xref: /dflybsd-src/sys/kern/subr_disk.c (revision dc77152fd0d00e1b1f8b8cb04d0706817b468ddd)
1 /*
2  * ----------------------------------------------------------------------------
3  * "THE BEER-WARE LICENSE" (Revision 42):
4  * <phk@FreeBSD.ORG> wrote this file.  As long as you retain this notice you
5  * can do whatever you want with this stuff. If we meet some day, and you think
6  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
7  * ----------------------------------------------------------------------------
8  *
9  * Copyright (c) 2004 Matthew Dillon.
10  * Copyright (c) 1982, 1986, 1988, 1993
11  *	The Regents of the University of California.  All rights reserved.
12  * (c) UNIX System Laboratories, Inc.
13  * All or some portions of this file are derived from material licensed
14  * to the University of California by American Telephone and Telegraph
15  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
16  * the permission of UNIX System Laboratories, Inc.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  * 3. All advertising materials mentioning features or use of this software
27  *    must display the following acknowledgement:
28  *	This product includes software developed by the University of
29  *	California, Berkeley and its contributors.
30  * 4. Neither the name of the University nor the names of its contributors
31  *    may be used to endorse or promote products derived from this software
32  *    without specific prior written permission.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
35  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
36  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
37  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
38  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
39  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
40  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
41  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
42  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
43  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44  * SUCH DAMAGE.
45  *
46  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
47  * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $
48  * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $
49  * $DragonFly: src/sys/kern/subr_disk.c,v 1.12 2004/06/02 19:31:02 dillon Exp $
50  */
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kernel.h>
55 #include <sys/proc.h>
56 #include <sys/sysctl.h>
57 #include <sys/buf.h>
58 #include <sys/conf.h>
59 #include <sys/disklabel.h>
60 #include <sys/diskslice.h>
61 #include <sys/disk.h>
62 #include <sys/malloc.h>
63 #include <sys/sysctl.h>
64 #include <machine/md_var.h>
65 #include <sys/ctype.h>
66 #include <sys/syslog.h>
67 #include <sys/device.h>
68 #include <sys/msgport.h>
69 #include <sys/msgport2.h>
70 #include <sys/buf2.h>
71 
72 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
73 
74 static d_strategy_t diskstrategy;
75 static d_open_t diskopen;
76 static d_close_t diskclose;
77 static d_ioctl_t diskioctl;
78 static d_psize_t diskpsize;
79 static d_clone_t diskclone;
80 static int disk_putport(lwkt_port_t port, lwkt_msg_t msg);
81 
82 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
83 
84 /*
85  * Create a slice and unit managed disk.
86  *
87  * Our port layer will be responsible for assigning pblkno and handling
88  * high level partition operations, then forwarding the requests to the
89  * raw device.
90  *
91  * The raw device (based on rawsw) is returned to the caller, NOT the
92  * slice and unit managed cdev.  The caller typically sets various
93  * driver parameters and IO limits on the returned rawdev which we must
94  * inherit when our managed device is opened.
95  */
96 dev_t
97 disk_create(int unit, struct disk *dp, int flags, struct cdevsw *rawsw)
98 {
99 	dev_t rawdev;
100 	struct cdevsw *devsw;
101 
102 	/*
103 	 * Create the raw backing device
104 	 */
105 	compile_devsw(rawsw);
106 	rawdev = make_dev(rawsw, dkmakeminor(unit, WHOLE_DISK_SLICE, RAW_PART),
107 			    UID_ROOT, GID_OPERATOR, 0640,
108 			    "%s%d", rawsw->d_name, unit);
109 
110 	/*
111 	 * Initialize our intercept port
112 	 */
113 	bzero(dp, sizeof(*dp));
114 	lwkt_initport(&dp->d_port, NULL);
115 	dp->d_port.mp_putport = disk_putport;
116 	dp->d_rawsw = rawsw;
117 
118 	/*
119 	 * We install a custom cdevsw rather then the passed cdevsw,
120 	 * and save our disk structure in d_data so we can get at it easily
121 	 * without any complex cloning code.
122 	 */
123 	devsw = cdevsw_add_override(rawdev, dkunitmask(), dkmakeunit(unit));
124 	devsw->d_port = &dp->d_port;
125 	devsw->d_data = dp;
126 	devsw->d_clone = diskclone;
127 	dp->d_devsw = devsw;
128 	dp->d_rawdev = rawdev;
129 	dp->d_cdev = make_dev(devsw,
130 			    dkmakeminor(unit, WHOLE_DISK_SLICE, RAW_PART),
131 			    UID_ROOT, GID_OPERATOR, 0640,
132 			    "%s%d", devsw->d_name, unit);
133 
134 	dp->d_dsflags = flags;
135 	LIST_INSERT_HEAD(&disklist, dp, d_list);
136 	return (dp->d_rawdev);
137 }
138 
139 /*
140  * This routine is called when an adapter detaches.  The higher level
141  * managed disk device is destroyed while the lower level raw device is
142  * released.
143  */
144 void
145 disk_destroy(struct disk *disk)
146 {
147 	if (disk->d_devsw) {
148 	    cdevsw_remove(disk->d_devsw, dkunitmask(), dkunit(disk->d_cdev));
149 	    LIST_REMOVE(disk, d_list);
150 	}
151 	if (disk->d_rawsw)
152 	    destroy_all_dev(disk->d_rawsw, dkunitmask(), dkunit(disk->d_rawdev));
153 	bzero(disk, sizeof(*disk));
154 }
155 
156 int
157 disk_dumpcheck(dev_t dev, u_int *count, u_int *blkno, u_int *secsize)
158 {
159 	struct disk *dp;
160 	struct disklabel *dl;
161 	u_int boff;
162 
163 	dp = dev->si_disk;
164 	if (!dp)
165 		return (ENXIO);
166 	if (!dp->d_slice)
167 		return (ENXIO);
168 	dl = dsgetlabel(dev, dp->d_slice);
169 	if (!dl)
170 		return (ENXIO);
171 	*count = Maxmem * (PAGE_SIZE / dl->d_secsize);
172 	if (dumplo <= LABELSECTOR ||
173 	    (dumplo + *count > dl->d_partitions[dkpart(dev)].p_size))
174 		return (EINVAL);
175 	boff = dl->d_partitions[dkpart(dev)].p_offset +
176 	    dp->d_slice->dss_slices[dkslice(dev)].ds_offset;
177 	*blkno = boff + dumplo;
178 	*secsize = dl->d_secsize;
179 	return (0);
180 
181 }
182 
183 void
184 disk_invalidate (struct disk *disk)
185 {
186 	if (disk->d_slice)
187 		dsgone(&disk->d_slice);
188 }
189 
190 struct disk *
191 disk_enumerate(struct disk *disk)
192 {
193 	if (!disk)
194 		return (LIST_FIRST(&disklist));
195 	else
196 		return (LIST_NEXT(disk, d_list));
197 }
198 
199 static
200 int
201 sysctl_disks(SYSCTL_HANDLER_ARGS)
202 {
203 	struct disk *disk;
204 	int error, first;
205 
206 	disk = NULL;
207 	first = 1;
208 
209 	while ((disk = disk_enumerate(disk))) {
210 		if (!first) {
211 			error = SYSCTL_OUT(req, " ", 1);
212 			if (error)
213 				return error;
214 		} else {
215 			first = 0;
216 		}
217 		error = SYSCTL_OUT(req, disk->d_rawdev->si_name, strlen(disk->d_rawdev->si_name));
218 		if (error)
219 			return error;
220 	}
221 	error = SYSCTL_OUT(req, "", 1);
222 	return error;
223 }
224 
225 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, 0, NULL,
226     sysctl_disks, "A", "names of available disks");
227 
228 /*
229  * The port intercept functions
230  */
231 static
232 int
233 disk_putport(lwkt_port_t port, lwkt_msg_t lmsg)
234 {
235 	struct disk *disk = (struct disk *)port;
236 	cdevallmsg_t msg = (cdevallmsg_t)lmsg;
237 	int error;
238 
239 	switch(msg->am_lmsg.ms_cmd.cm_op) {
240 	case CDEV_CMD_OPEN:
241 		error = diskopen(
242 			    msg->am_open.msg.dev,
243 			    msg->am_open.oflags,
244 			    msg->am_open.devtype,
245 			    msg->am_open.td);
246 		break;
247 	case CDEV_CMD_CLOSE:
248 		error = diskclose(
249 			    msg->am_close.msg.dev,
250 			    msg->am_close.fflag,
251 			    msg->am_close.devtype,
252 			    msg->am_close.td);
253 		break;
254 	case CDEV_CMD_IOCTL:
255 		error = diskioctl(
256 			    msg->am_ioctl.msg.dev,
257 			    msg->am_ioctl.cmd,
258 			    msg->am_ioctl.data,
259 			    msg->am_ioctl.fflag,
260 			    msg->am_ioctl.td);
261 		break;
262 	case CDEV_CMD_STRATEGY:
263 		diskstrategy(msg->am_strategy.bp);
264 		error = 0;
265 		break;
266 	case CDEV_CMD_PSIZE:
267 		msg->am_psize.result = diskpsize(msg->am_psize.msg.dev);
268 		error = 0;      /* XXX */
269 		break;
270 	case CDEV_CMD_READ:
271 		error = physio(msg->am_read.msg.dev,
272 				msg->am_read.uio, msg->am_read.ioflag);
273 		break;
274 	case CDEV_CMD_WRITE:
275 		error = physio(msg->am_write.msg.dev,
276 				msg->am_write.uio, msg->am_write.ioflag);
277 		break;
278 	case CDEV_CMD_POLL:
279 	case CDEV_CMD_KQFILTER:
280 		error = ENODEV;
281 	case CDEV_CMD_MMAP:
282 		error = -1;
283 		break;
284 	case CDEV_CMD_DUMP:
285 		error = disk_dumpcheck(msg->am_dump.msg.dev,
286 				&msg->am_dump.count,
287 				&msg->am_dump.blkno,
288 				&msg->am_dump.secsize);
289 		if (error == 0) {
290 			msg->am_dump.msg.dev = disk->d_rawdev;
291 			error = lwkt_forwardmsg(disk->d_rawdev->si_port,
292 						&msg->am_dump.msg.msg);
293 			printf("error2 %d\n", error);
294 		}
295 		break;
296 	default:
297 		error = ENOTSUP;
298 		break;
299 	}
300 	return(error);
301 }
302 
303 /*
304  * When new device entries are instantiated, make sure they inherit our
305  * si_disk structure and block and iosize limits from the raw device.
306  *
307  * This routine is always called synchronously in the context of the
308  * client.
309  *
310  * XXX The various io and block size constraints are not always initialized
311  * properly by devices.
312  */
313 static
314 int
315 diskclone(dev_t dev)
316 {
317 	struct disk *dp;
318 
319 	dp = dev->si_devsw->d_data;
320 	KKASSERT(dp != NULL);
321 	dev->si_disk = dp;
322 	dev->si_iosize_max = dp->d_rawdev->si_iosize_max;
323 	dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys;
324 	dev->si_bsize_best = dp->d_rawdev->si_bsize_best;
325 	return(0);
326 }
327 
328 /*
329  * Open a disk device or partition.
330  */
331 static
332 int
333 diskopen(dev_t dev, int oflags, int devtype, struct thread *td)
334 {
335 	struct disk *dp;
336 	int error;
337 
338 	/*
339 	 * dp can't be NULL here XXX.
340 	 */
341 	error = 0;
342 	dp = dev->si_disk;
343 	if (dp == NULL)
344 		return (ENXIO);
345 
346 	/*
347 	 * Deal with open races
348 	 */
349 	while (dp->d_flags & DISKFLAG_LOCK) {
350 		dp->d_flags |= DISKFLAG_WANTED;
351 		error = tsleep(dp, PCATCH, "diskopen", hz);
352 		if (error)
353 			return (error);
354 	}
355 	dp->d_flags |= DISKFLAG_LOCK;
356 
357 	/*
358 	 * Open the underlying raw device.
359 	 */
360 	if (!dsisopen(dp->d_slice)) {
361 #if 0
362 		if (!pdev->si_iosize_max)
363 			pdev->si_iosize_max = dev->si_iosize_max;
364 #endif
365 		error = dev_dopen(dp->d_rawdev, oflags, devtype, td);
366 	}
367 
368 	/*
369 	 * Inherit properties from the underlying device now that it is
370 	 * open.
371 	 */
372 	diskclone(dev);
373 
374 	if (error)
375 		goto out;
376 
377 	error = dsopen(dev, devtype, dp->d_dsflags, &dp->d_slice, &dp->d_label);
378 
379 	if (!dsisopen(dp->d_slice))
380 		dev_dclose(dp->d_rawdev, oflags, devtype, td);
381 out:
382 	dp->d_flags &= ~DISKFLAG_LOCK;
383 	if (dp->d_flags & DISKFLAG_WANTED) {
384 		dp->d_flags &= ~DISKFLAG_WANTED;
385 		wakeup(dp);
386 	}
387 
388 	return(error);
389 }
390 
391 /*
392  * Close a disk device or partition
393  */
394 static
395 int
396 diskclose(dev_t dev, int fflag, int devtype, struct thread *td)
397 {
398 	struct disk *dp;
399 	int error;
400 
401 	error = 0;
402 	dp = dev->si_disk;
403 
404 	dsclose(dev, devtype, dp->d_slice);
405 	if (!dsisopen(dp->d_slice))
406 		error = dev_dclose(dp->d_rawdev, fflag, devtype, td);
407 	return (error);
408 }
409 
410 /*
411  * Execute strategy routine
412  */
413 static
414 void
415 diskstrategy(struct buf *bp)
416 {
417 	struct disk *dp;
418 
419 	dp = bp->b_dev->si_disk;
420 
421 	if (dp == NULL) {
422 		bp->b_error = ENXIO;
423 		bp->b_flags |= B_ERROR;
424 		biodone(bp);
425 		return;
426 	}
427 	KKASSERT(bp->b_dev->si_disk == dp);
428 
429 	if (dscheck(bp, dp->d_slice) <= 0) {
430 		biodone(bp);
431 		return;
432 	}
433 	bp->b_dev = dp->d_rawdev;
434 	dev_dstrategy(dp->d_rawdev, bp);
435 }
436 
437 /*
438  * First execute the ioctl on the disk device, and if it isn't supported
439  * try running it on the backing device.
440  */
441 static
442 int
443 diskioctl(dev_t dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
444 {
445 	struct disk *dp;
446 	int error;
447 
448 	dp = dev->si_disk;
449 	if (dp == NULL)
450 		return (ENXIO);
451 
452 	error = dsioctl(dev, cmd, data, fflag, &dp->d_slice);
453 	if (error == ENOIOCTL)
454 		error = dev_dioctl(dp->d_rawdev, cmd, data, fflag, td);
455 	return (error);
456 }
457 
458 /*
459  *
460  */
461 static
462 int
463 diskpsize(dev_t dev)
464 {
465 	struct disk *dp;
466 
467 	dp = dev->si_disk;
468 	if (dp == NULL)
469 		return (-1);
470 	return(dssize(dev, &dp->d_slice));
471 #if 0
472 	if (dp != dev->si_disk) {
473 		dev->si_drv1 = pdev->si_drv1;
474 		dev->si_drv2 = pdev->si_drv2;
475 		/* XXX: don't set bp->b_dev->si_disk (?) */
476 	}
477 #endif
478 }
479 
480 SYSCTL_DECL(_debug_sizeof);
481 
482 SYSCTL_INT(_debug_sizeof, OID_AUTO, disklabel, CTLFLAG_RD,
483     0, sizeof(struct disklabel), "sizeof(struct disklabel)");
484 
485 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
486     0, sizeof(struct diskslices), "sizeof(struct diskslices)");
487 
488 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
489     0, sizeof(struct disk), "sizeof(struct disk)");
490 
491 
492 /*
493  * Seek sort for disks.
494  *
495  * The buf_queue keep two queues, sorted in ascending block order.  The first
496  * queue holds those requests which are positioned after the current block
497  * (in the first request); the second, which starts at queue->switch_point,
498  * holds requests which came in after their block number was passed.  Thus
499  * we implement a one way scan, retracting after reaching the end of the drive
500  * to the first request on the second queue, at which time it becomes the
501  * first queue.
502  *
503  * A one-way scan is natural because of the way UNIX read-ahead blocks are
504  * allocated.
505  */
506 void
507 bufqdisksort(struct buf_queue_head *bufq, struct buf *bp)
508 {
509 	struct buf *bq;
510 	struct buf *bn;
511 	struct buf *be;
512 
513 	be = TAILQ_LAST(&bufq->queue, buf_queue);
514 	/*
515 	 * If the queue is empty or we are an
516 	 * ordered transaction, then it's easy.
517 	 */
518 	if ((bq = bufq_first(bufq)) == NULL ||
519 	    (bp->b_flags & B_ORDERED) != 0) {
520 		bufq_insert_tail(bufq, bp);
521 		return;
522 	} else if (bufq->insert_point != NULL) {
523 
524 		/*
525 		 * A certain portion of the list is
526 		 * "locked" to preserve ordering, so
527 		 * we can only insert after the insert
528 		 * point.
529 		 */
530 		bq = bufq->insert_point;
531 	} else {
532 
533 		/*
534 		 * If we lie before the last removed (currently active)
535 		 * request, and are not inserting ourselves into the
536 		 * "locked" portion of the list, then we must add ourselves
537 		 * to the second request list.
538 		 */
539 		if (bp->b_pblkno < bufq->last_pblkno) {
540 
541 			bq = bufq->switch_point;
542 			/*
543 			 * If we are starting a new secondary list,
544 			 * then it's easy.
545 			 */
546 			if (bq == NULL) {
547 				bufq->switch_point = bp;
548 				bufq_insert_tail(bufq, bp);
549 				return;
550 			}
551 			/*
552 			 * If we lie ahead of the current switch point,
553 			 * insert us before the switch point and move
554 			 * the switch point.
555 			 */
556 			if (bp->b_pblkno < bq->b_pblkno) {
557 				bufq->switch_point = bp;
558 				TAILQ_INSERT_BEFORE(bq, bp, b_act);
559 				return;
560 			}
561 		} else {
562 			if (bufq->switch_point != NULL)
563 				be = TAILQ_PREV(bufq->switch_point,
564 						buf_queue, b_act);
565 			/*
566 			 * If we lie between last_pblkno and bq,
567 			 * insert before bq.
568 			 */
569 			if (bp->b_pblkno < bq->b_pblkno) {
570 				TAILQ_INSERT_BEFORE(bq, bp, b_act);
571 				return;
572 			}
573 		}
574 	}
575 
576 	/*
577 	 * Request is at/after our current position in the list.
578 	 * Optimize for sequential I/O by seeing if we go at the tail.
579 	 */
580 	if (bp->b_pblkno > be->b_pblkno) {
581 		TAILQ_INSERT_AFTER(&bufq->queue, be, bp, b_act);
582 		return;
583 	}
584 
585 	/* Otherwise, insertion sort */
586 	while ((bn = TAILQ_NEXT(bq, b_act)) != NULL) {
587 
588 		/*
589 		 * We want to go after the current request if it is the end
590 		 * of the first request list, or if the next request is a
591 		 * larger cylinder than our request.
592 		 */
593 		if (bn == bufq->switch_point
594 		 || bp->b_pblkno < bn->b_pblkno)
595 			break;
596 		bq = bn;
597 	}
598 	TAILQ_INSERT_AFTER(&bufq->queue, bq, bp, b_act);
599 }
600 
601 
602 /*
603  * Attempt to read a disk label from a device using the indicated strategy
604  * routine.  The label must be partly set up before this: secpercyl, secsize
605  * and anything required in the strategy routine (e.g., dummy bounds for the
606  * partition containing the label) must be filled in before calling us.
607  * Returns NULL on success and an error string on failure.
608  */
609 char *
610 readdisklabel(dev_t dev, struct disklabel *lp)
611 {
612 	struct buf *bp;
613 	struct disklabel *dlp;
614 	char *msg = NULL;
615 
616 	bp = geteblk((int)lp->d_secsize);
617 	bp->b_dev = dev;
618 	bp->b_blkno = LABELSECTOR * ((int)lp->d_secsize/DEV_BSIZE);
619 	bp->b_bcount = lp->d_secsize;
620 	bp->b_flags &= ~B_INVAL;
621 	bp->b_flags |= B_READ;
622 	BUF_STRATEGY(bp, 1);
623 	if (biowait(bp))
624 		msg = "I/O error";
625 	else for (dlp = (struct disklabel *)bp->b_data;
626 	    dlp <= (struct disklabel *)((char *)bp->b_data +
627 	    lp->d_secsize - sizeof(*dlp));
628 	    dlp = (struct disklabel *)((char *)dlp + sizeof(long))) {
629 		if (dlp->d_magic != DISKMAGIC || dlp->d_magic2 != DISKMAGIC) {
630 			if (msg == NULL)
631 				msg = "no disk label";
632 		} else if (dlp->d_npartitions > MAXPARTITIONS ||
633 			   dkcksum(dlp) != 0)
634 			msg = "disk label corrupted";
635 		else {
636 			*lp = *dlp;
637 			msg = NULL;
638 			break;
639 		}
640 	}
641 	bp->b_flags |= B_INVAL | B_AGE;
642 	brelse(bp);
643 	return (msg);
644 }
645 
646 /*
647  * Check new disk label for sensibility before setting it.
648  */
649 int
650 setdisklabel(struct disklabel *olp, struct disklabel *nlp, u_long openmask)
651 {
652 	int i;
653 	struct partition *opp, *npp;
654 
655 	/*
656 	 * Check it is actually a disklabel we are looking at.
657 	 */
658 	if (nlp->d_magic != DISKMAGIC || nlp->d_magic2 != DISKMAGIC ||
659 	    dkcksum(nlp) != 0)
660 		return (EINVAL);
661 	/*
662 	 * For each partition that we think is open,
663 	 */
664 	while ((i = ffs((long)openmask)) != 0) {
665 		i--;
666 		/*
667 	 	 * Check it is not changing....
668 	 	 */
669 		openmask &= ~(1 << i);
670 		if (nlp->d_npartitions <= i)
671 			return (EBUSY);
672 		opp = &olp->d_partitions[i];
673 		npp = &nlp->d_partitions[i];
674 		if (npp->p_offset != opp->p_offset || npp->p_size < opp->p_size)
675 			return (EBUSY);
676 		/*
677 		 * Copy internally-set partition information
678 		 * if new label doesn't include it.		XXX
679 		 * (If we are using it then we had better stay the same type)
680 		 * This is possibly dubious, as someone else noted (XXX)
681 		 */
682 		if (npp->p_fstype == FS_UNUSED && opp->p_fstype != FS_UNUSED) {
683 			npp->p_fstype = opp->p_fstype;
684 			npp->p_fsize = opp->p_fsize;
685 			npp->p_frag = opp->p_frag;
686 			npp->p_cpg = opp->p_cpg;
687 		}
688 	}
689  	nlp->d_checksum = 0;
690  	nlp->d_checksum = dkcksum(nlp);
691 	*olp = *nlp;
692 	return (0);
693 }
694 
695 /*
696  * Write disk label back to device after modification.
697  */
698 int
699 writedisklabel(dev_t dev, struct disklabel *lp)
700 {
701 	struct buf *bp;
702 	struct disklabel *dlp;
703 	int error = 0;
704 
705 	if (lp->d_partitions[RAW_PART].p_offset != 0)
706 		return (EXDEV);			/* not quite right */
707 	bp = geteblk((int)lp->d_secsize);
708 	bp->b_dev = dkmodpart(dev, RAW_PART);
709 	bp->b_blkno = LABELSECTOR * ((int)lp->d_secsize/DEV_BSIZE);
710 	bp->b_bcount = lp->d_secsize;
711 #if 1
712 	/*
713 	 * We read the label first to see if it's there,
714 	 * in which case we will put ours at the same offset into the block..
715 	 * (I think this is stupid [Julian])
716 	 * Note that you can't write a label out over a corrupted label!
717 	 * (also stupid.. how do you write the first one? by raw writes?)
718 	 */
719 	bp->b_flags &= ~B_INVAL;
720 	bp->b_flags |= B_READ;
721 	BUF_STRATEGY(bp, 1);
722 	error = biowait(bp);
723 	if (error)
724 		goto done;
725 	for (dlp = (struct disklabel *)bp->b_data;
726 	    dlp <= (struct disklabel *)
727 	      ((char *)bp->b_data + lp->d_secsize - sizeof(*dlp));
728 	    dlp = (struct disklabel *)((char *)dlp + sizeof(long))) {
729 		if (dlp->d_magic == DISKMAGIC && dlp->d_magic2 == DISKMAGIC &&
730 		    dkcksum(dlp) == 0) {
731 			*dlp = *lp;
732 			bp->b_flags &= ~(B_DONE | B_READ);
733 			bp->b_flags |= B_WRITE;
734 			bp->b_dev = dkmodpart(dev, RAW_PART);
735 #ifdef __alpha__
736 			alpha_fix_srm_checksum(bp);
737 #endif
738 			BUF_STRATEGY(bp, 1);
739 			error = biowait(bp);
740 			goto done;
741 		}
742 	}
743 	error = ESRCH;
744 done:
745 #else
746 	bzero(bp->b_data, lp->d_secsize);
747 	dlp = (struct disklabel *)bp->b_data;
748 	*dlp = *lp;
749 	bp->b_flags &= ~B_INVAL;
750 	bp->b_flags |= B_WRITE;
751 	BUF_STRATEGY(bp, 1);
752 	error = biowait(bp);
753 #endif
754 	bp->b_flags |= B_INVAL | B_AGE;
755 	brelse(bp);
756 	return (error);
757 }
758 
759 /*
760  * Disk error is the preface to plaintive error messages
761  * about failing disk transfers.  It prints messages of the form
762 
763 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
764 
765  * if the offset of the error in the transfer and a disk label
766  * are both available.  blkdone should be -1 if the position of the error
767  * is unknown; the disklabel pointer may be null from drivers that have not
768  * been converted to use them.  The message is printed with printf
769  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
770  * The message should be completed (with at least a newline) with printf
771  * or addlog, respectively.  There is no trailing space.
772  */
773 void
774 diskerr(struct buf *bp, dev_t dev, char *what, int pri,
775 	int blkdone, struct disklabel *lp)
776 {
777 	int unit = dkunit(dev);
778 	int slice = dkslice(dev);
779 	int part = dkpart(dev);
780 	char partname[2];
781 	char *sname;
782 	daddr_t sn;
783 
784 	sname = dsname(dev, unit, slice, part, partname);
785 	printf("%s%s: %s %sing fsbn ", sname, partname, what,
786 	      bp->b_flags & B_READ ? "read" : "writ");
787 	sn = bp->b_blkno;
788 	if (bp->b_bcount <= DEV_BSIZE) {
789 		printf("%ld", (long)sn);
790 	} else {
791 		if (blkdone >= 0) {
792 			sn += blkdone;
793 			printf("%ld of ", (long)sn);
794 		}
795 		printf("%ld-%ld", (long)bp->b_blkno,
796 		    (long)(bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE));
797 	}
798 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
799 #ifdef tahoe
800 		sn *= DEV_BSIZE / lp->d_secsize;		/* XXX */
801 #endif
802 		sn += lp->d_partitions[part].p_offset;
803 		/*
804 		 * XXX should add slice offset and not print the slice,
805 		 * but we don't know the slice pointer.
806 		 * XXX should print bp->b_pblkno so that this will work
807 		 * independent of slices, labels and bad sector remapping,
808 		 * but some drivers don't set bp->b_pblkno.
809 		 */
810 		printf(" (%s bn %ld; cn %ld", sname, (long)sn,
811 		    (long)(sn / lp->d_secpercyl));
812 		sn %= (long)lp->d_secpercyl;
813 		printf(" tn %ld sn %ld)", (long)(sn / lp->d_nsectors),
814 		    (long)(sn % lp->d_nsectors));
815 	}
816 }
817