xref: /dragonfly/sys/kern/subr_diskslice.c (revision 4a65f651)
1 /*-
2  * Copyright (c) 1994 Bruce D. Evans.
3  * All rights reserved.
4  *
5  * Copyright (c) 1990 The Regents of the University of California.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * William Jolitz.
10  *
11  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
12  * All rights reserved.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. All advertising materials mentioning features or use of this software
23  *    must display the following acknowledgement:
24  *	This product includes software developed by the University of
25  *	California, Berkeley and its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	from: @(#)wd.c	7.2 (Berkeley) 5/9/91
43  *	from: wd.c,v 1.55 1994/10/22 01:57:12 phk Exp $
44  *	from: @(#)ufs_disksubr.c	7.16 (Berkeley) 5/4/91
45  *	from: ufs_disksubr.c,v 1.8 1994/06/07 01:21:39 phk Exp $
46  * $FreeBSD: src/sys/kern/subr_diskslice.c,v 1.82.2.6 2001/07/24 09:49:41 dd Exp $
47  * $DragonFly: src/sys/kern/subr_diskslice.c,v 1.51 2008/08/29 20:08:36 dillon Exp $
48  */
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/buf.h>
53 #include <sys/conf.h>
54 #include <sys/disklabel.h>
55 #include <sys/disklabel32.h>
56 #include <sys/disklabel64.h>
57 #include <sys/diskslice.h>
58 #include <sys/disk.h>
59 #include <sys/diskmbr.h>
60 #include <sys/fcntl.h>
61 #include <sys/malloc.h>
62 #include <sys/stat.h>
63 #include <sys/syslog.h>
64 #include <sys/proc.h>
65 #include <sys/vnode.h>
66 #include <sys/device.h>
67 #include <sys/thread2.h>
68 
69 #include <vfs/ufs/dinode.h>	/* XXX used only for fs.h */
70 #include <vfs/ufs/fs.h>		/* XXX used only to get BBSIZE/SBSIZE */
71 #include <sys/devfs.h>
72 
73 static int  dsreadandsetlabel(cdev_t dev, u_int flags,
74 			   struct diskslices *ssp, struct diskslice *sp,
75 			   struct disk_info *info);
76 static void free_ds_label (struct diskslices *ssp, int slice);
77 static void set_ds_label (struct diskslices *ssp, int slice, disklabel_t lp,
78 			   disklabel_ops_t ops);
79 static void set_ds_wlabel (struct diskslices *ssp, int slice, int wlabel);
80 
81 /*
82  * Determine the size of the transfer, and make sure it is
83  * within the boundaries of the partition. Adjust transfer
84  * if needed, and signal errors or early completion.
85  *
86  * XXX TODO:
87  *	o Split buffers that are too big for the device.
88  *	o Check for overflow.
89  *	o Finish cleaning this up.
90  *
91  * This function returns 1 on success, 0 if transfer equates
92  * to EOF (end of disk) or -1 on failure.  The appropriate
93  * 'errno' value is also set in bp->b_error and bp->b_flags
94  * is marked with B_ERROR.
95  */
96 struct bio *
97 dscheck(cdev_t dev, struct bio *bio, struct diskslices *ssp)
98 {
99 	struct buf *bp = bio->bio_buf;
100 	struct bio *nbio;
101 	disklabel_t lp;
102 	disklabel_ops_t ops;
103 	long nsec;
104 	u_int64_t secno;
105 	u_int64_t endsecno;
106 	u_int64_t slicerel_secno;
107 	struct diskslice *sp;
108 	u_int32_t part;
109 	u_int32_t slice;
110 	int shift;
111 	int mask;
112 
113 	slice = dkslice(dev);
114 	part  = dkpart(dev);
115 
116 	if (bio->bio_offset < 0) {
117 		kprintf("dscheck(%s): negative bio_offset %lld\n",
118 			devtoname(dev), (long long)bio->bio_offset);
119 		goto bad;
120 	}
121 	if (slice >= ssp->dss_nslices) {
122 		kprintf("dscheck(%s): slice too large %d/%d\n",
123 			devtoname(dev), slice, ssp->dss_nslices);
124 		goto bad;
125 	}
126 	sp = &ssp->dss_slices[slice];
127 	/*
128 	 * Calculate secno and nsec
129 	 */
130 	if (ssp->dss_secmult == 1) {
131 		shift = DEV_BSHIFT;
132 		goto doshift;
133 	} else if (ssp->dss_secshift != -1) {
134 		shift = DEV_BSHIFT + ssp->dss_secshift;
135 doshift:
136 		mask = (1 << shift) - 1;
137 		if ((int)bp->b_bcount & mask)
138 			goto bad_bcount;
139 		if ((int)bio->bio_offset & mask)
140 			goto bad_blkno;
141 		secno = bio->bio_offset >> shift;
142 		nsec = bp->b_bcount >> shift;
143 	} else {
144 		if (bp->b_bcount % ssp->dss_secsize)
145 			goto bad_bcount;
146 		if (bio->bio_offset % ssp->dss_secsize)
147 			goto bad_blkno;
148 		secno = bio->bio_offset / ssp->dss_secsize;
149 		nsec = bp->b_bcount / ssp->dss_secsize;
150 	}
151 
152 	/*
153 	 * Calculate slice-relative sector number end slice-relative
154 	 * limit.
155 	 */
156 	if (slice == WHOLE_DISK_SLICE) {
157 		/*
158 		 * Labels have not been allowed on whole-disks for a while.
159 		 * This really puts the nail in the coffin.
160 		 *
161 		 * Accesses to the WHOLE_DISK_SLICE do not use a disklabel
162 		 * and partition numbers are special-cased.  Currently numbers
163 		 * less then 128 are not allowed.  Partition numbers >= 128
164 		 * are encoded in the high 8 bits of the 64 bit buffer offset
165 		 * and are fed directly through to the device with no
166 		 * further interpretation.  In particular, no sector
167 		 * translation interpretation should occur because the
168 		 * sector size for the special raw access may not be the
169 		 * same as the nominal sector size for the device.
170 		 */
171 		lp.opaque = NULL;
172 		if (part < 128) {
173 			kprintf("dscheck(%s): illegal partition number (%d) "
174 				"for WHOLE_DISK_SLICE access\n",
175 				devtoname(dev), part);
176 			goto bad;
177 		} else if (part != WHOLE_SLICE_PART) {
178 			nbio = push_bio(bio);
179 			nbio->bio_offset = bio->bio_offset |
180 					   (u_int64_t)part << 56;
181 			return(nbio);
182 		} else {
183 			/*
184 			 * If writing to the raw disk request a
185 			 * reprobe on the last close.
186 			 */
187 			if (bp->b_cmd == BUF_CMD_WRITE)
188 				sp->ds_flags |= DSF_REPROBE;
189 		}
190 
191 		/*
192 		 * sp->ds_size is for the whole disk in the WHOLE_DISK_SLICE,
193 		 * there are no reserved areas.
194 		 */
195 		endsecno = sp->ds_size;
196 		slicerel_secno = secno;
197 	} else if (part == WHOLE_SLICE_PART) {
198 		/*
199 		 * NOTE! opens on a whole-slice partition will not attempt
200 		 * to read a disklabel in, so there may not be an in-core
201 		 * disklabel even if there is one on the disk.
202 		 */
203 		endsecno = sp->ds_size;
204 		slicerel_secno = secno;
205 	} else if ((lp = sp->ds_label).opaque != NULL) {
206 		/*
207 		 * A label is present, extract the partition.  Snooping of
208 		 * the disklabel is not supported even if accessible.  Of
209 		 * course, the reserved area is still write protected.
210 		 */
211 		ops = sp->ds_ops;
212 		if (ops->op_getpartbounds(ssp, lp, part,
213 					  &slicerel_secno, &endsecno)) {
214 			kprintf("dscheck(%s): partition %d out of bounds\n",
215 				devtoname(dev), part);
216 			goto bad;
217 		}
218 		slicerel_secno += secno;
219 	} else {
220 		/*
221 		 * Attempt to access partition when no disklabel present
222 		 */
223 		kprintf("dscheck(%s): attempt to access non-existent partition\n",
224 			devtoname(dev));
225 		goto bad;
226 	}
227 
228 	/*
229 	 * Disallow writes to reserved areas unless ds_wlabel allows it.
230 	 * If the reserved area is written to request a reprobe of the
231 	 * disklabel when the slice is closed.
232 	 */
233 	if (slicerel_secno < sp->ds_reserved && nsec &&
234 	    bp->b_cmd == BUF_CMD_WRITE) {
235 		if (sp->ds_wlabel == 0) {
236 			bp->b_error = EROFS;
237 			goto error;
238 		}
239 		sp->ds_flags |= DSF_REPROBE;
240 	}
241 
242 	/*
243 	 * If we get here, bio_offset must be on a block boundary and
244 	 * the sector size must be a power of 2.
245 	 */
246 	if ((bio->bio_offset & (ssp->dss_secsize - 1)) ||
247 	    (ssp->dss_secsize ^ (ssp->dss_secsize - 1)) !=
248 	    ((ssp->dss_secsize << 1) - 1)) {
249 		kprintf("%s: invalid BIO offset, not sector aligned or"
250 			" invalid sector size (not power of 2) %08llx %d\n",
251 			devtoname(dev), (long long)bio->bio_offset,
252 			ssp->dss_secsize);
253 		goto bad;
254 	}
255 
256 	/*
257 	 * EOF handling
258 	 */
259 	if (secno + nsec > endsecno) {
260 		/*
261 		 * Return an error if beyond the end of the disk, or
262 		 * if B_BNOCLIP is set.  Tell the system that we do not
263 		 * need to keep the buffer around.
264 		 */
265 		if (secno > endsecno || (bp->b_flags & B_BNOCLIP))
266 			goto bad;
267 
268 		/*
269 		 * If exactly at end of disk, return an EOF.  Throw away
270 		 * the buffer contents, if any, by setting B_INVAL.
271 		 */
272 		if (secno == endsecno) {
273 			bp->b_resid = bp->b_bcount;
274 			bp->b_flags |= B_INVAL;
275 			goto done;
276 		}
277 
278 		/*
279 		 * Else truncate
280 		 */
281 		nsec = endsecno - secno;
282 		bp->b_bcount = nsec * ssp->dss_secsize;
283 	}
284 
285 	nbio = push_bio(bio);
286 	nbio->bio_offset = (off_t)(sp->ds_offset + slicerel_secno) *
287 			   ssp->dss_secsize;
288 	return (nbio);
289 
290 bad_bcount:
291 	kprintf(
292 	"dscheck(%s): b_bcount %d is not on a sector boundary (ssize %d)\n",
293 	    devtoname(dev), bp->b_bcount, ssp->dss_secsize);
294 	goto bad;
295 
296 bad_blkno:
297 	kprintf(
298 	"dscheck(%s): bio_offset %lld is not on a sector boundary (ssize %d)\n",
299 	    devtoname(dev), (long long)bio->bio_offset, ssp->dss_secsize);
300 bad:
301 	bp->b_error = EINVAL;
302 	/* fall through */
303 error:
304 	/*
305 	 * Terminate the I/O with a ranging error.  Since the buffer is
306 	 * either illegal or beyond the file EOF, mark it B_INVAL as well.
307 	 */
308 	bp->b_resid = bp->b_bcount;
309 	bp->b_flags |= B_ERROR | B_INVAL;
310 done:
311 	/*
312 	 * Caller must biodone() the originally passed bio if NULL is
313 	 * returned.
314 	 */
315 	return (NULL);
316 }
317 
318 /*
319  * dsclose() - close a cooked disk slice.
320  *
321  * WARNING!  The passed diskslices and related diskslice structures may
322  *	     be invalidated or replaced by this function, callers must
323  *	     reload from the disk structure for continued access.
324  */
325 void
326 dsclose(cdev_t dev, int mode, struct diskslices *ssp)
327 {
328 	u_int32_t part;
329 	u_int32_t slice;
330 	struct diskslice *sp;
331 
332 	slice = dkslice(dev);
333 	part  = dkpart(dev);
334 	if (slice < ssp->dss_nslices) {
335 		sp = &ssp->dss_slices[slice];
336 		dsclrmask(sp, part);
337 		if (sp->ds_flags & DSF_REPROBE) {
338 			sp->ds_flags &= ~DSF_REPROBE;
339 			if (slice == WHOLE_DISK_SLICE) {
340 				disk_msg_send_sync(DISK_DISK_REPROBE,
341 						   dev->si_disk, NULL);
342 				devfs_config();
343 			} else {
344 				disk_msg_send_sync(DISK_SLICE_REPROBE,
345 						   dev->si_disk, sp);
346 				devfs_config();
347 			}
348 			/* ssp and sp may both be invalid after reprobe */
349 		}
350 	}
351 }
352 
353 void
354 dsgone(struct diskslices **sspp)
355 {
356 	int slice;
357 	struct diskslice *sp;
358 	struct diskslices *ssp;
359 
360 	kprintf("dsgone is called... fear!\n");
361 
362 	for (slice = 0, ssp = *sspp; slice < ssp->dss_nslices; slice++) {
363 		sp = &ssp->dss_slices[slice];
364 		free_ds_label(ssp, slice);
365 	}
366 	kfree(ssp, M_DEVBUF);
367 	*sspp = NULL;
368 }
369 
370 /*
371  * For the "write" commands (DIOCSDINFO and DIOCWDINFO), this
372  * is subject to the same restriction as dsopen().
373  */
374 int
375 dsioctl(cdev_t dev, u_long cmd, caddr_t data, int flags,
376 	struct diskslices **sspp, struct disk_info *info)
377 {
378 	int error;
379 	disklabel_t lp;
380 	disklabel_t lptmp;
381 	disklabel_ops_t ops;
382 	int old_wlabel;
383 	u_int32_t openmask[DKMAXPARTITIONS/(sizeof(u_int32_t)*8)];
384 	int part;
385 	int slice;
386 	struct diskslice *sp;
387 	struct diskslices *ssp;
388 
389 	slice = dkslice(dev);
390 	part = dkpart(dev);
391 	ssp = *sspp;
392 	if (slice >= ssp->dss_nslices)
393 		return (EINVAL);
394 	sp = &ssp->dss_slices[slice];
395 	lp = sp->ds_label;
396 	ops = sp->ds_ops;	/* may be NULL if no label */
397 
398 	switch (cmd) {
399 	case DIOCGDVIRGIN32:
400 		ops = &disklabel32_ops;
401 		/* fall through */
402 	case DIOCGDVIRGIN64:
403 		if (cmd != DIOCGDVIRGIN32)
404 			ops = &disklabel64_ops;
405 		/*
406 		 * You can only retrieve a virgin disklabel on the whole
407 		 * disk slice or whole-slice partition.
408 		 */
409 		if (slice != WHOLE_DISK_SLICE &&
410 		    part != WHOLE_SLICE_PART) {
411 			return(EINVAL);
412 		}
413 
414 		lp.opaque = data;
415 		ops->op_makevirginlabel(lp, ssp, sp, info);
416 		return (0);
417 
418 	case DIOCGDINFO32:
419 	case DIOCGDINFO64:
420 		/*
421 		 * You can only retrieve a disklabel on the whole
422 		 * slice partition.
423 		 *
424 		 * We do not support labels directly on whole-disks
425 		 * any more (that is, disks without slices), unless the
426 		 * device driver has asked for a compatible label (e.g.
427 		 * for a CD) to allow booting off of storage that is
428 		 * otherwise unlabeled.
429 		 */
430 		error = 0;
431 		if (part != WHOLE_SLICE_PART)
432 			return(EINVAL);
433 		if (slice == WHOLE_DISK_SLICE &&
434 		    (info->d_dsflags & DSO_COMPATLABEL) == 0) {
435 			return (ENODEV);
436 		}
437 		if (sp->ds_label.opaque == NULL) {
438 			error = dsreadandsetlabel(dev, info->d_dsflags,
439 						  ssp, sp, info);
440 			ops = sp->ds_ops;	/* may be NULL */
441 		}
442 
443 		/*
444 		 * The type of label we found must match the type of
445 		 * label requested.
446 		 */
447 		if (error == 0 && IOCPARM_LEN(cmd) != ops->labelsize)
448 			error = ENOATTR;
449 		if (error == 0)
450 			bcopy(sp->ds_label.opaque, data, ops->labelsize);
451 		return (error);
452 
453 	case DIOCGPART:
454 		{
455 			struct partinfo *dpart = (void *)data;
456 
457 			/*
458 			 * The disk management layer may not have read the
459 			 * disklabel yet because simply opening a slice no
460 			 * longer 'probes' the disk that way.  Be sure we
461 			 * have tried.
462 			 *
463 			 * We ignore any error.
464 			 */
465 			if (sp->ds_label.opaque == NULL &&
466 			    part == WHOLE_SLICE_PART &&
467 			    slice != WHOLE_DISK_SLICE) {
468 				dsreadandsetlabel(dev, info->d_dsflags,
469 						  ssp, sp, info);
470 				ops = sp->ds_ops;	/* may be NULL */
471 			}
472 
473 			bzero(dpart, sizeof(*dpart));
474 			dpart->media_offset   = (u_int64_t)sp->ds_offset *
475 						info->d_media_blksize;
476 			dpart->media_size     = (u_int64_t)sp->ds_size *
477 						info->d_media_blksize;
478 			dpart->media_blocks   = sp->ds_size;
479 			dpart->media_blksize  = info->d_media_blksize;
480 			dpart->reserved_blocks= sp->ds_reserved;
481 			dpart->fstype_uuid = sp->ds_type_uuid;
482 			dpart->storage_uuid = sp->ds_stor_uuid;
483 
484 			if (slice != WHOLE_DISK_SLICE &&
485 			    part != WHOLE_SLICE_PART) {
486 				u_int64_t start;
487 				u_int64_t blocks;
488 				if (lp.opaque == NULL)
489 					return(EINVAL);
490 				if (ops->op_getpartbounds(ssp, lp, part,
491 							  &start, &blocks)) {
492 					return(EINVAL);
493 				}
494 				ops->op_loadpartinfo(lp, part, dpart);
495 				dpart->media_offset += start *
496 						       info->d_media_blksize;
497 				dpart->media_size = blocks *
498 						    info->d_media_blksize;
499 				dpart->media_blocks = blocks;
500 
501 				/*
502 				 * partition starting sector (p_offset)
503 				 * requires slice's reserved areas to be
504 				 * adjusted.
505 				 */
506 				if (dpart->reserved_blocks > start)
507 					dpart->reserved_blocks -= start;
508 				else
509 					dpart->reserved_blocks = 0;
510 			}
511 
512 			/*
513 			 * Load remaining fields from the info structure
514 			 */
515 			dpart->d_nheads =	info->d_nheads;
516 			dpart->d_ncylinders =	info->d_ncylinders;
517 			dpart->d_secpertrack =	info->d_secpertrack;
518 			dpart->d_secpercyl =	info->d_secpercyl;
519 		}
520 		return (0);
521 
522 	case DIOCGSLICEINFO:
523 		bcopy(ssp, data, (char *)&ssp->dss_slices[ssp->dss_nslices] -
524 				 (char *)ssp);
525 		return (0);
526 
527 	case DIOCSDINFO32:
528 		ops = &disklabel32_ops;
529 		/* fall through */
530 	case DIOCSDINFO64:
531 		if (cmd != DIOCSDINFO32)
532 			ops = &disklabel64_ops;
533 		/*
534 		 * You can write a disklabel on the whole disk slice or
535 		 * whole-slice partition.
536 		 */
537 		if (slice != WHOLE_DISK_SLICE &&
538 		    part != WHOLE_SLICE_PART) {
539 			return(EINVAL);
540 		}
541 
542 		/*
543 		 * We no longer support writing disklabels directly to media
544 		 * without there being a slice.  Keep this as a separate
545 		 * conditional.
546 		 */
547 		if (slice == WHOLE_DISK_SLICE)
548 			return (ENODEV);
549 		if (!(flags & FWRITE))
550 			return (EBADF);
551 
552 		/*
553 		 * If an existing label is present it must be the same
554 		 * type as the label being passed by the ioctl.
555 		 */
556 		if (sp->ds_label.opaque && sp->ds_ops != ops)
557 			return (ENOATTR);
558 
559 		/*
560 		 * Create a temporary copy of the existing label
561 		 * (if present) so setdisklabel can compare it against
562 		 * the new label.
563 		 */
564 		lp.opaque = kmalloc(ops->labelsize, M_DEVBUF, M_WAITOK);
565 		if (sp->ds_label.opaque == NULL)
566 			bzero(lp.opaque, ops->labelsize);
567 		else
568 			bcopy(sp->ds_label.opaque, lp.opaque, ops->labelsize);
569 		if (sp->ds_label.opaque == NULL) {
570 			bzero(openmask, sizeof(openmask));
571 		} else {
572 			bcopy(sp->ds_openmask, openmask, sizeof(openmask));
573 		}
574 		lptmp.opaque = data;
575 		error = ops->op_setdisklabel(lp, lptmp, ssp, sp, openmask);
576 		disk_msg_send_sync(DISK_SLICE_REPROBE, dev->si_disk, sp);
577 		devfs_config();
578 		if (error != 0) {
579 			kfree(lp.opaque, M_DEVBUF);
580 			return (error);
581 		}
582 		free_ds_label(ssp, slice);
583 		set_ds_label(ssp, slice, lp, ops);
584 		return (0);
585 
586 	case DIOCSYNCSLICEINFO:
587 		/*
588 		 * This ioctl can only be done on the whole disk
589 		 */
590 		if (slice != WHOLE_DISK_SLICE || part != WHOLE_SLICE_PART)
591 			return (EINVAL);
592 
593 		if (*(int *)data == 0) {
594 			for (slice = 0; slice < ssp->dss_nslices; slice++) {
595 				struct diskslice *ds = &ssp->dss_slices[slice];
596 
597 				switch(dscountmask(ds)) {
598 				case 0:
599 					break;
600 				case 1:
601 					if (slice != WHOLE_DISK_SLICE)
602 						return (EBUSY);
603 					if (!dschkmask(ds, RAW_PART))
604 						return (EBUSY);
605 					break;
606 				default:
607 					return (EBUSY);
608 				}
609 			}
610 		}
611 
612 		disk_msg_send_sync(DISK_DISK_REPROBE, dev->si_disk, NULL);
613 		devfs_config();
614 		return 0;
615 
616 	case DIOCWDINFO32:
617 	case DIOCWDINFO64:
618 		error = dsioctl(dev, ((cmd == DIOCWDINFO32) ?
619 					DIOCSDINFO32 : DIOCSDINFO64),
620 				data, flags, &ssp, info);
621 		if (error == 0 && sp->ds_label.opaque == NULL)
622 			error = EINVAL;
623 		if (part != WHOLE_SLICE_PART)
624 			error = EINVAL;
625 		if (error != 0)
626 			return (error);
627 
628 		/*
629 		 * Allow the reserved area to be written, reload ops
630 		 * because the DIOCSDINFO op above may have installed
631 		 * a new label type.
632 		 */
633 		ops = sp->ds_ops;
634 		old_wlabel = sp->ds_wlabel;
635 		set_ds_wlabel(ssp, slice, TRUE);
636 		error = ops->op_writedisklabel(dev, ssp, sp, sp->ds_label);
637 		disk_msg_send_sync(DISK_SLICE_REPROBE, dev->si_disk, sp);
638 		devfs_config();
639 		set_ds_wlabel(ssp, slice, old_wlabel);
640 		/* XXX should invalidate in-core label if write failed. */
641 		return (error);
642 
643 	case DIOCWLABEL:
644 		if (slice == WHOLE_DISK_SLICE)
645 			return (ENODEV);
646 		if (!(flags & FWRITE))
647 			return (EBADF);
648 		set_ds_wlabel(ssp, slice, *(int *)data != 0);
649 		return (0);
650 
651 	default:
652 		return (ENOIOCTL);
653 	}
654 }
655 
656 int
657 dsisopen(struct diskslices *ssp)
658 {
659 	int slice;
660 
661 	if (ssp == NULL)
662 		return (0);
663 	for (slice = 0; slice < ssp->dss_nslices; slice++) {
664 		if (dscountmask(&ssp->dss_slices[slice]))
665 			return (1);
666 	}
667 	return (0);
668 }
669 
670 /*
671  * Allocate a slices "struct" and initialize it to contain only an empty
672  * compatibility slice (pointing to itself), a whole disk slice (covering
673  * the disk as described by the label), and (nslices - BASE_SLICES) empty
674  * slices beginning at BASE_SLICE.
675  *
676  * Note that the compatibility slice is no longer really a compatibility
677  * slice.  It is slice 0 if a GPT label is present, and the dangerously
678  * dedicated slice if no slice table otherwise exists.  Else it is 0-sized.
679  */
680 struct diskslices *
681 dsmakeslicestruct(int nslices, struct disk_info *info)
682 {
683 	struct diskslice *sp;
684 	struct diskslices *ssp;
685 
686 	ssp = kmalloc(offsetof(struct diskslices, dss_slices) +
687 		     nslices * sizeof *sp, M_DEVBUF, M_WAITOK);
688 	ssp->dss_first_bsd_slice = COMPATIBILITY_SLICE;
689 	ssp->dss_nslices = nslices;
690 	ssp->dss_oflags = 0;
691 
692 	/*
693 	 * Figure out if we can use shifts or whether we have to
694 	 * use mod/multply to translate byte offsets into sector numbers.
695 	 */
696 	if ((info->d_media_blksize ^ (info->d_media_blksize - 1)) ==
697 	     (info->d_media_blksize << 1) - 1) {
698 		ssp->dss_secmult = info->d_media_blksize / DEV_BSIZE;
699 		if (ssp->dss_secmult & (ssp->dss_secmult - 1))
700 			ssp->dss_secshift = -1;
701 		else
702 			ssp->dss_secshift = ffs(ssp->dss_secmult) - 1;
703 	} else {
704 		ssp->dss_secmult = 0;
705 		ssp->dss_secshift = -1;
706 	}
707 	ssp->dss_secsize = info->d_media_blksize;
708 	sp = &ssp->dss_slices[0];
709 	bzero(sp, nslices * sizeof *sp);
710 	sp[WHOLE_DISK_SLICE].ds_size = info->d_media_blocks;
711 	return (ssp);
712 }
713 
714 char *
715 dsname(cdev_t dev, int unit, int slice, int part, char *partname)
716 {
717 	return dev->si_name;
718 }
719 
720 /*
721  * This should only be called when the unit is inactive and the strategy
722  * routine should not allow it to become active unless we call it.  Our
723  * strategy routine must be special to allow activity.
724  */
725 int
726 dsopen(cdev_t dev, int mode, u_int flags,
727        struct diskslices **sspp, struct disk_info *info)
728 {
729 	struct diskslice *sp;
730 	struct diskslices *ssp;
731 	int slice;
732 	int part;
733 
734 	ssp = *sspp;
735 	dev->si_bsize_phys = info->d_media_blksize;
736 	slice = dkslice(dev);
737 	part = dkpart(dev);
738 	sp = &ssp->dss_slices[slice];
739 	dssetmask(sp, part);
740 
741 	return 0;
742 }
743 
744 /*
745  * Attempt to read the disklabel.  If successful, store it in sp->ds_label.
746  *
747  * If we cannot read the disklabel and DSO_COMPATLABEL is set, we construct
748  * a fake label covering the whole disk.
749  */
750 static
751 int
752 dsreadandsetlabel(cdev_t dev, u_int flags,
753 		  struct diskslices *ssp, struct diskslice *sp,
754 		  struct disk_info *info)
755 {
756 	disklabel_t lp;
757 	disklabel_ops_t ops;
758 	const char *msg;
759 	const char *sname;
760 	char partname[2];
761 	int slice = dkslice(dev);
762 
763 	/*
764 	 * Probe the disklabel
765 	 */
766 	lp.opaque = NULL;
767 	sname = dsname(dev, dkunit(dev), slice, WHOLE_SLICE_PART, partname);
768 	ops = &disklabel32_ops;
769 	msg = ops->op_readdisklabel(dev, sp, &lp, info);
770 	if (msg && strcmp(msg, "no disk label") == 0) {
771 		ops = &disklabel64_ops;
772 		msg = disklabel64_ops.op_readdisklabel(dev, sp, &lp, info);
773 	}
774 
775 	/*
776 	 * If we failed and COMPATLABEL is set, create a dummy disklabel.
777 	 */
778 	if (msg != NULL && (flags & DSO_COMPATLABEL)) {
779 		msg = NULL;
780 		if (sp->ds_size >= 0x100000000ULL)
781 			ops = &disklabel64_ops;
782 		else
783 			ops = &disklabel32_ops;
784 		lp = ops->op_clone_label(info, sp);
785 	}
786 	if (msg != NULL) {
787 		if (sp->ds_type == DOSPTYP_386BSD /* XXX */)
788 			log(LOG_WARNING, "%s: cannot find label (%s)\n",
789 			    sname, msg);
790 		if (lp.opaque)
791 			kfree(lp.opaque, M_DEVBUF);
792 	} else {
793 		set_ds_label(ssp, slice, lp, ops);
794 		set_ds_wlabel(ssp, slice, FALSE);
795 	}
796 	return (msg ? EINVAL : 0);
797 }
798 
799 int64_t
800 dssize(cdev_t dev, struct diskslices **sspp)
801 {
802 	disklabel_t lp;
803 	disklabel_ops_t ops;
804 	int part;
805 	int slice;
806 	struct diskslices *ssp;
807 	u_int64_t start;
808 	u_int64_t blocks;
809 
810 	slice = dkslice(dev);
811 	part = dkpart(dev);
812 	ssp = *sspp;
813 	if (ssp == NULL || slice >= ssp->dss_nslices
814 	    || !dschkmask(&ssp->dss_slices[slice], part)) {
815 		if (dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred) != 0)
816 			return (-1);
817 		dev_dclose(dev, FREAD, S_IFCHR);
818 		ssp = *sspp;
819 	}
820 	lp = ssp->dss_slices[slice].ds_label;
821 	if (lp.opaque == NULL)
822 		return (-1);
823 	ops = ssp->dss_slices[slice].ds_ops;
824 	if (ops->op_getpartbounds(ssp, lp, part, &start, &blocks))
825 		return (-1);
826 	return ((int64_t)blocks);
827 }
828 
829 static void
830 free_ds_label(struct diskslices *ssp, int slice)
831 {
832 	struct diskslice *sp;
833 	disklabel_t lp;
834 
835 	sp = &ssp->dss_slices[slice];
836 	lp = sp->ds_label;
837 	if (lp.opaque != NULL) {
838 		kfree(lp.opaque, M_DEVBUF);
839 		lp.opaque = NULL;
840 		set_ds_label(ssp, slice, lp, NULL);
841 	}
842 }
843 
844 static void
845 set_ds_label(struct diskslices *ssp, int slice,
846 	     disklabel_t lp, disklabel_ops_t ops)
847 {
848 	struct diskslice *sp = &ssp->dss_slices[slice];
849 
850 	sp->ds_label = lp;
851 	sp->ds_ops = ops;
852 	if (lp.opaque && slice != WHOLE_DISK_SLICE)
853 		ops->op_adjust_label_reserved(ssp, slice, sp);
854 	else
855 		sp->ds_reserved = 0;
856 }
857 
858 static void
859 set_ds_wlabel(struct diskslices *ssp, int slice, int wlabel)
860 {
861 	ssp->dss_slices[slice].ds_wlabel = wlabel;
862 }
863 
864