xref: /netbsd-src/sys/dev/ld.c (revision 179b12252ecaf3553d9c2b7458ce62b6a2203d0c)
1 /*	$NetBSD: ld.c,v 1.67 2010/08/19 17:59:10 jmcneill 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 1.67 2010/08/19 17:59:10 jmcneill 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_suspend(device_t, const pmf_qual_t *);
71 static bool	ld_shutdown(device_t, int);
72 static void	ldstart(struct ld_softc *, struct buf *);
73 static void	ld_set_properties(struct ld_softc *);
74 static void	ld_config_interrupts (device_t);
75 static int	ldlastclose(device_t);
76 
77 extern struct	cfdriver ld_cd;
78 
79 static dev_type_open(ldopen);
80 static dev_type_close(ldclose);
81 static dev_type_read(ldread);
82 static dev_type_write(ldwrite);
83 static dev_type_ioctl(ldioctl);
84 static dev_type_strategy(ldstrategy);
85 static dev_type_dump(lddump);
86 static dev_type_size(ldsize);
87 
88 const struct bdevsw ld_bdevsw = {
89 	ldopen, ldclose, ldstrategy, ldioctl, lddump, ldsize, D_DISK
90 };
91 
92 const struct cdevsw ld_cdevsw = {
93 	ldopen, ldclose, ldread, ldwrite, ldioctl,
94 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
95 };
96 
97 static struct	dkdriver lddkdriver = { ldstrategy, ldminphys };
98 
99 void
100 ldattach(struct ld_softc *sc)
101 {
102 	char tbuf[9];
103 
104 	mutex_init(&sc->sc_mutex, MUTEX_DEFAULT, IPL_VM);
105 
106 	if ((sc->sc_flags & LDF_ENABLED) == 0) {
107 		aprint_normal_dev(sc->sc_dv, "disabled\n");
108 		return;
109 	}
110 
111 	/* Initialise and attach the disk structure. */
112 	disk_init(&sc->sc_dk, device_xname(sc->sc_dv), &lddkdriver);
113 	disk_attach(&sc->sc_dk);
114 
115 	if (sc->sc_maxxfer > MAXPHYS)
116 		sc->sc_maxxfer = MAXPHYS;
117 
118 	/* Build synthetic geometry if necessary. */
119 	if (sc->sc_nheads == 0 || sc->sc_nsectors == 0 ||
120 	    sc->sc_ncylinders == 0) {
121 		uint64_t ncyl;
122 
123 		if (sc->sc_secperunit <= 528 * 2048)		/* 528MB */
124 			sc->sc_nheads = 16;
125 		else if (sc->sc_secperunit <= 1024 * 2048)	/* 1GB */
126 			sc->sc_nheads = 32;
127 		else if (sc->sc_secperunit <= 21504 * 2048)	/* 21GB */
128 			sc->sc_nheads = 64;
129 		else if (sc->sc_secperunit <= 43008 * 2048)	/* 42GB */
130 			sc->sc_nheads = 128;
131 		else
132 			sc->sc_nheads = 255;
133 
134 		sc->sc_nsectors = 63;
135 		sc->sc_ncylinders = INT_MAX;
136 		ncyl = sc->sc_secperunit /
137 		    (sc->sc_nheads * sc->sc_nsectors);
138 		if (ncyl < INT_MAX)
139 			sc->sc_ncylinders = (int)ncyl;
140 	}
141 
142 	format_bytes(tbuf, sizeof(tbuf), sc->sc_secperunit *
143 	    sc->sc_secsize);
144 	aprint_normal_dev(sc->sc_dv, "%s, %d cyl, %d head, %d sec, "
145 	    "%d bytes/sect x %"PRIu64" sectors\n",
146 	    tbuf, sc->sc_ncylinders, sc->sc_nheads,
147 	    sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
148 
149 	ld_set_properties(sc);
150 
151 #if NRND > 0
152 	/* Attach the device into the rnd source list. */
153 	rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dv),
154 	    RND_TYPE_DISK, 0);
155 #endif
156 
157 	/* Register with PMF */
158 	if (!pmf_device_register1(sc->sc_dv, ld_suspend, NULL, ld_shutdown))
159 		aprint_error_dev(sc->sc_dv,
160 		    "couldn't establish power handler\n");
161 
162 	bufq_alloc(&sc->sc_bufq, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
163 
164 	/* Discover wedges on this disk. */
165 	config_interrupts(sc->sc_dv, ld_config_interrupts);
166 }
167 
168 int
169 ldadjqparam(struct ld_softc *sc, int xmax)
170 {
171 	int s;
172 
173 	s = splbio();
174 	sc->sc_maxqueuecnt = xmax;
175 	splx(s);
176 
177 	return (0);
178 }
179 
180 int
181 ldbegindetach(struct ld_softc *sc, int flags)
182 {
183 	int s, rv = 0;
184 
185 	if ((sc->sc_flags & LDF_ENABLED) == 0)
186 		return (0);
187 
188 	rv = disk_begindetach(&sc->sc_dk, ldlastclose, sc->sc_dv, flags);
189 
190 	if (rv != 0)
191 		return rv;
192 
193 	s = splbio();
194 	sc->sc_maxqueuecnt = 0;
195 	sc->sc_flags |= LDF_DETACH;
196 	while (sc->sc_queuecnt > 0) {
197 		sc->sc_flags |= LDF_DRAIN;
198 		rv = tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 0);
199 		if (rv)
200 			break;
201 	}
202 	splx(s);
203 
204 	return (rv);
205 }
206 
207 void
208 ldenddetach(struct ld_softc *sc)
209 {
210 	int s, bmaj, cmaj, i, mn;
211 
212 	if ((sc->sc_flags & LDF_ENABLED) == 0)
213 		return;
214 
215 	/* Wait for commands queued with the hardware to complete. */
216 	if (sc->sc_queuecnt != 0)
217 		if (tsleep(&sc->sc_queuecnt, PRIBIO, "lddtch", 30 * hz))
218 			printf("%s: not drained\n", device_xname(sc->sc_dv));
219 
220 	/* Locate the major numbers. */
221 	bmaj = bdevsw_lookup_major(&ld_bdevsw);
222 	cmaj = cdevsw_lookup_major(&ld_cdevsw);
223 
224 	/* Kill off any queued buffers. */
225 	s = splbio();
226 	bufq_drain(sc->sc_bufq);
227 	splx(s);
228 
229 	bufq_free(sc->sc_bufq);
230 
231 	/* Nuke the vnodes for any open instances. */
232 	for (i = 0; i < MAXPARTITIONS; i++) {
233 		mn = DISKMINOR(device_unit(sc->sc_dv), i);
234 		vdevgone(bmaj, mn, mn, VBLK);
235 		vdevgone(cmaj, mn, mn, VCHR);
236 	}
237 
238 	/* Delete all of our wedges. */
239 	dkwedge_delall(&sc->sc_dk);
240 
241 	/* Detach from the disk list. */
242 	disk_detach(&sc->sc_dk);
243 	disk_destroy(&sc->sc_dk);
244 
245 #if NRND > 0
246 	/* Unhook the entropy source. */
247 	rnd_detach_source(&sc->sc_rnd_source);
248 #endif
249 
250 	/* Deregister with PMF */
251 	pmf_device_deregister(sc->sc_dv);
252 
253 	/*
254 	 * XXX We can't really flush the cache here, beceause the
255 	 * XXX device may already be non-existent from the controller's
256 	 * XXX perspective.
257 	 */
258 #if 0
259 	/* Flush the device's cache. */
260 	if (sc->sc_flush != NULL)
261 		if ((*sc->sc_flush)(sc, 0) != 0)
262 			aprint_error_dev(&sc->sc_dv, "unable to flush cache\n");
263 #endif
264 	mutex_destroy(&sc->sc_mutex);
265 }
266 
267 /* ARGSUSED */
268 static bool
269 ld_suspend(device_t dev, const pmf_qual_t *qual)
270 {
271 	return ld_shutdown(dev, 0);
272 }
273 
274 /* ARGSUSED */
275 static bool
276 ld_shutdown(device_t dev, int flags)
277 {
278 	struct ld_softc *sc = device_private(dev);
279 
280 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, LDFL_POLL) != 0) {
281 		printf("%s: unable to flush cache\n", device_xname(dev));
282 		return false;
283 	}
284 
285 	return true;
286 }
287 
288 /* ARGSUSED */
289 static int
290 ldopen(dev_t dev, int flags, int fmt, struct lwp *l)
291 {
292 	struct ld_softc *sc;
293 	int error, unit, part;
294 
295 	unit = DISKUNIT(dev);
296 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
297 		return (ENXIO);
298 	if ((sc->sc_flags & LDF_ENABLED) == 0)
299 		return (ENODEV);
300 	part = DISKPART(dev);
301 
302 	mutex_enter(&sc->sc_dk.dk_openlock);
303 
304 	if (sc->sc_dk.dk_openmask == 0) {
305 		/* Load the partition info if not already loaded. */
306 		if ((sc->sc_flags & LDF_VLABEL) == 0)
307 			ldgetdisklabel(sc);
308 	}
309 
310 	/* Check that the partition exists. */
311 	if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
312 	    sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
313 		error = ENXIO;
314 		goto bad1;
315 	}
316 
317 	/* Ensure only one open at a time. */
318 	switch (fmt) {
319 	case S_IFCHR:
320 		sc->sc_dk.dk_copenmask |= (1 << part);
321 		break;
322 	case S_IFBLK:
323 		sc->sc_dk.dk_bopenmask |= (1 << part);
324 		break;
325 	}
326 	sc->sc_dk.dk_openmask =
327 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
328 
329 	error = 0;
330  bad1:
331 	mutex_exit(&sc->sc_dk.dk_openlock);
332 	return (error);
333 }
334 
335 static int
336 ldlastclose(device_t self)
337 {
338 	struct ld_softc *sc = device_private(self);
339 
340 	if (sc->sc_flush != NULL && (*sc->sc_flush)(sc, 0) != 0)
341 		aprint_error_dev(self, "unable to flush cache\n");
342 	if ((sc->sc_flags & LDF_KLABEL) == 0)
343 		sc->sc_flags &= ~LDF_VLABEL;
344 
345 	return 0;
346 }
347 
348 /* ARGSUSED */
349 static int
350 ldclose(dev_t dev, int flags, int fmt, struct lwp *l)
351 {
352 	struct ld_softc *sc;
353 	int part, unit;
354 
355 	unit = DISKUNIT(dev);
356 	part = DISKPART(dev);
357 	sc = device_lookup_private(&ld_cd, unit);
358 
359 	mutex_enter(&sc->sc_dk.dk_openlock);
360 
361 	switch (fmt) {
362 	case S_IFCHR:
363 		sc->sc_dk.dk_copenmask &= ~(1 << part);
364 		break;
365 	case S_IFBLK:
366 		sc->sc_dk.dk_bopenmask &= ~(1 << part);
367 		break;
368 	}
369 	sc->sc_dk.dk_openmask =
370 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
371 
372 	if (sc->sc_dk.dk_openmask == 0)
373 		ldlastclose(sc->sc_dv);
374 
375 	mutex_exit(&sc->sc_dk.dk_openlock);
376 	return (0);
377 }
378 
379 /* ARGSUSED */
380 static int
381 ldread(dev_t dev, struct uio *uio, int ioflag)
382 {
383 
384 	return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
385 }
386 
387 /* ARGSUSED */
388 static int
389 ldwrite(dev_t dev, struct uio *uio, int ioflag)
390 {
391 
392 	return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
393 }
394 
395 /* ARGSUSED */
396 static int
397 ldioctl(dev_t dev, u_long cmd, void *addr, int32_t flag, struct lwp *l)
398 {
399 	struct ld_softc *sc;
400 	int part, unit, error;
401 #ifdef __HAVE_OLD_DISKLABEL
402 	struct disklabel newlabel;
403 #endif
404 	struct disklabel *lp;
405 
406 	unit = DISKUNIT(dev);
407 	part = DISKPART(dev);
408 	sc = device_lookup_private(&ld_cd, unit);
409 
410 	error = disk_ioctl(&sc->sc_dk, cmd, addr, flag, l);
411 	if (error != EPASSTHROUGH)
412 		return (error);
413 
414 	error = 0;
415 	switch (cmd) {
416 	case DIOCGDINFO:
417 		memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
418 		return (0);
419 
420 #ifdef __HAVE_OLD_DISKLABEL
421 	case ODIOCGDINFO:
422 		newlabel = *(sc->sc_dk.dk_label);
423 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
424 			return ENOTTY;
425 		memcpy(addr, &newlabel, sizeof(struct olddisklabel));
426 		return (0);
427 #endif
428 
429 	case DIOCGPART:
430 		((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
431 		((struct partinfo *)addr)->part =
432 		    &sc->sc_dk.dk_label->d_partitions[part];
433 		break;
434 
435 	case DIOCWDINFO:
436 	case DIOCSDINFO:
437 #ifdef __HAVE_OLD_DISKLABEL
438 	case ODIOCWDINFO:
439 	case ODIOCSDINFO:
440 
441 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
442 			memset(&newlabel, 0, sizeof newlabel);
443 			memcpy(&newlabel, addr, sizeof (struct olddisklabel));
444 			lp = &newlabel;
445 		} else
446 #endif
447 		lp = (struct disklabel *)addr;
448 
449 		if ((flag & FWRITE) == 0)
450 			return (EBADF);
451 
452 		mutex_enter(&sc->sc_dk.dk_openlock);
453 		sc->sc_flags |= LDF_LABELLING;
454 
455 		error = setdisklabel(sc->sc_dk.dk_label,
456 		    lp, /*sc->sc_dk.dk_openmask : */0,
457 		    sc->sc_dk.dk_cpulabel);
458 		if (error == 0 && (cmd == DIOCWDINFO
459 #ifdef __HAVE_OLD_DISKLABEL
460 		    || cmd == ODIOCWDINFO
461 #endif
462 		    ))
463 			error = writedisklabel(
464 			    MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
465 			    ldstrategy, sc->sc_dk.dk_label,
466 			    sc->sc_dk.dk_cpulabel);
467 
468 		sc->sc_flags &= ~LDF_LABELLING;
469 		mutex_exit(&sc->sc_dk.dk_openlock);
470 		break;
471 
472 	case DIOCKLABEL:
473 		if ((flag & FWRITE) == 0)
474 			return (EBADF);
475 		if (*(int *)addr)
476 			sc->sc_flags |= LDF_KLABEL;
477 		else
478 			sc->sc_flags &= ~LDF_KLABEL;
479 		break;
480 
481 	case DIOCWLABEL:
482 		if ((flag & FWRITE) == 0)
483 			return (EBADF);
484 		if (*(int *)addr)
485 			sc->sc_flags |= LDF_WLABEL;
486 		else
487 			sc->sc_flags &= ~LDF_WLABEL;
488 		break;
489 
490 	case DIOCGDEFLABEL:
491 		ldgetdefaultlabel(sc, (struct disklabel *)addr);
492 		break;
493 
494 #ifdef __HAVE_OLD_DISKLABEL
495 	case ODIOCGDEFLABEL:
496 		ldgetdefaultlabel(sc, &newlabel);
497 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
498 			return ENOTTY;
499 		memcpy(addr, &newlabel, sizeof (struct olddisklabel));
500 		break;
501 #endif
502 
503 	case DIOCCACHESYNC:
504 		/*
505 		 * XXX Do we really need to care about having a writable
506 		 * file descriptor here?
507 		 */
508 		if ((flag & FWRITE) == 0)
509 			error = EBADF;
510 		else if (sc->sc_flush)
511 			error = (*sc->sc_flush)(sc, 0);
512 		else
513 			error = 0;	/* XXX Error out instead? */
514 		break;
515 
516 	case DIOCAWEDGE:
517 	    {
518 	    	struct dkwedge_info *dkw = (void *) addr;
519 
520 		if ((flag & FWRITE) == 0)
521 			return (EBADF);
522 
523 		/* If the ioctl happens here, the parent is us. */
524 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
525 			sizeof(dkw->dkw_parent));
526 		return (dkwedge_add(dkw));
527 	    }
528 
529 	case DIOCDWEDGE:
530 	    {
531 	    	struct dkwedge_info *dkw = (void *) addr;
532 
533 		if ((flag & FWRITE) == 0)
534 			return (EBADF);
535 
536 		/* If the ioctl happens here, the parent is us. */
537 		strlcpy(dkw->dkw_parent, device_xname(sc->sc_dv),
538 			sizeof(dkw->dkw_parent));
539 		return (dkwedge_del(dkw));
540 	    }
541 
542 	case DIOCLWEDGES:
543 	    {
544 	    	struct dkwedge_list *dkwl = (void *) addr;
545 
546 		return (dkwedge_list(&sc->sc_dk, dkwl, l));
547 	    }
548 	case DIOCGSTRATEGY:
549 	    {
550 		struct disk_strategy *dks = (void *)addr;
551 
552 		mutex_enter(&sc->sc_mutex);
553 		strlcpy(dks->dks_name, bufq_getstrategyname(sc->sc_bufq),
554 		    sizeof(dks->dks_name));
555 		mutex_exit(&sc->sc_mutex);
556 		dks->dks_paramlen = 0;
557 
558 		return 0;
559 	    }
560 	case DIOCSSTRATEGY:
561 	    {
562 		struct disk_strategy *dks = (void *)addr;
563 		struct bufq_state *new, *old;
564 
565 		if ((flag & FWRITE) == 0)
566 			return EPERM;
567 
568 		if (dks->dks_param != NULL)
569 			return EINVAL;
570 
571 		dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
572 		error = bufq_alloc(&new, dks->dks_name,
573 		    BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
574 		if (error)
575 			return error;
576 
577 		mutex_enter(&sc->sc_mutex);
578 		old = sc->sc_bufq;
579 		bufq_move(new, old);
580 		sc->sc_bufq = new;
581 		mutex_exit(&sc->sc_mutex);
582 		bufq_free(old);
583 
584 		return 0;
585 	    }
586 	default:
587 		error = ENOTTY;
588 		break;
589 	}
590 
591 	return (error);
592 }
593 
594 static void
595 ldstrategy(struct buf *bp)
596 {
597 	struct ld_softc *sc;
598 	struct disklabel *lp;
599 	daddr_t blkno;
600 	int s, part;
601 
602 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
603 	part = DISKPART(bp->b_dev);
604 
605 	if ((sc->sc_flags & LDF_DETACH) != 0) {
606 		bp->b_error = EIO;
607 		goto done;
608 	}
609 
610 	lp = sc->sc_dk.dk_label;
611 
612 	/*
613 	 * The transfer must be a whole number of blocks and the offset must
614 	 * not be negative.
615 	 */
616 	if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
617 		bp->b_error = EINVAL;
618 		goto done;
619 	}
620 
621 	/* If it's a null transfer, return immediately. */
622 	if (bp->b_bcount == 0)
623 		goto done;
624 
625 	/*
626 	 * Do bounds checking and adjust the transfer.  If error, process.
627 	 * If past the end of partition, just return.
628 	 */
629 	if (part != RAW_PART &&
630 	    bounds_check_with_label(&sc->sc_dk, bp,
631 	    (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
632 		goto done;
633 	}
634 
635 	/*
636 	 * Convert the block number to absolute and put it in terms
637 	 * of the device's logical block size.
638 	 */
639 	if (lp->d_secsize == DEV_BSIZE)
640 		blkno = bp->b_blkno;
641 	else if (lp->d_secsize > DEV_BSIZE)
642 		blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
643 	else
644 		blkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
645 
646 	if (part != RAW_PART)
647 		blkno += lp->d_partitions[part].p_offset;
648 
649 	bp->b_rawblkno = blkno;
650 
651 	s = splbio();
652 	ldstart(sc, bp);
653 	splx(s);
654 	return;
655 
656  done:
657 	bp->b_resid = bp->b_bcount;
658 	biodone(bp);
659 }
660 
661 static void
662 ldstart(struct ld_softc *sc, struct buf *bp)
663 {
664 	int error;
665 
666 	mutex_enter(&sc->sc_mutex);
667 
668 	if (bp != NULL)
669 		bufq_put(sc->sc_bufq, bp);
670 
671 	while (sc->sc_queuecnt < sc->sc_maxqueuecnt) {
672 		/* See if there is work to do. */
673 		if ((bp = bufq_peek(sc->sc_bufq)) == NULL)
674 			break;
675 
676 		disk_busy(&sc->sc_dk);
677 		sc->sc_queuecnt++;
678 
679 		if (__predict_true((error = (*sc->sc_start)(sc, bp)) == 0)) {
680 			/*
681 			 * The back-end is running the job; remove it from
682 			 * the queue.
683 			 */
684 			(void) bufq_get(sc->sc_bufq);
685 		} else  {
686 			disk_unbusy(&sc->sc_dk, 0, (bp->b_flags & B_READ));
687 			sc->sc_queuecnt--;
688 			if (error == EAGAIN) {
689 				/*
690 				 * Temporary resource shortage in the
691 				 * back-end; just defer the job until
692 				 * later.
693 				 *
694 				 * XXX We might consider a watchdog timer
695 				 * XXX to make sure we are kicked into action.
696 				 */
697 				break;
698 			} else {
699 				(void) bufq_get(sc->sc_bufq);
700 				bp->b_error = error;
701 				bp->b_resid = bp->b_bcount;
702 				mutex_exit(&sc->sc_mutex);
703 				biodone(bp);
704 				mutex_enter(&sc->sc_mutex);
705 			}
706 		}
707 	}
708 
709 	mutex_exit(&sc->sc_mutex);
710 }
711 
712 void
713 lddone(struct ld_softc *sc, struct buf *bp)
714 {
715 
716 	if (bp->b_error != 0) {
717 		diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
718 		printf("\n");
719 	}
720 
721 	disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid,
722 	    (bp->b_flags & B_READ));
723 #if NRND > 0
724 	rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
725 #endif
726 	biodone(bp);
727 
728 	mutex_enter(&sc->sc_mutex);
729 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
730 		if ((sc->sc_flags & LDF_DRAIN) != 0) {
731 			sc->sc_flags &= ~LDF_DRAIN;
732 			wakeup(&sc->sc_queuecnt);
733 		}
734 		mutex_exit(&sc->sc_mutex);
735 		ldstart(sc, NULL);
736 	} else
737 		mutex_exit(&sc->sc_mutex);
738 }
739 
740 static int
741 ldsize(dev_t dev)
742 {
743 	struct ld_softc *sc;
744 	int part, unit, omask, size;
745 
746 	unit = DISKUNIT(dev);
747 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
748 		return (ENODEV);
749 	if ((sc->sc_flags & LDF_ENABLED) == 0)
750 		return (ENODEV);
751 	part = DISKPART(dev);
752 
753 	omask = sc->sc_dk.dk_openmask & (1 << part);
754 
755 	if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
756 		return (-1);
757 	else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
758 		size = -1;
759 	else
760 		size = sc->sc_dk.dk_label->d_partitions[part].p_size *
761 		    (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
762 	if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
763 		return (-1);
764 
765 	return (size);
766 }
767 
768 /*
769  * Load the label information from the specified device.
770  */
771 static void
772 ldgetdisklabel(struct ld_softc *sc)
773 {
774 	const char *errstring;
775 
776 	ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
777 
778 	/* Call the generic disklabel extraction routine. */
779 	errstring = readdisklabel(MAKEDISKDEV(0, device_unit(sc->sc_dv),
780 	    RAW_PART), ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
781 	if (errstring != NULL)
782 		printf("%s: %s\n", device_xname(sc->sc_dv), errstring);
783 
784 	/* In-core label now valid. */
785 	sc->sc_flags |= LDF_VLABEL;
786 }
787 
788 /*
789  * Construct a ficticious label.
790  */
791 static void
792 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
793 {
794 
795 	memset(lp, 0, sizeof(struct disklabel));
796 
797 	lp->d_secsize = sc->sc_secsize;
798 	lp->d_ntracks = sc->sc_nheads;
799 	lp->d_nsectors = sc->sc_nsectors;
800 	lp->d_ncylinders = sc->sc_ncylinders;
801 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
802 	lp->d_type = DTYPE_LD;
803 	strlcpy(lp->d_typename, "unknown", sizeof(lp->d_typename));
804 	strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
805 	lp->d_secperunit = sc->sc_secperunit;
806 	lp->d_rpm = 7200;
807 	lp->d_interleave = 1;
808 	lp->d_flags = 0;
809 
810 	lp->d_partitions[RAW_PART].p_offset = 0;
811 	lp->d_partitions[RAW_PART].p_size =
812 	    lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
813 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
814 	lp->d_npartitions = RAW_PART + 1;
815 
816 	lp->d_magic = DISKMAGIC;
817 	lp->d_magic2 = DISKMAGIC;
818 	lp->d_checksum = dkcksum(lp);
819 }
820 
821 /*
822  * Take a dump.
823  */
824 static int
825 lddump(dev_t dev, daddr_t blkno, void *vav, size_t size)
826 {
827 	char *va = vav;
828 	struct ld_softc *sc;
829 	struct disklabel *lp;
830 	int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
831 	static int dumping;
832 
833 	unit = DISKUNIT(dev);
834 	if ((sc = device_lookup_private(&ld_cd, unit)) == NULL)
835 		return (ENXIO);
836 	if ((sc->sc_flags & LDF_ENABLED) == 0)
837 		return (ENODEV);
838 	if (sc->sc_dump == NULL)
839 		return (ENXIO);
840 
841 	/* Check if recursive dump; if so, punt. */
842 	if (dumping)
843 		return (EFAULT);
844 	dumping = 1;
845 
846 	/* Convert to disk sectors.  Request must be a multiple of size. */
847 	part = DISKPART(dev);
848 	lp = sc->sc_dk.dk_label;
849 	if ((size % lp->d_secsize) != 0)
850 		return (EFAULT);
851 	towrt = size / lp->d_secsize;
852 	blkno = dbtob(blkno) / lp->d_secsize;	/* blkno in DEV_BSIZE units */
853 
854 	nsects = lp->d_partitions[part].p_size;
855 	sectoff = lp->d_partitions[part].p_offset;
856 
857 	/* Check transfer bounds against partition size. */
858 	if ((blkno < 0) || ((blkno + towrt) > nsects))
859 		return (EINVAL);
860 
861 	/* Offset block number to start of partition. */
862 	blkno += sectoff;
863 
864 	/* Start dumping and return when done. */
865 	maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
866 	while (towrt > 0) {
867 		nblk = min(maxblkcnt, towrt);
868 
869 		if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
870 			return (rv);
871 
872 		towrt -= nblk;
873 		blkno += nblk;
874 		va += nblk * sc->sc_secsize;
875 	}
876 
877 	dumping = 0;
878 	return (0);
879 }
880 
881 /*
882  * Adjust the size of a transfer.
883  */
884 static void
885 ldminphys(struct buf *bp)
886 {
887 	struct ld_softc *sc;
888 
889 	sc = device_lookup_private(&ld_cd, DISKUNIT(bp->b_dev));
890 
891 	if (bp->b_bcount > sc->sc_maxxfer)
892 		bp->b_bcount = sc->sc_maxxfer;
893 	minphys(bp);
894 }
895 
896 static void
897 ld_set_properties(struct ld_softc *ld)
898 {
899 	prop_dictionary_t disk_info, odisk_info, geom;
900 
901 	disk_info = prop_dictionary_create();
902 
903 	geom = prop_dictionary_create();
904 
905 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
906 	    ld->sc_secperunit);
907 
908 	prop_dictionary_set_uint32(geom, "sector-size",
909 	    ld->sc_secsize);
910 
911 	prop_dictionary_set_uint16(geom, "sectors-per-track",
912 	    ld->sc_nsectors);
913 
914 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
915 	    ld->sc_nheads);
916 
917 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
918 	    ld->sc_ncylinders);
919 
920 	prop_dictionary_set(disk_info, "geometry", geom);
921 	prop_object_release(geom);
922 
923 	prop_dictionary_set(device_properties(ld->sc_dv),
924 	    "disk-info", disk_info);
925 
926 	/*
927 	 * Don't release disk_info here; we keep a reference to it.
928 	 * disk_detach() will release it when we go away.
929 	 */
930 
931 	odisk_info = ld->sc_dk.dk_info;
932 	ld->sc_dk.dk_info = disk_info;
933 	if (odisk_info)
934 		prop_object_release(odisk_info);
935 }
936 
937 static void
938 ld_config_interrupts(device_t d)
939 {
940 	struct ld_softc *sc = device_private(d);
941 	dkwedge_discover(&sc->sc_dk);
942 }
943