		if ((flag & FWRITE) == 0)
494 			error = EBADF;
495 		else if (sc->sc_flush)
496 			error = (*sc->sc_flush)(sc);
497 		else
498 			error = 0;	/* XXX Error out instead? */
499 		break;
500 
501 	case DIOCAWEDGE:
502 	    {
503 	    	struct dkwedge_info *dkw = (void *) addr;
504 
505 		if ((flag & FWRITE) == 0)
506 			return (EBADF);
507 
508 		/* If the ioctl happens here, the parent is us. */
509 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
510 		return (dkwedge_add(dkw));
511 	    }
512 
513 	case DIOCDWEDGE:
514 	    {
515 	    	struct dkwedge_info *dkw = (void *) addr;
516 
517 		if ((flag & FWRITE) == 0)
518 			return (EBADF);
519 
520 		/* If the ioctl happens here, the parent is us. */
521 		strcpy(dkw->dkw_parent, sc->sc_dv.dv_xname);
522 		return (dkwedge_del(dkw));
523 	    }
524 
525 	case DIOCLWEDGES:
526 	    {
527 	    	struct dkwedge_list *dkwl = (void *) addr;
528 
529 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
530 	    }
531 	case DIOCGSTRATEGY:
532 	    {
533 		struct disk_strategy *dks = (void *)addr;
534 
535 		mutex_enter(&sc->sc_mutex);
536 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
537 		    sizeof(dks->dks_name));
538 		mutex_exit(&sc->sc_mutex);
539 		dks->dks_paramlen = 0;
540 
541 		return 0;
542 	    }
543 	case DIOCSSTRATEGY:
544 	    {
545 		struct disk_strategy *dks = (void *)addr;
546 		struct bufq_state *new, *old;
547 
548 		if ((flag & FWRITE) == 0)
549 			return EPERM;
550 
551 		if (dks->dks_param != NULL)
552 			return EINVAL;
553 
554 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
555 		error = bufq_alloc(&new, dks->dks_name,
556 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
557 		if (error)
558 			return error;
559 
560 		mutex_enter(&sc->sc_mutex);
561 		old = sc->sc_bufq;
562 		bufq_move(new, old);
563 		sc->sc_bufq = new;
564 		mutex_exit(&sc->sc_mutex);
565 		bufq_free(old);
566 
567 		return 0;
568 	    }
569 	default:
570 		error = ENOTTY;
571 		break;
572 	}
573 
574 	return (error);
575 }
576 
577 static void
578 ldstrategy(struct buf *bp)
579 {
580 	struct ld_softc *sc;
581 	struct disklabel *lp;
582 	daddr_t blkno;
583 	int s, part;
584 
585 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
586 	part = DISKPART(bp->b_dev);
587 
588 	if ((sc->sc_flags & LDF_DETACH) != 0) {
589 		bp->b_error = EIO;
590 		goto done;
591 	}
592 
593 	lp = sc->sc_dk.dk_label;
594 
595 	/*
596 	 * The transfer must be a whole number of blocks and the offset must
597 	 * not be negative.
598 	 */
599 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
600 		bp->b_error = EINVAL;
601 		goto done;
602 	}
603 
604 	/* If it's a null transfer, return immediately. */
605 	if (bp->b_bcount == 0)
606 		goto done;
607 
608 	/*
609 	 * Do bounds checking and adjust the transfer.  If error, process.
610 	 * If past the end of partition, just return.
611 	 */
612 	if (part != RAW_PART &&
613 	    bounds_check_with_label(&sc->sc_dk, bp,
614 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
615 		goto done;
616 	}
617 
618 	/*
619 	 * Convert the block number to absolute and put it in terms
620 	 * of the device's logical block size.
621 	 */
622 	if (lp->d_secsize == DEV_BSIZE)
623 		blkno = bp->b_blkno;
624 	else if (lp->d_secsize > DEV_BSIZE)
625 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
626 	else
627 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
628 
629 	if (part != RAW_PART)
630 		blkno += lp->d_partitions[part].p_offset;
631 
632 	bp->b_rawblkno = blkno;
633 
634 	s = splbio();
635 	ldstart(sc, bp);
636 	splx(s);
637 	return;
638 
639  done:
640 	bp->b_resid = bp->b_bcount;
641 	biodone(bp);
642 }
643 
644 static void
645 ldstart(struct ld_softc *sc, struct buf *bp)
646 {
647 	int error;
648 
649 	mutex_enter(&sc->sc_mutex);
650 
651 	if (bp != NULL)
652 		BUFQ_PUT(sc->sc_bufq, bp);
653 
654 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
655 		/* See if there is work to do. */
656 		if ((bp = BUFQ_PEEK(sc->sc_bufq)) == NULL)
657 			break;
658 
659 		disk_busy(&sc->sc_dk);
660 		sc->sc_queuecnt++;
661 
662 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
663 			/*
664 			 * The back-end is running the job; remove it from
665 			 * the queue.
666 			 */
667 			(void) BUFQ_GET(sc->sc_bufq);
668 		} else  {
669 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
670 			sc->sc_queuecnt--;
671 			if (error == EAGAIN) {
672 				/*
673 				 * Temporary resource shortage in the
674 				 * back-end; just defer the job until
675 				 * later.
676 				 *
677 				 * XXX We might consider a watchdog timer
678 				 * XXX to make sure we are kicked into action.
679 				 */
680 				break;
681 			} else {
682 				(void) BUFQ_GET(sc->sc_bufq);
683 				bp->b_error = error;
684 				bp->b_resid = bp->b_bcount;
685 				mutex_exit(&sc->sc_mutex);
686 				biodone(bp);
687 				mutex_enter(&sc->sc_mutex);
688 			}
689 		}
690 	}
691 
692 	mutex_exit(&sc->sc_mutex);
693 }
694 
695 void
696 lddone(struct ld_softc *sc, struct buf *bp)
697 {
698 
699 	if (bp->b_error != 0) {
700 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
701 		printf("\n");
702 	}
703 
704 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
705 	    (bp->b_flags & B_READ));
706 #if NRND > 0
707 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
708 #endif
709 	biodone(bp);
710 
711 	mutex_enter(&sc->sc_mutex);
712 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
713 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
714 			sc->sc_flags &= ~LDF_DRAIN;
715 			wakeup(&sc->sc_queuecnt);
716 		}
717 		mutex_exit(&sc->sc_mutex);
718 		ldstart(sc, NULL);
719 	} else
720 		mutex_exit(&sc->sc_mutex);
721 }
722 
723 static int
724 ldsize(dev_t dev)
725 {
726 	struct ld_softc *sc;
727 	int part, unit, omask, size;
728 
729 	unit = DISKUNIT(dev);
730 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
731 		return (ENODEV);
732 	if ((sc->sc_flags & LDF_ENABLED) == 0)
733 		return (ENODEV);
734 	part = DISKPART(dev);
735 
736 	omask = sc->sc_dk.dk_openmask & (1 << part);
737 
738 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
739 		return (-1);
740 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
741 		size = -1;
742 	else
743 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
744 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
745 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
746 		return (-1);
747 
748 	return (size);
749 }
750 
751 /*
752  * Load the label information from the specified device.
753  */
754 static void
755 ldgetdisklabel(struct ld_softc *sc)
756 {
757 	const char *errstring;
758 
759 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
760 
761 	/* Call the generic disklabel extraction routine. */
762 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(&sc->sc_dv),
763 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
764 	if (errstring != NULL)
765 		printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
766 
767 	/* In-core label now valid. */
768 	sc->sc_flags |= LDF_VLABEL;
769 }
770 
771 /*
772  * Construct a ficticious label.
773  */
774 static void
775 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
776 {
777 
778 	memset(lp, 0, sizeof(struct disklabel));
779 
780 	lp->d_secsize = sc->sc_secsize;
781 	lp->d_ntracks = sc->sc_nheads;
782 	lp->d_nsectors = sc->sc_nsectors;
783 	lp->d_ncylinders = sc->sc_ncylinders;
784 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
785 	lp->d_type = DTYPE_LD;
786 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
787 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
788 	lp->d_secperunit = sc->sc_secperunit;
789 	lp->d_rpm = 7200;
790 	lp->d_interleave = 1;
791 	lp->d_flags = 0;
792 
793 	lp->d_partitions[RAW_PART].p_offset = 0;
794 	lp->d_partitions[RAW_PART].p_size =
795 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
796 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
797 	lp->d_npartitions = RAW_PART + 1;
798 
799 	lp->d_magic = DISKMAGIC;
800 	lp->d_magic2 = DISKMAGIC;
801 	lp->d_checksum = dkcksum(lp);
802 }
803 
804 /*
805  * Take a dump.
806  */
807 static int
808 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
809 {
810 	char *va = vav;
811 	struct ld_softc *sc;
812 	struct disklabel *lp;
813 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
814 	static int dumping;
815 
816 	unit = DISKUNIT(dev);
817 	if ((sc = device_lookup(&ld_cd, unit)) == NULL)
818 		return (ENXIO);
819 	if ((sc->sc_flags & LDF_ENABLED) == 0)
820 		return (ENODEV);
821 	if (sc->sc_dump == NULL)
822 		return (ENXIO);
823 
824 	/* Check if recursive dump; if so, punt. */
825 	if (dumping)
826 		return (EFAULT);
827 	dumping = 1;
828 
829 	/* Convert to disk sectors.  Request must be a multiple of size. */
830 	part = DISKPART(dev);
831 	lp = sc->sc_dk.dk_label;
832 	if ((size % lp->d_secsize) != 0)
833 		return (EFAULT);
834 	towrt = size / lp->d_secsize;
835 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
836 
837 	nsects = lp->d_partitions[part].p_size;
838 	sectoff = lp->d_partitions[part].p_offset;
839 
840 	/* Check transfer bounds against partition size. */
841 	if ((blkno < 0) || ((blkno + towrt) > nsects))
842 		return (EINVAL);
843 
844 	/* Offset block number to start of partition. */
845 	blkno += sectoff;
846 
847 	/* Start dumping and return when done. */
848 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
849 	while (towrt > 0) {
850 		nblk = min(maxblkcnt, towrt);
851 
852 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
853 			return (rv);
854 
855 		towrt -= nblk;
856 		blkno += nblk;
857 		va += nblk * sc->sc_secsize;
858 	}
859 
860 	dumping = 0;
861 	return (0);
862 }
863 
864 /*
865  * Adjust the size of a transfer.
866  */
867 static void
868 ldminphys(struct buf *bp)
869 {
870 	struct ld_softc *sc;
871 
872 	sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
873 
874 	if (bp->b_bcount > sc->sc_maxxfer)
875 		bp->b_bcount = sc->sc_maxxfer;
876 	minphys(bp);
877 }
878 
879 static void
880 ld_set_properties(struct ld_softc *ld)
881 {
882 	prop_dictionary_t disk_info, odisk_info, geom;
883 
884 	disk_info = prop_dictionary_create();
885 
886 	geom = prop_dictionary_create();
887 
888 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
889 	    ld->sc_secperunit);
890 
891 	prop_dictionary_set_uint32(geom, "sector-size",
892 	    ld->sc_secsize);
893 
894 	prop_dictionary_set_uint16(geom, "sectors-per-track",
895 	    ld->sc_nsectors);
896 
897 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
898 	    ld->sc_nheads);
899 
900 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
901 	    ld->sc_ncylinders);
902 
903 	prop_dictionary_set(disk_info, "geometry", geom);
904 	prop_object_release(geom);
905 
906 	prop_dictionary_set(device_properties(&ld->sc_dv),
907 	    "disk-info", disk_info);
908 
909 	/*
910 	 * Don't release disk_info here; we keep a reference to it.
911 	 * disk_detach() will release it when we go away.
912 	 */
913 
914 	odisk_info = ld->sc_dk.dk_info;
915 	ld->sc_dk.dk_info = disk_info;
916 	if (odisk_info)
917 		prop_object_release(odisk_info);
918 }
919 
920 static void
921 ld_config_interrupts (struct device *d)
922 {
923 	struct ld_softc *sc = (struct ld_softc *)d;
924 	dkwedge_discover(&sc->sc_dk);
925 }
926