xref: /netbsd-src/sys/dev/ld.c (revision ba65fde2d7fefa7d39838fa5fa855e62bd606b5e)
1 /*	$NetBSD: ld.c,v 1.70 2012/10/27 17:18:14 chs 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 1.70 2012/10/27 17:18:14 chs Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/device.h>
43 #include <sys/queue.h>
44 #include <sys/proc.h>
45 #include <sys/buf.h>
46 #include <sys/bufq.h>
47 #include <sys/endian.h>
48 #include <sys/disklabel.h>
49 #include <sys/disk.h>
50 #include <sys/dkio.h>
51 #include <sys/stat.h>
52 #include <sys/conf.h>
53 #include <sys/fcntl.h>
54 #include <sys/vnode.h>
55 #include <sys/syslog.h>
56 #include <sys/mutex.h>
57 #include <sys/rnd.h>
58 
59 #include <dev/ldvar.h>
60 
61 #include <prop/proplib.h>
62 
63 static void	ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
64 static void	ldgetdisklabel(struct ld_softc *);
65 static void	ldminphys(struct buf *bp);
66 static bool	ld_suspend(device_t, const pmf_qual_t *);
67 static bool	ld_shutdown(device_t, int);
68 static void	ldstart(struct ld_softc *, struct buf *);
69 static void	ld_set_properties(struct ld_softc *);
70 static void	ld_config_interrupts (device_t);
71 static int	ldlastclose(device_t);
72 
73 extern struct	cfdriver ld_cd;
74 
75 static dev_type_open(ldopen);
76 static dev_type_close(ldclose);
77 static dev_type_read(ldread);
78 static dev_type_write(ldwrite);
79 static dev_type_ioctl(ldioctl);
80 static dev_type_strategy(ldstrategy);
81 static dev_type_dump(lddump);
82 static dev_type_size(ldsize);
83 
84 const struct bdevsw ld_bdevsw = {
85 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
86 };
87 
88 const struct cdevsw ld_cdevsw = {
89 	ldopen, ldclose, ldread, ldwrite, ldioctl,
90 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
91 };
92 
93 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
94 
95 void
96 ldattach(struct ld_softc *sc)
97 {
98 	char tbuf[9];
99 
100 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
101 
102 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
103 		aprint_normal_dev(sc->sc_dv, "disabled\n");
104 		return;
105 	}
106 
107 	/* Initialise and attach the disk structure. */
108 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
109 	disk_attach(&sc->sc_dk);
110 
111 	if (sc->sc_maxxfer > MAXPHYS)
112 		sc->sc_maxxfer = MAXPHYS;
113 
114 	/* Build synthetic geometry if necessary. */
115 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
116 	    sc->sc_ncylinders == 0) {
117 		uint64_t ncyl;
118 
119 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
120 			sc->sc_nheads = 16;
121 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
122 			sc->sc_nheads = 32;
123 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
124 			sc->sc_nheads = 64;
125 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
126 			sc->sc_nheads = 128;
127 		else
128 			sc->sc_nheads = 255;
129 
130 		sc->sc_nsectors = 63;
131 		sc->sc_ncylinders = INT_MAX;
132 		ncyl = sc->sc_secperunit /
133 		    (sc->sc_nheads * sc->sc_nsectors);
134 		if (ncyl < INT_MAX)
135 			sc->sc_ncylinders = (int)ncyl;
136 	}
137 
138 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
139 	    sc->sc_secsize);
140 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
141 	    "%d bytes/sect x %"PRIu64" sectors\n",
142 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
143 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
144 	sc->sc_disksize512 = sc->sc_secperunit * sc->sc_secsize / DEV_BSIZE;
145 
146 	ld_set_properties(sc);
147 
148 	/* Attach the device into the rnd source list. */
149 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
150 	    RND_TYPE_DISK, 0);
151 
152 	/* Register with PMF */
153 	if (!pmf_device_register1(sc->sc_dv, ld_suspend, NULL, ld_shutdown))
154 		aprint_error_dev(sc->sc_dv,
155 		    "couldn't establish power handler\n");
156 
157 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
158 
159 	/* Discover wedges on this disk. */
160 	config_interrupts(sc->sc_dv, ld_config_interrupts);
161 }
162 
163 int
164 ldadjqparam(struct ld_softc *sc, int xmax)
165 {
166 	int s;
167 
168 	s = splbio();
169 	sc->sc_maxqueuecnt = xmax;
170 	splx(s);
171 
172 	return (0);
173 }
174 
175 int
176 ldbegindetach(struct ld_softc *sc, int flags)
177 {
178 	int s, rv = 0;
179 
180 	if ((sc->sc_flags & LDF_ENABLED) == 0)
181 		return (0);
182 
183 	rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
184 
185 	if (rv != 0)
186 		return rv;
187 
188 	s = splbio();
189 	sc->sc_maxqueuecnt = 0;
190 	sc->sc_flags |= LDF_DETACH;
191 	while (sc->sc_queuecnt > 0) {
192 		sc->sc_flags |= LDF_DRAIN;
193 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
194 		if (rv)
195 			break;
196 	}
197 	splx(s);
198 
199 	return (rv);
200 }
201 
202 void
203 ldenddetach(struct ld_softc *sc)
204 {
205 	int s, bmaj, cmaj, i, mn;
206 
207 	if ((sc->sc_flags & LDF_ENABLED) == 0)
208 		return;
209 
210 	/* Wait for commands queued with the hardware to complete. */
211 	if (sc->sc_queuecnt != 0)
212 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
213 			printf("%s: not drained\n", device_xname(sc->sc_dv));
214 
215 	/* Locate the major numbers. */
216 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
217 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
218 
219 	/* Kill off any queued buffers. */
220 	s = splbio();
221 	bufq_drain(sc->sc_bufq);
222 	splx(s);
223 
224 	bufq_free(sc->sc_bufq);
225 
226 	/* Nuke the vnodes for any open instances. */
227 	for (i = 0; i < MAXPARTITIONS; i++) {
228 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
229 		vdevgone(bmaj, mn, mn, VBLK);
230 		vdevgone(cmaj, mn, mn, VCHR);
231 	}
232 
233 	/* Delete all of our wedges. */
234 	dkwedge_delall(&sc->sc_dk);
235 
236 	/* Detach from the disk list. */
237 	disk_detach(&sc->sc_dk);
238 	disk_destroy(&sc->sc_dk);
239 
240 	/* Unhook the entropy source. */
241 	rnd_detach_source(&sc->sc_rnd_source);
242 
243 	/* Deregister with PMF */
244 	pmf_device_deregister(sc->sc_dv);
245 
246 	/*
247 	 * XXX We can't really flush the cache here, beceause the
248 	 * XXX device may already be non-existent from the controller's
249 	 * XXX perspective.
250 	 */
251 #if 0
252 	/* Flush the device's cache. */
253 	if (sc->sc_flush != NULL)
254 		if ((*sc->sc_flush)(sc, 0) != 0)
255 			aprint_error_dev(sc->sc_dv, "unable to flush cache\n");
256 #endif
257 	mutex_destroy(&sc->sc_mutex);
258 }
259 
260 /* ARGSUSED */
261 static bool
262 ld_suspend(device_t dev, const pmf_qual_t *qual)
263 {
264 	return ld_shutdown(dev, 0);
265 }
266 
267 /* ARGSUSED */
268 static bool
269 ld_shutdown(device_t dev, int flags)
270 {
271 	struct ld_softc *sc = device_private(dev);
272 
273 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
274 		printf("%s: unable to flush cache\n", device_xname(dev));
275 		return false;
276 	}
277 
278 	return true;
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_private(&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 static int
329 ldlastclose(device_t self)
330 {
331 	struct ld_softc *sc = device_private(self);
332 
333 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
334 		aprint_error_dev(self, "unable to flush cache\n");
335 	if ((sc->sc_flags & LDF_KLABEL) == 0)
336 		sc->sc_flags &= ~LDF_VLABEL;
337 
338 	return 0;
339 }
340 
341 /* ARGSUSED */
342 static int
343 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
344 {
345 	struct ld_softc *sc;
346 	int part, unit;
347 
348 	unit = DISKUNIT(dev);
349 	part = DISKPART(dev);
350 	sc = device_lookup_private(&ld_cd, unit);
351 
352 	mutex_enter(&sc->sc_dk.dk_openlock);
353 
354 	switch (fmt) {
355 	case S_IFCHR:
356 		sc->sc_dk.dk_copenmask &= ~(1 << part);
357 		break;
358 	case S_IFBLK:
359 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
360 		break;
361 	}
362 	sc->sc_dk.dk_openmask =
363 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
364 
365 	if (sc->sc_dk.dk_openmask == 0)
366 		ldlastclose(sc->sc_dv);
367 
368 	mutex_exit(&sc->sc_dk.dk_openlock);
369 	return (0);
370 }
371 
372 /* ARGSUSED */
373 static int
374 ldread(dev_t dev, struct uio *uio, int ioflag)
375 {
376 
377 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
378 }
379 
380 /* ARGSUSED */
381 static int
382 ldwrite(dev_t dev, struct uio *uio, int ioflag)
383 {
384 
385 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
386 }
387 
388 /* ARGSUSED */
389 static int
390 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
391 {
392 	struct ld_softc *sc;
393 	int part, unit, error;
394 #ifdef __HAVE_OLD_DISKLABEL
395 	struct disklabel newlabel;
396 #endif
397 	struct disklabel *lp;
398 
399 	unit = DISKUNIT(dev);
400 	part = DISKPART(dev);
401 	sc = device_lookup_private(&ld_cd, unit);
402 
403 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
404 	if (error != EPASSTHROUGH)
405 		return (error);
406 
407 	error = 0;
408 	switch (cmd) {
409 	case DIOCGDINFO:
410 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
411 		return (0);
412 
413 #ifdef __HAVE_OLD_DISKLABEL
414 	case ODIOCGDINFO:
415 		newlabel = *(sc->sc_dk.dk_label);
416 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
417 			return ENOTTY;
418 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
419 		return (0);
420 #endif
421 
422 	case DIOCGPART:
423 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
424 		((struct partinfo *)addr)->part =
425 		    &sc->sc_dk.dk_label->d_partitions[part];
426 		break;
427 
428 	case DIOCWDINFO:
429 	case DIOCSDINFO:
430 #ifdef __HAVE_OLD_DISKLABEL
431 	case ODIOCWDINFO:
432 	case ODIOCSDINFO:
433 
434 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
435 			memset(&newlabel, 0, sizeof newlabel);
436 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
437 			lp = &newlabel;
438 		} else
439 #endif
440 		lp = (struct disklabel *)addr;
441 
442 		if ((flag & FWRITE) == 0)
443 			return (EBADF);
444 
445 		mutex_enter(&sc->sc_dk.dk_openlock);
446 		sc->sc_flags |= LDF_LABELLING;
447 
448 		error = setdisklabel(sc->sc_dk.dk_label,
449 		    lp, /*sc->sc_dk.dk_openmask : */0,
450 		    sc->sc_dk.dk_cpulabel);
451 		if (error == 0 && (cmd == DIOCWDINFO
452 #ifdef __HAVE_OLD_DISKLABEL
453 		    || cmd == ODIOCWDINFO
454 #endif
455 		    ))
456 			error = writedisklabel(
457 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
458 			    ldstrategy, sc->sc_dk.dk_label,
459 			    sc->sc_dk.dk_cpulabel);
460 
461 		sc->sc_flags &= ~LDF_LABELLING;
462 		mutex_exit(&sc->sc_dk.dk_openlock);
463 		break;
464 
465 	case DIOCKLABEL:
466 		if ((flag & FWRITE) == 0)
467 			return (EBADF);
468 		if (*(int *)addr)
469 			sc->sc_flags |= LDF_KLABEL;
470 		else
471 			sc->sc_flags &= ~LDF_KLABEL;
472 		break;
473 
474 	case DIOCWLABEL:
475 		if ((flag & FWRITE) == 0)
476 			return (EBADF);
477 		if (*(int *)addr)
478 			sc->sc_flags |= LDF_WLABEL;
479 		else
480 			sc->sc_flags &= ~LDF_WLABEL;
481 		break;
482 
483 	case DIOCGDEFLABEL:
484 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
485 		break;
486 
487 #ifdef __HAVE_OLD_DISKLABEL
488 	case ODIOCGDEFLABEL:
489 		ldgetdefaultlabel(sc, &newlabel);
490 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
491 			return ENOTTY;
492 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
493 		break;
494 #endif
495 
496 	case DIOCCACHESYNC:
497 		/*
498 		 * XXX Do we really need to care about having a writable
499 		 * file descriptor here?
500 		 */
501 		if ((flag & FWRITE) == 0)
502 			error = EBADF;
503 		else if (sc->sc_flush)
504 			error = (*sc->sc_flush)(sc, 0);
505 		else
506 			error = 0;	/* XXX Error out instead? */
507 		break;
508 
509 	case DIOCAWEDGE:
510 	    {
511 	    	struct dkwedge_info *dkw = (void *) addr;
512 
513 		if ((flag & FWRITE) == 0)
514 			return (EBADF);
515 
516 		/* If the ioctl happens here, the parent is us. */
517 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
518 			sizeof(dkw->dkw_parent));
519 		return (dkwedge_add(dkw));
520 	    }
521 
522 	case DIOCDWEDGE:
523 	    {
524 	    	struct dkwedge_info *dkw = (void *) addr;
525 
526 		if ((flag & FWRITE) == 0)
527 			return (EBADF);
528 
529 		/* If the ioctl happens here, the parent is us. */
530 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
531 			sizeof(dkw->dkw_parent));
532 		return (dkwedge_del(dkw));
533 	    }
534 
535 	case DIOCLWEDGES:
536 	    {
537 	    	struct dkwedge_list *dkwl = (void *) addr;
538 
539 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
540 	    }
541 	case DIOCGSTRATEGY:
542 	    {
543 		struct disk_strategy *dks = (void *)addr;
544 
545 		mutex_enter(&sc->sc_mutex);
546 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
547 		    sizeof(dks->dks_name));
548 		mutex_exit(&sc->sc_mutex);
549 		dks->dks_paramlen = 0;
550 
551 		return 0;
552 	    }
553 	case DIOCSSTRATEGY:
554 	    {
555 		struct disk_strategy *dks = (void *)addr;
556 		struct bufq_state *new, *old;
557 
558 		if ((flag & FWRITE) == 0)
559 			return EPERM;
560 
561 		if (dks->dks_param != NULL)
562 			return EINVAL;
563 
564 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
565 		error = bufq_alloc(&new, dks->dks_name,
566 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
567 		if (error)
568 			return error;
569 
570 		mutex_enter(&sc->sc_mutex);
571 		old = sc->sc_bufq;
572 		bufq_move(new, old);
573 		sc->sc_bufq = new;
574 		mutex_exit(&sc->sc_mutex);
575 		bufq_free(old);
576 
577 		return 0;
578 	    }
579 	default:
580 		error = ENOTTY;
581 		break;
582 	}
583 
584 	return (error);
585 }
586 
587 static void
588 ldstrategy(struct buf *bp)
589 {
590 	struct ld_softc *sc;
591 	struct disklabel *lp;
592 	daddr_t blkno;
593 	int s, part;
594 
595 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
596 	part = DISKPART(bp->b_dev);
597 
598 	if ((sc->sc_flags & LDF_DETACH) != 0) {
599 		bp->b_error = EIO;
600 		goto done;
601 	}
602 
603 	lp = sc->sc_dk.dk_label;
604 
605 	/*
606 	 * The transfer must be a whole number of blocks and the offset must
607 	 * not be negative.
608 	 */
609 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
610 		bp->b_error = EINVAL;
611 		goto done;
612 	}
613 
614 	/* If it's a null transfer, return immediately. */
615 	if (bp->b_bcount == 0)
616 		goto done;
617 
618 	/*
619 	 * Do bounds checking and adjust the transfer.  If error, process.
620 	 * If past the end of partition, just return.
621 	 */
622 	if (part == RAW_PART) {
623 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
624 		    sc->sc_disksize512) <= 0)
625 			goto done;
626 	} else {
627 		if (bounds_check_with_label(&sc->sc_dk, bp,
628 		    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0)
629 			goto done;
630 	}
631 
632 	/*
633 	 * Convert the block number to absolute and put it in terms
634 	 * of the device's logical block size.
635 	 */
636 	if (lp->d_secsize == DEV_BSIZE)
637 		blkno = bp->b_blkno;
638 	else if (lp->d_secsize > DEV_BSIZE)
639 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
640 	else
641 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
642 
643 	if (part != RAW_PART)
644 		blkno += lp->d_partitions[part].p_offset;
645 
646 	bp->b_rawblkno = blkno;
647 
648 	s = splbio();
649 	ldstart(sc, bp);
650 	splx(s);
651 	return;
652 
653  done:
654 	bp->b_resid = bp->b_bcount;
655 	biodone(bp);
656 }
657 
658 static void
659 ldstart(struct ld_softc *sc, struct buf *bp)
660 {
661 	int error;
662 
663 	mutex_enter(&sc->sc_mutex);
664 
665 	if (bp != NULL)
666 		bufq_put(sc->sc_bufq, bp);
667 
668 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
669 		/* See if there is work to do. */
670 		if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
671 			break;
672 
673 		disk_busy(&sc->sc_dk);
674 		sc->sc_queuecnt++;
675 
676 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
677 			/*
678 			 * The back-end is running the job; remove it from
679 			 * the queue.
680 			 */
681 			(void) bufq_get(sc->sc_bufq);
682 		} else  {
683 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
684 			sc->sc_queuecnt--;
685 			if (error == EAGAIN) {
686 				/*
687 				 * Temporary resource shortage in the
688 				 * back-end; just defer the job until
689 				 * later.
690 				 *
691 				 * XXX We might consider a watchdog timer
692 				 * XXX to make sure we are kicked into action.
693 				 */
694 				break;
695 			} else {
696 				(void) bufq_get(sc->sc_bufq);
697 				bp->b_error = error;
698 				bp->b_resid = bp->b_bcount;
699 				mutex_exit(&sc->sc_mutex);
700 				biodone(bp);
701 				mutex_enter(&sc->sc_mutex);
702 			}
703 		}
704 	}
705 
706 	mutex_exit(&sc->sc_mutex);
707 }
708 
709 void
710 lddone(struct ld_softc *sc, struct buf *bp)
711 {
712 
713 	if (bp->b_error != 0) {
714 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
715 		printf("\n");
716 	}
717 
718 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
719 	    (bp->b_flags & B_READ));
720 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
721 	biodone(bp);
722 
723 	mutex_enter(&sc->sc_mutex);
724 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
725 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
726 			sc->sc_flags &= ~LDF_DRAIN;
727 			wakeup(&sc->sc_queuecnt);
728 		}
729 		mutex_exit(&sc->sc_mutex);
730 		ldstart(sc, NULL);
731 	} else
732 		mutex_exit(&sc->sc_mutex);
733 }
734 
735 static int
736 ldsize(dev_t dev)
737 {
738 	struct ld_softc *sc;
739 	int part, unit, omask, size;
740 
741 	unit = DISKUNIT(dev);
742 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
743 		return (ENODEV);
744 	if ((sc->sc_flags & LDF_ENABLED) == 0)
745 		return (ENODEV);
746 	part = DISKPART(dev);
747 
748 	omask = sc->sc_dk.dk_openmask & (1 << part);
749 
750 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
751 		return (-1);
752 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
753 		size = -1;
754 	else
755 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
756 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
757 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
758 		return (-1);
759 
760 	return (size);
761 }
762 
763 /*
764  * Load the label information from the specified device.
765  */
766 static void
767 ldgetdisklabel(struct ld_softc *sc)
768 {
769 	const char *errstring;
770 
771 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
772 
773 	/* Call the generic disklabel extraction routine. */
774 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
775 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
776 	if (errstring != NULL)
777 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
778 
779 	/* In-core label now valid. */
780 	sc->sc_flags |= LDF_VLABEL;
781 }
782 
783 /*
784  * Construct a ficticious label.
785  */
786 static void
787 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
788 {
789 
790 	memset(lp, 0, sizeof(struct disklabel));
791 
792 	lp->d_secsize = sc->sc_secsize;
793 	lp->d_ntracks = sc->sc_nheads;
794 	lp->d_nsectors = sc->sc_nsectors;
795 	lp->d_ncylinders = sc->sc_ncylinders;
796 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
797 	lp->d_type = DTYPE_LD;
798 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
799 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
800 	lp->d_secperunit = sc->sc_secperunit;
801 	lp->d_rpm = 7200;
802 	lp->d_interleave = 1;
803 	lp->d_flags = 0;
804 
805 	lp->d_partitions[RAW_PART].p_offset = 0;
806 	lp->d_partitions[RAW_PART].p_size =
807 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
808 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
809 	lp->d_npartitions = RAW_PART + 1;
810 
811 	lp->d_magic = DISKMAGIC;
812 	lp->d_magic2 = DISKMAGIC;
813 	lp->d_checksum = dkcksum(lp);
814 }
815 
816 /*
817  * Take a dump.
818  */
819 static int
820 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
821 {
822 	char *va = vav;
823 	struct ld_softc *sc;
824 	struct disklabel *lp;
825 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
826 	static int dumping;
827 
828 	unit = DISKUNIT(dev);
829 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
830 		return (ENXIO);
831 	if ((sc->sc_flags & LDF_ENABLED) == 0)
832 		return (ENODEV);
833 	if (sc->sc_dump == NULL)
834 		return (ENXIO);
835 
836 	/* Check if recursive dump; if so, punt. */
837 	if (dumping)
838 		return (EFAULT);
839 	dumping = 1;
840 
841 	/* Convert to disk sectors.  Request must be a multiple of size. */
842 	part = DISKPART(dev);
843 	lp = sc->sc_dk.dk_label;
844 	if ((size % lp->d_secsize) != 0)
845 		return (EFAULT);
846 	towrt = size / lp->d_secsize;
847 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
848 
849 	nsects = lp->d_partitions[part].p_size;
850 	sectoff = lp->d_partitions[part].p_offset;
851 
852 	/* Check transfer bounds against partition size. */
853 	if ((blkno < 0) || ((blkno + towrt) > nsects))
854 		return (EINVAL);
855 
856 	/* Offset block number to start of partition. */
857 	blkno += sectoff;
858 
859 	/* Start dumping and return when done. */
860 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
861 	while (towrt > 0) {
862 		nblk = min(maxblkcnt, towrt);
863 
864 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
865 			return (rv);
866 
867 		towrt -= nblk;
868 		blkno += nblk;
869 		va += nblk * sc->sc_secsize;
870 	}
871 
872 	dumping = 0;
873 	return (0);
874 }
875 
876 /*
877  * Adjust the size of a transfer.
878  */
879 static void
880 ldminphys(struct buf *bp)
881 {
882 	struct ld_softc *sc;
883 
884 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
885 
886 	if (bp->b_bcount > sc->sc_maxxfer)
887 		bp->b_bcount = sc->sc_maxxfer;
888 	minphys(bp);
889 }
890 
891 static void
892 ld_set_properties(struct ld_softc *ld)
893 {
894 	prop_dictionary_t disk_info, odisk_info, geom;
895 
896 	disk_info = prop_dictionary_create();
897 
898 	geom = prop_dictionary_create();
899 
900 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
901 	    ld->sc_secperunit);
902 
903 	prop_dictionary_set_uint32(geom, "sector-size",
904 	    ld->sc_secsize);
905 
906 	prop_dictionary_set_uint16(geom, "sectors-per-track",
907 	    ld->sc_nsectors);
908 
909 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
910 	    ld->sc_nheads);
911 
912 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
913 	    ld->sc_ncylinders);
914 
915 	prop_dictionary_set(disk_info, "geometry", geom);
916 	prop_object_release(geom);
917 
918 	prop_dictionary_set(device_properties(ld->sc_dv),
919 	    "disk-info", disk_info);
920 
921 	/*
922 	 * Don't release disk_info here; we keep a reference to it.
923 	 * disk_detach() will release it when we go away.
924 	 */
925 
926 	odisk_info = ld->sc_dk.dk_info;
927 	ld->sc_dk.dk_info = disk_info;
928 	if (odisk_info)
929 		prop_object_release(odisk_info);
930 }
931 
932 static void
933 ld_config_interrupts(device_t d)
934 {
935 	struct ld_softc *sc = device_private(d);
936 	dkwedge_discover(&sc->sc_dk);
937 }
938