xref: /netbsd-src/sys/dev/ld.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: ld.c,v 1.66 2009/07/23 21:38:33 dyoung 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Disk driver for use by RAID controllers.
34  */
35 
36 #include <sys/cdefs.h>
37 __KERNEL_RCSID(0, "$NetBSD: ld.c,v 1.66 2009/07/23 21:38:33 dyoung Exp $");
38 
39 #include "rnd.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/device.h>
45 #include <sys/queue.h>
46 #include <sys/proc.h>
47 #include <sys/buf.h>
48 #include <sys/bufq.h>
49 #include <sys/endian.h>
50 #include <sys/disklabel.h>
51 #include <sys/disk.h>
52 #include <sys/dkio.h>
53 #include <sys/stat.h>
54 #include <sys/conf.h>
55 #include <sys/fcntl.h>
56 #include <sys/vnode.h>
57 #include <sys/syslog.h>
58 #include <sys/mutex.h>
59 #if NRND > 0
60 #include <sys/rnd.h>
61 #endif
62 
63 #include <dev/ldvar.h>
64 
65 #include <prop/proplib.h>
66 
67 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
68 static void	ldgetdisklabel(struct ld_softc *);
69 static void	ldminphys(struct buf *bp);
70 static bool	ld_shutdown(device_t, int);
71 static void	ldstart(struct ld_softc *, struct buf *);
72 static void	ld_set_properties(struct ld_softc *);
73 static void	ld_config_interrupts (device_t);
74 static int	ldlastclose(device_t);
75 
76 extern struct	cfdriver ld_cd;
77 
78 static dev_type_open(ldopen);
79 static dev_type_close(ldclose);
80 static dev_type_read(ldread);
81 static dev_type_write(ldwrite);
82 static dev_type_ioctl(ldioctl);
83 static dev_type_strategy(ldstrategy);
84 static dev_type_dump(lddump);
85 static dev_type_size(ldsize);
86 
87 const struct bdevsw ld_bdevsw = {
88 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
89 };
90 
91 const struct cdevsw ld_cdevsw = {
92 	ldopen, ldclose, ldread, ldwrite, ldioctl,
93 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
94 };
95 
96 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
97 
98 void
99 ldattach(struct ld_softc *sc)
100 {
101 	char tbuf[9];
102 
103 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
104 
105 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
106 		aprint_normal_dev(sc->sc_dv, "disabled\n");
107 		return;
108 	}
109 
110 	/* Initialise and attach the disk structure. */
111 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
112 	disk_attach(&sc->sc_dk);
113 
114 	if (sc->sc_maxxfer > MAXPHYS)
115 		sc->sc_maxxfer = MAXPHYS;
116 
117 	/* Build synthetic geometry if necessary. */
118 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
119 	    sc->sc_ncylinders == 0) {
120 		uint64_t ncyl;
121 
122 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
123 			sc->sc_nheads = 16;
124 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
125 			sc->sc_nheads = 32;
126 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
127 			sc->sc_nheads = 64;
128 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
129 			sc->sc_nheads = 128;
130 		else
131 			sc->sc_nheads = 255;
132 
133 		sc->sc_nsectors = 63;
134 		sc->sc_ncylinders = INT_MAX;
135 		ncyl = sc->sc_secperunit /
136 		    (sc->sc_nheads * sc->sc_nsectors);
137 		if (ncyl < INT_MAX)
138 			sc->sc_ncylinders = (int)ncyl;
139 	}
140 
141 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
142 	    sc->sc_secsize);
143 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
144 	    "%d bytes/sect x %"PRIu64" sectors\n",
145 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
146 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
147 
148 	ld_set_properties(sc);
149 
150 #if NRND > 0
151 	/* Attach the device into the rnd source list. */
152 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
153 	    RND_TYPE_DISK, 0);
154 #endif
155 
156 	/* Register with PMF */
157 	if (!pmf_device_register1(sc->sc_dv, NULL, NULL, ld_shutdown))
158 		aprint_error_dev(sc->sc_dv,
159 		    "couldn't establish power handler\n");
160 
161 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
162 
163 	/* Discover wedges on this disk. */
164 	config_interrupts(sc->sc_dv, ld_config_interrupts);
165 }
166 
167 int
168 ldadjqparam(struct ld_softc *sc, int xmax)
169 {
170 	int s;
171 
172 	s = splbio();
173 	sc->sc_maxqueuecnt = xmax;
174 	splx(s);
175 
176 	return (0);
177 }
178 
179 int
180 ldbegindetach(struct ld_softc *sc, int flags)
181 {
182 	int s, rv = 0;
183 
184 	if ((sc->sc_flags & LDF_ENABLED) == 0)
185 		return (0);
186 
187 	rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
188 
189 	if (rv != 0)
190 		return rv;
191 
192 	s = splbio();
193 	sc->sc_maxqueuecnt = 0;
194 	sc->sc_flags |= LDF_DETACH;
195 	while (sc->sc_queuecnt > 0) {
196 		sc->sc_flags |= LDF_DRAIN;
197 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
198 		if (rv)
199 			break;
200 	}
201 	splx(s);
202 
203 	return (rv);
204 }
205 
206 void
207 ldenddetach(struct ld_softc *sc)
208 {
209 	int s, bmaj, cmaj, i, mn;
210 
211 	if ((sc->sc_flags & LDF_ENABLED) == 0)
212 		return;
213 
214 	/* Wait for commands queued with the hardware to complete. */
215 	if (sc->sc_queuecnt != 0)
216 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
217 			printf("%s: not drained\n", device_xname(sc->sc_dv));
218 
219 	/* Locate the major numbers. */
220 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
221 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
222 
223 	/* Kill off any queued buffers. */
224 	s = splbio();
225 	bufq_drain(sc->sc_bufq);
226 	splx(s);
227 
228 	bufq_free(sc->sc_bufq);
229 
230 	/* Nuke the vnodes for any open instances. */
231 	for (i = 0; i < MAXPARTITIONS; i++) {
232 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
233 		vdevgone(bmaj, mn, mn, VBLK);
234 		vdevgone(cmaj, mn, mn, VCHR);
235 	}
236 
237 	/* Delete all of our wedges. */
238 	dkwedge_delall(&sc->sc_dk);
239 
240 	/* Detach from the disk list. */
241 	disk_detach(&sc->sc_dk);
242 	disk_destroy(&sc->sc_dk);
243 
244 #if NRND > 0
245 	/* Unhook the entropy source. */
246 	rnd_detach_source(&sc->sc_rnd_source);
247 #endif
248 
249 	/* Deregister with PMF */
250 	pmf_device_deregister(sc->sc_dv);
251 
252 	/*
253 	 * XXX We can't really flush the cache here, beceause the
254 	 * XXX device may already be non-existent from the controller's
255 	 * XXX perspective.
256 	 */
257 #if 0
258 	/* Flush the device's cache. */
259 	if (sc->sc_flush != NULL)
260 		if ((*sc->sc_flush)(sc, 0) != 0)
261 			aprint_error_dev(&sc->sc_dv, "unable to flush cache\n");
262 #endif
263 	mutex_destroy(&sc->sc_mutex);
264 }
265 
266 /* ARGSUSED */
267 static bool
268 ld_shutdown(device_t dev, int flags)
269 {
270 	struct ld_softc *sc = device_private(dev);
271 
272 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
273 		printf("%s: unable to flush cache\n", device_xname(dev));
274 		return false;
275 	}
276 
277 	return true;
278 }
279 
280 /* ARGSUSED */
281 static int
282 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
283 {
284 	struct ld_softc *sc;
285 	int error, unit, part;
286 
287 	unit = DISKUNIT(dev);
288 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
289 		return (ENXIO);
290 	if ((sc->sc_flags & LDF_ENABLED) == 0)
291 		return (ENODEV);
292 	part = DISKPART(dev);
293 
294 	mutex_enter(&sc->sc_dk.dk_openlock);
295 
296 	if (sc->sc_dk.dk_openmask == 0) {
297 		/* Load the partition info if not already loaded. */
298 		if ((sc->sc_flags & LDF_VLABEL) == 0)
299 			ldgetdisklabel(sc);
300 	}
301 
302 	/* Check that the partition exists. */
303 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
304 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
305 		error = ENXIO;
306 		goto bad1;
307 	}
308 
309 	/* Ensure only one open at a time. */
310 	switch (fmt) {
311 	case S_IFCHR:
312 		sc->sc_dk.dk_copenmask |= (1 << part);
313 		break;
314 	case S_IFBLK:
315 		sc->sc_dk.dk_bopenmask |= (1 << part);
316 		break;
317 	}
318 	sc->sc_dk.dk_openmask =
319 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
320 
321 	error = 0;
322  bad1:
323 	mutex_exit(&sc->sc_dk.dk_openlock);
324 	return (error);
325 }
326 
327 static int
328 ldlastclose(device_t self)
329 {
330 	struct ld_softc *sc = device_private(self);
331 
332 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
333 		aprint_error_dev(self, "unable to flush cache\n");
334 	if ((sc->sc_flags & LDF_KLABEL) == 0)
335 		sc->sc_flags &= ~LDF_VLABEL;
336 
337 	return 0;
338 }
339 
340 /* ARGSUSED */
341 static int
342 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
343 {
344 	struct ld_softc *sc;
345 	int part, unit;
346 
347 	unit = DISKUNIT(dev);
348 	part = DISKPART(dev);
349 	sc = device_lookup_private(&ld_cd, unit);
350 
351 	mutex_enter(&sc->sc_dk.dk_openlock);
352 
353 	switch (fmt) {
354 	case S_IFCHR:
355 		sc->sc_dk.dk_copenmask &= ~(1 << part);
356 		break;
357 	case S_IFBLK:
358 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
359 		break;
360 	}
361 	sc->sc_dk.dk_openmask =
362 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
363 
364 	if (sc->sc_dk.dk_openmask == 0)
365 		ldlastclose(sc->sc_dv);
366 
367 	mutex_exit(&sc->sc_dk.dk_openlock);
368 	return (0);
369 }
370 
371 /* ARGSUSED */
372 static int
373 ldread(dev_t dev, struct uio *uio, int ioflag)
374 {
375 
376 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
377 }
378 
379 /* ARGSUSED */
380 static int
381 ldwrite(dev_t dev, struct uio *uio, int ioflag)
382 {
383 
384 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
385 }
386 
387 /* ARGSUSED */
388 static int
389 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
390 {
391 	struct ld_softc *sc;
392 	int part, unit, error;
393 #ifdef __HAVE_OLD_DISKLABEL
394 	struct disklabel newlabel;
395 #endif
396 	struct disklabel *lp;
397 
398 	unit = DISKUNIT(dev);
399 	part = DISKPART(dev);
400 	sc = device_lookup_private(&ld_cd, unit);
401 
402 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
403 	if (error != EPASSTHROUGH)
404 		return (error);
405 
406 	error = 0;
407 	switch (cmd) {
408 	case DIOCGDINFO:
409 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
410 		return (0);
411 
412 #ifdef __HAVE_OLD_DISKLABEL
413 	case ODIOCGDINFO:
414 		newlabel = *(sc->sc_dk.dk_label);
415 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
416 			return ENOTTY;
417 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
418 		return (0);
419 #endif
420 
421 	case DIOCGPART:
422 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
423 		((struct partinfo *)addr)->part =
424 		    &sc->sc_dk.dk_label->d_partitions[part];
425 		break;
426 
427 	case DIOCWDINFO:
428 	case DIOCSDINFO:
429 #ifdef __HAVE_OLD_DISKLABEL
430 	case ODIOCWDINFO:
431 	case ODIOCSDINFO:
432 
433 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
434 			memset(&newlabel, 0, sizeof newlabel);
435 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
436 			lp = &newlabel;
437 		} else
438 #endif
439 		lp = (struct disklabel *)addr;
440 
441 		if ((flag & FWRITE) == 0)
442 			return (EBADF);
443 
444 		mutex_enter(&sc->sc_dk.dk_openlock);
445 		sc->sc_flags |= LDF_LABELLING;
446 
447 		error = setdisklabel(sc->sc_dk.dk_label,
448 		    lp, /*sc->sc_dk.dk_openmask : */0,
449 		    sc->sc_dk.dk_cpulabel);
450 		if (error == 0 && (cmd == DIOCWDINFO
451 #ifdef __HAVE_OLD_DISKLABEL
452 		    || cmd == ODIOCWDINFO
453 #endif
454 		    ))
455 			error = writedisklabel(
456 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
457 			    ldstrategy, sc->sc_dk.dk_label,
458 			    sc->sc_dk.dk_cpulabel);
459 
460 		sc->sc_flags &= ~LDF_LABELLING;
461 		mutex_exit(&sc->sc_dk.dk_openlock);
462 		break;
463 
464 	case DIOCKLABEL:
465 		if ((flag & FWRITE) == 0)
466 			return (EBADF);
467 		if (*(int *)addr)
468 			sc->sc_flags |= LDF_KLABEL;
469 		else
470 			sc->sc_flags &= ~LDF_KLABEL;
471 		break;
472 
473 	case DIOCWLABEL:
474 		if ((flag & FWRITE) == 0)
475 			return (EBADF);
476 		if (*(int *)addr)
477 			sc->sc_flags |= LDF_WLABEL;
478 		else
479 			sc->sc_flags &= ~LDF_WLABEL;
480 		break;
481 
482 	case DIOCGDEFLABEL:
483 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
484 		break;
485 
486 #ifdef __HAVE_OLD_DISKLABEL
487 	case ODIOCGDEFLABEL:
488 		ldgetdefaultlabel(sc, &newlabel);
489 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
490 			return ENOTTY;
491 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
492 		break;
493 #endif
494 
495 	case DIOCCACHESYNC:
496 		/*
497 		 * XXX Do we really need to care about having a writable
498 		 * file descriptor here?
499 		 */
500 		if ((flag & FWRITE) == 0)
501 			error = EBADF;
502 		else if (sc->sc_flush)
503 			error = (*sc->sc_flush)(sc, 0);
504 		else
505 			error = 0;	/* XXX Error out instead? */
506 		break;
507 
508 	case DIOCAWEDGE:
509 	    {
510 	    	struct dkwedge_info *dkw = (void *) addr;
511 
512 		if ((flag & FWRITE) == 0)
513 			return (EBADF);
514 
515 		/* If the ioctl happens here, the parent is us. */
516 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
517 			sizeof(dkw->dkw_parent));
518 		return (dkwedge_add(dkw));
519 	    }
520 
521 	case DIOCDWEDGE:
522 	    {
523 	    	struct dkwedge_info *dkw = (void *) addr;
524 
525 		if ((flag & FWRITE) == 0)
526 			return (EBADF);
527 
528 		/* If the ioctl happens here, the parent is us. */
529 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
530 			sizeof(dkw->dkw_parent));
531 		return (dkwedge_del(dkw));
532 	    }
533 
534 	case DIOCLWEDGES:
535 	    {
536 	    	struct dkwedge_list *dkwl = (void *) addr;
537 
538 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
539 	    }
540 	case DIOCGSTRATEGY:
541 	    {
542 		struct disk_strategy *dks = (void *)addr;
543 
544 		mutex_enter(&sc->sc_mutex);
545 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
546 		    sizeof(dks->dks_name));
547 		mutex_exit(&sc->sc_mutex);
548 		dks->dks_paramlen = 0;
549 
550 		return 0;
551 	    }
552 	case DIOCSSTRATEGY:
553 	    {
554 		struct disk_strategy *dks = (void *)addr;
555 		struct bufq_state *new, *old;
556 
557 		if ((flag & FWRITE) == 0)
558 			return EPERM;
559 
560 		if (dks->dks_param != NULL)
561 			return EINVAL;
562 
563 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
564 		error = bufq_alloc(&new, dks->dks_name,
565 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
566 		if (error)
567 			return error;
568 
569 		mutex_enter(&sc->sc_mutex);
570 		old = sc->sc_bufq;
571 		bufq_move(new, old);
572 		sc->sc_bufq = new;
573 		mutex_exit(&sc->sc_mutex);
574 		bufq_free(old);
575 
576 		return 0;
577 	    }
578 	default:
579 		error = ENOTTY;
580 		break;
581 	}
582 
583 	return (error);
584 }
585 
586 static void
587 ldstrategy(struct buf *bp)
588 {
589 	struct ld_softc *sc;
590 	struct disklabel *lp;
591 	daddr_t blkno;
592 	int s, part;
593 
594 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
595 	part = DISKPART(bp->b_dev);
596 
597 	if ((sc->sc_flags & LDF_DETACH) != 0) {
598 		bp->b_error = EIO;
599 		goto done;
600 	}
601 
602 	lp = sc->sc_dk.dk_label;
603 
604 	/*
605 	 * The transfer must be a whole number of blocks and the offset must
606 	 * not be negative.
607 	 */
608 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
609 		bp->b_error = EINVAL;
610 		goto done;
611 	}
612 
613 	/* If it's a null transfer, return immediately. */
614 	if (bp->b_bcount == 0)
615 		goto done;
616 
617 	/*
618 	 * Do bounds checking and adjust the transfer.  If error, process.
619 	 * If past the end of partition, just return.
620 	 */
621 	if (part != RAW_PART &&
622 	    bounds_check_with_label(&sc->sc_dk, bp,
623 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
624 		goto done;
625 	}
626 
627 	/*
628 	 * Convert the block number to absolute and put it in terms
629 	 * of the device's logical block size.
630 	 */
631 	if (lp->d_secsize == DEV_BSIZE)
632 		blkno = bp->b_blkno;
633 	else if (lp->d_secsize > DEV_BSIZE)
634 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
635 	else
636 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
637 
638 	if (part != RAW_PART)
639 		blkno += lp->d_partitions[part].p_offset;
640 
641 	bp->b_rawblkno = blkno;
642 
643 	s = splbio();
644 	ldstart(sc, bp);
645 	splx(s);
646 	return;
647 
648  done:
649 	bp->b_resid = bp->b_bcount;
650 	biodone(bp);
651 }
652 
653 static void
654 ldstart(struct ld_softc *sc, struct buf *bp)
655 {
656 	int error;
657 
658 	mutex_enter(&sc->sc_mutex);
659 
660 	if (bp != NULL)
661 		bufq_put(sc->sc_bufq, bp);
662 
663 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
664 		/* See if there is work to do. */
665 		if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
666 			break;
667 
668 		disk_busy(&sc->sc_dk);
669 		sc->sc_queuecnt++;
670 
671 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
672 			/*
673 			 * The back-end is running the job; remove it from
674 			 * the queue.
675 			 */
676 			(void) bufq_get(sc->sc_bufq);
677 		} else  {
678 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
679 			sc->sc_queuecnt--;
680 			if (error == EAGAIN) {
681 				/*
682 				 * Temporary resource shortage in the
683 				 * back-end; just defer the job until
684 				 * later.
685 				 *
686 				 * XXX We might consider a watchdog timer
687 				 * XXX to make sure we are kicked into action.
688 				 */
689 				break;
690 			} else {
691 				(void) bufq_get(sc->sc_bufq);
692 				bp->b_error = error;
693 				bp->b_resid = bp->b_bcount;
694 				mutex_exit(&sc->sc_mutex);
695 				biodone(bp);
696 				mutex_enter(&sc->sc_mutex);
697 			}
698 		}
699 	}
700 
701 	mutex_exit(&sc->sc_mutex);
702 }
703 
704 void
705 lddone(struct ld_softc *sc, struct buf *bp)
706 {
707 
708 	if (bp->b_error != 0) {
709 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
710 		printf("\n");
711 	}
712 
713 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
714 	    (bp->b_flags & B_READ));
715 #if NRND > 0
716 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
717 #endif
718 	biodone(bp);
719 
720 	mutex_enter(&sc->sc_mutex);
721 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
722 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
723 			sc->sc_flags &= ~LDF_DRAIN;
724 			wakeup(&sc->sc_queuecnt);
725 		}
726 		mutex_exit(&sc->sc_mutex);
727 		ldstart(sc, NULL);
728 	} else
729 		mutex_exit(&sc->sc_mutex);
730 }
731 
732 static int
733 ldsize(dev_t dev)
734 {
735 	struct ld_softc *sc;
736 	int part, unit, omask, size;
737 
738 	unit = DISKUNIT(dev);
739 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
740 		return (ENODEV);
741 	if ((sc->sc_flags & LDF_ENABLED) == 0)
742 		return (ENODEV);
743 	part = DISKPART(dev);
744 
745 	omask = sc->sc_dk.dk_openmask & (1 << part);
746 
747 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
748 		return (-1);
749 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
750 		size = -1;
751 	else
752 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
753 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
754 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
755 		return (-1);
756 
757 	return (size);
758 }
759 
760 /*
761  * Load the label information from the specified device.
762  */
763 static void
764 ldgetdisklabel(struct ld_softc *sc)
765 {
766 	const char *errstring;
767 
768 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
769 
770 	/* Call the generic disklabel extraction routine. */
771 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
772 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
773 	if (errstring != NULL)
774 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
775 
776 	/* In-core label now valid. */
777 	sc->sc_flags |= LDF_VLABEL;
778 }
779 
780 /*
781  * Construct a ficticious label.
782  */
783 static void
784 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
785 {
786 
787 	memset(lp, 0, sizeof(struct disklabel));
788 
789 	lp->d_secsize = sc->sc_secsize;
790 	lp->d_ntracks = sc->sc_nheads;
791 	lp->d_nsectors = sc->sc_nsectors;
792 	lp->d_ncylinders = sc->sc_ncylinders;
793 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
794 	lp->d_type = DTYPE_LD;
795 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
796 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
797 	lp->d_secperunit = sc->sc_secperunit;
798 	lp->d_rpm = 7200;
799 	lp->d_interleave = 1;
800 	lp->d_flags = 0;
801 
802 	lp->d_partitions[RAW_PART].p_offset = 0;
803 	lp->d_partitions[RAW_PART].p_size =
804 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
805 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
806 	lp->d_npartitions = RAW_PART + 1;
807 
808 	lp->d_magic = DISKMAGIC;
809 	lp->d_magic2 = DISKMAGIC;
810 	lp->d_checksum = dkcksum(lp);
811 }
812 
813 /*
814  * Take a dump.
815  */
816 static int
817 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
818 {
819 	char *va = vav;
820 	struct ld_softc *sc;
821 	struct disklabel *lp;
822 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
823 	static int dumping;
824 
825 	unit = DISKUNIT(dev);
826 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
827 		return (ENXIO);
828 	if ((sc->sc_flags & LDF_ENABLED) == 0)
829 		return (ENODEV);
830 	if (sc->sc_dump == NULL)
831 		return (ENXIO);
832 
833 	/* Check if recursive dump; if so, punt. */
834 	if (dumping)
835 		return (EFAULT);
836 	dumping = 1;
837 
838 	/* Convert to disk sectors.  Request must be a multiple of size. */
839 	part = DISKPART(dev);
840 	lp = sc->sc_dk.dk_label;
841 	if ((size % lp->d_secsize) != 0)
842 		return (EFAULT);
843 	towrt = size / lp->d_secsize;
844 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
845 
846 	nsects = lp->d_partitions[part].p_size;
847 	sectoff = lp->d_partitions[part].p_offset;
848 
849 	/* Check transfer bounds against partition size. */
850 	if ((blkno < 0) || ((blkno + towrt) > nsects))
851 		return (EINVAL);
852 
853 	/* Offset block number to start of partition. */
854 	blkno += sectoff;
855 
856 	/* Start dumping and return when done. */
857 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
858 	while (towrt > 0) {
859 		nblk = min(maxblkcnt, towrt);
860 
861 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
862 			return (rv);
863 
864 		towrt -= nblk;
865 		blkno += nblk;
866 		va += nblk * sc->sc_secsize;
867 	}
868 
869 	dumping = 0;
870 	return (0);
871 }
872 
873 /*
874  * Adjust the size of a transfer.
875  */
876 static void
877 ldminphys(struct buf *bp)
878 {
879 	struct ld_softc *sc;
880 
881 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
882 
883 	if (bp->b_bcount > sc->sc_maxxfer)
884 		bp->b_bcount = sc->sc_maxxfer;
885 	minphys(bp);
886 }
887 
888 static void
889 ld_set_properties(struct ld_softc *ld)
890 {
891 	prop_dictionary_t disk_info, odisk_info, geom;
892 
893 	disk_info = prop_dictionary_create();
894 
895 	geom = prop_dictionary_create();
896 
897 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
898 	    ld->sc_secperunit);
899 
900 	prop_dictionary_set_uint32(geom, "sector-size",
901 	    ld->sc_secsize);
902 
903 	prop_dictionary_set_uint16(geom, "sectors-per-track",
904 	    ld->sc_nsectors);
905 
906 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
907 	    ld->sc_nheads);
908 
909 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
910 	    ld->sc_ncylinders);
911 
912 	prop_dictionary_set(disk_info, "geometry", geom);
913 	prop_object_release(geom);
914 
915 	prop_dictionary_set(device_properties(ld->sc_dv),
916 	    "disk-info", disk_info);
917 
918 	/*
919 	 * Don't release disk_info here; we keep a reference to it.
920 	 * disk_detach() will release it when we go away.
921 	 */
922 
923 	odisk_info = ld->sc_dk.dk_info;
924 	ld->sc_dk.dk_info = disk_info;
925 	if (odisk_info)
926 		prop_object_release(odisk_info);
927 }
928 
929 static void
930 ld_config_interrupts(device_t d)
931 {
932 	struct ld_softc *sc = device_private(d);
933 	dkwedge_discover(&sc->sc_dk);
934 }
935