xref: /openbsd-src/sys/kern/subr_disk.c (revision 2777ee89d0e541ec819d05abee114837837abbec)
1 /*	$OpenBSD: subr_disk.c,v 1.223 2016/02/08 22:12:51 krw Exp $	*/
2 /*	$NetBSD: subr_disk.c,v 1.17 1996/03/16 23:17:08 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1995 Jason R. Thorpe.  All rights reserved.
6  * Copyright (c) 1982, 1986, 1988, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
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. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/fcntl.h>
46 #include <sys/buf.h>
47 #include <sys/stat.h>
48 #include <sys/syslog.h>
49 #include <sys/device.h>
50 #include <sys/time.h>
51 #include <sys/disklabel.h>
52 #include <sys/conf.h>
53 #include <sys/lock.h>
54 #include <sys/disk.h>
55 #include <sys/reboot.h>
56 #include <sys/dkio.h>
57 #include <sys/vnode.h>
58 #include <sys/task.h>
59 #include <sys/stdint.h>
60 
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 
64 #include <net/if.h>
65 
66 #include <dev/rndvar.h>
67 #include <dev/cons.h>
68 
69 #include <lib/libz/zlib.h>
70 
71 #include "softraid.h"
72 
73 #ifdef DEBUG
74 #define DPRINTF(x...)	printf(x)
75 #else
76 #define DPRINTF(x...)
77 #endif
78 
79 /*
80  * A global list of all disks attached to the system.  May grow or
81  * shrink over time.
82  */
83 struct	disklist_head disklist;	/* TAILQ_HEAD */
84 int	disk_count;		/* number of drives in global disklist */
85 int	disk_change;		/* set if a disk has been attached/detached
86 				 * since last we looked at this variable. This
87 				 * is reset by hw_sysctl()
88 				 */
89 
90 u_char	bootduid[8];		/* DUID of boot disk. */
91 u_char	rootduid[8];		/* DUID of root disk. */
92 
93 /* softraid callback, do not use! */
94 void (*softraid_disk_attach)(struct disk *, int);
95 
96 void sr_map_root(void);
97 
98 struct disk_attach_task {
99 	struct task task;
100 	struct disk *dk;
101 };
102 
103 void disk_attach_callback(void *);
104 
105 int spoofgptlabel(struct buf *, void (*)(struct buf *), struct disklabel *);
106 
107 int gpt_chk_mbr(struct dos_partition *, u_int64_t);
108 int gpt_chk_hdr(struct gpt_header *, struct disklabel *);
109 int gpt_chk_parts(struct gpt_header *, struct gpt_partition *);
110 int gpt_get_fstype(struct uuid *);
111 
112 /*
113  * Compute checksum for disk label.
114  */
115 u_int
116 dkcksum(struct disklabel *lp)
117 {
118 	u_int16_t *start, *end;
119 	u_int16_t sum = 0;
120 
121 	start = (u_int16_t *)lp;
122 	end = (u_int16_t *)&lp->d_partitions[lp->d_npartitions];
123 	while (start < end)
124 		sum ^= *start++;
125 	return (sum);
126 }
127 
128 int
129 initdisklabel(struct disklabel *lp)
130 {
131 	int i;
132 
133 	/* minimal requirements for archetypal disk label */
134 	if (lp->d_secsize < DEV_BSIZE)
135 		lp->d_secsize = DEV_BSIZE;
136 	if (DL_GETDSIZE(lp) == 0)
137 		DL_SETDSIZE(lp, MAXDISKSIZE);
138 	if (lp->d_secpercyl == 0)
139 		return (ERANGE);
140 	lp->d_npartitions = MAXPARTITIONS;
141 	for (i = 0; i < RAW_PART; i++) {
142 		DL_SETPSIZE(&lp->d_partitions[i], 0);
143 		DL_SETPOFFSET(&lp->d_partitions[i], 0);
144 	}
145 	if (DL_GETPSIZE(&lp->d_partitions[RAW_PART]) == 0)
146 		DL_SETPSIZE(&lp->d_partitions[RAW_PART], DL_GETDSIZE(lp));
147 	DL_SETPOFFSET(&lp->d_partitions[RAW_PART], 0);
148 	DL_SETBSTART(lp, 0);
149 	DL_SETBEND(lp, DL_GETDSIZE(lp));
150 	lp->d_version = 1;
151 	lp->d_bbsize = 8192;
152 	lp->d_sbsize = 64*1024;			/* XXX ? */
153 	return (0);
154 }
155 
156 /*
157  * Check an incoming block to make sure it is a disklabel, convert it to
158  * a newer version if needed, etc etc.
159  */
160 int
161 checkdisklabel(void *rlp, struct disklabel *lp, u_int64_t boundstart,
162     u_int64_t boundend)
163 {
164 	struct disklabel *dlp = rlp;
165 	struct __partitionv0 *v0pp;
166 	struct partition *pp;
167 	u_int64_t disksize;
168 	int error = 0;
169 	int i;
170 
171 	if (dlp->d_magic != DISKMAGIC || dlp->d_magic2 != DISKMAGIC)
172 		error = ENOENT;	/* no disk label */
173 	else if (dlp->d_npartitions > MAXPARTITIONS)
174 		error = E2BIG;	/* too many partitions */
175 	else if (dlp->d_secpercyl == 0)
176 		error = EINVAL;	/* invalid label */
177 	else if (dlp->d_secsize == 0)
178 		error = ENOSPC;	/* disk too small */
179 	else if (dkcksum(dlp) != 0)
180 		error = EINVAL;	/* incorrect checksum */
181 
182 	if (error) {
183 		u_int16_t *start, *end, sum = 0;
184 
185 		/* If it is byte-swapped, attempt to convert it */
186 		if (swap32(dlp->d_magic) != DISKMAGIC ||
187 		    swap32(dlp->d_magic2) != DISKMAGIC ||
188 		    swap16(dlp->d_npartitions) > MAXPARTITIONS)
189 			return (error);
190 
191 		/*
192 		 * Need a byte-swap aware dkcksum variant
193 		 * inlined, because dkcksum uses a sub-field
194 		 */
195 		start = (u_int16_t *)dlp;
196 		end = (u_int16_t *)&dlp->d_partitions[
197 		    swap16(dlp->d_npartitions)];
198 		while (start < end)
199 			sum ^= *start++;
200 		if (sum != 0)
201 			return (error);
202 
203 		dlp->d_magic = swap32(dlp->d_magic);
204 		dlp->d_type = swap16(dlp->d_type);
205 
206 		/* d_typename and d_packname are strings */
207 
208 		dlp->d_secsize = swap32(dlp->d_secsize);
209 		dlp->d_nsectors = swap32(dlp->d_nsectors);
210 		dlp->d_ntracks = swap32(dlp->d_ntracks);
211 		dlp->d_ncylinders = swap32(dlp->d_ncylinders);
212 		dlp->d_secpercyl = swap32(dlp->d_secpercyl);
213 		dlp->d_secperunit = swap32(dlp->d_secperunit);
214 
215 		/* d_uid is a string */
216 
217 		dlp->d_acylinders = swap32(dlp->d_acylinders);
218 
219 		dlp->d_flags = swap32(dlp->d_flags);
220 
221 		for (i = 0; i < NDDATA; i++)
222 			dlp->d_drivedata[i] = swap32(dlp->d_drivedata[i]);
223 
224 		dlp->d_secperunith = swap16(dlp->d_secperunith);
225 		dlp->d_version = swap16(dlp->d_version);
226 
227 		for (i = 0; i < NSPARE; i++)
228 			dlp->d_spare[i] = swap32(dlp->d_spare[i]);
229 
230 		dlp->d_magic2 = swap32(dlp->d_magic2);
231 
232 		dlp->d_npartitions = swap16(dlp->d_npartitions);
233 		dlp->d_bbsize = swap32(dlp->d_bbsize);
234 		dlp->d_sbsize = swap32(dlp->d_sbsize);
235 
236 		for (i = 0; i < MAXPARTITIONS; i++) {
237 			pp = &dlp->d_partitions[i];
238 			pp->p_size = swap32(pp->p_size);
239 			pp->p_offset = swap32(pp->p_offset);
240 			if (dlp->d_version == 0) {
241 				v0pp = (struct __partitionv0 *)pp;
242 				v0pp->p_fsize = swap32(v0pp->p_fsize);
243 			} else {
244 				pp->p_offseth = swap16(pp->p_offseth);
245 				pp->p_sizeh = swap16(pp->p_sizeh);
246 			}
247 			pp->p_cpg = swap16(pp->p_cpg);
248 		}
249 
250 		dlp->d_checksum = 0;
251 		dlp->d_checksum = dkcksum(dlp);
252 		error = 0;
253 	}
254 
255 	/* XXX should verify lots of other fields and whine a lot */
256 
257 	/* Initial passed in lp contains the real disk size. */
258 	disksize = DL_GETDSIZE(lp);
259 
260 	if (lp != dlp)
261 		*lp = *dlp;
262 
263 	if (lp->d_version == 0) {
264 		lp->d_version = 1;
265 		lp->d_secperunith = 0;
266 
267 		v0pp = (struct __partitionv0 *)lp->d_partitions;
268 		pp = lp->d_partitions;
269 		for (i = 0; i < lp->d_npartitions; i++, pp++, v0pp++) {
270 			pp->p_fragblock = DISKLABELV1_FFS_FRAGBLOCK(v0pp->
271 			    p_fsize, v0pp->p_frag);
272 			pp->p_offseth = 0;
273 			pp->p_sizeh = 0;
274 		}
275 	}
276 
277 #ifdef DEBUG
278 	if (DL_GETDSIZE(lp) != disksize)
279 		printf("on-disk disklabel has incorrect disksize (%llu)\n",
280 		    DL_GETDSIZE(lp));
281 	if (DL_GETPSIZE(&lp->d_partitions[RAW_PART]) != disksize)
282 		printf("on-disk disklabel RAW_PART has incorrect size (%llu)\n",
283 		    DL_GETPSIZE(&lp->d_partitions[RAW_PART]));
284 	if (DL_GETPOFFSET(&lp->d_partitions[RAW_PART]) != 0)
285 		printf("on-disk disklabel RAW_PART offset != 0 (%llu)\n",
286 		    DL_GETPOFFSET(&lp->d_partitions[RAW_PART]));
287 #endif
288 	DL_SETDSIZE(lp, disksize);
289 	DL_SETPSIZE(&lp->d_partitions[RAW_PART], disksize);
290 	DL_SETPOFFSET(&lp->d_partitions[RAW_PART], 0);
291 	DL_SETBSTART(lp, boundstart);
292 	DL_SETBEND(lp, boundend < DL_GETDSIZE(lp) ? boundend : DL_GETDSIZE(lp));
293 
294 	lp->d_checksum = 0;
295 	lp->d_checksum = dkcksum(lp);
296 	return (0);
297 }
298 
299 /*
300  * Read a disk sector.
301  */
302 int
303 readdisksector(struct buf *bp, void (*strat)(struct buf *),
304     struct disklabel *lp, u_int64_t sector)
305 {
306 	bp->b_blkno = DL_SECTOBLK(lp, sector);
307 	bp->b_bcount = lp->d_secsize;
308 	bp->b_error = 0;
309 	CLR(bp->b_flags, B_READ | B_WRITE | B_DONE | B_ERROR);
310 	SET(bp->b_flags, B_BUSY | B_READ | B_RAW);
311 
312 	(*strat)(bp);
313 
314 	return (biowait(bp));
315 }
316 
317 /*
318  * If dos partition table requested, attempt to load it and
319  * find disklabel inside a DOS partition. Return buffer
320  * for use in signalling errors if requested.
321  *
322  * We would like to check if each MBR has a valid BOOT_MAGIC, but
323  * we cannot because it doesn't always exist. So.. we assume the
324  * MBR is valid.
325  */
326 int
327 readdoslabel(struct buf *bp, void (*strat)(struct buf *),
328     struct disklabel *lp, daddr_t *partoffp, int spoofonly)
329 {
330 	struct disklabel *gptlp;
331 	u_int64_t dospartoff = 0, dospartend = DL_GETBEND(lp);
332 	int i, ourpart = -1, wander = 1, n = 0, loop = 0, offset;
333 	struct dos_partition dp[NDOSPART], *dp2;
334 	u_int64_t sector = DOSBBSECTOR;
335 	u_int32_t extoff = 0;
336 	int error;
337 
338 	if (lp->d_secpercyl == 0)
339 		return (EINVAL);	/* invalid label */
340 	if (lp->d_secsize == 0)
341 		return (ENOSPC);	/* disk too small */
342 
343 	/* do DOS partitions in the process of getting disklabel? */
344 
345 	/*
346 	 * Read dos partition table, follow extended partitions.
347 	 * Map the partitions to disklabel entries i-p
348 	 */
349 	while (wander && loop < DOS_MAXEBR) {
350 		loop++;
351 		wander = 0;
352 		if (sector < extoff)
353 			sector = extoff;
354 
355 		/* read MBR/EBR */
356 		error = readdisksector(bp, strat, lp, sector);
357 		if (error) {
358 /*wrong*/		if (partoffp)
359 /*wrong*/			*partoffp = -1;
360 			return (error);
361 		}
362 
363 		bcopy(bp->b_data + DOSPARTOFF, dp, sizeof(dp));
364 
365 		if (n == 0 && sector == DOSBBSECTOR) {
366 			u_int16_t mbrtest;
367 
368 			/* Check the end of sector marker. */
369 			mbrtest = ((bp->b_data[510] << 8) & 0xff00) |
370 			    (bp->b_data[511] & 0xff);
371 			if (mbrtest != 0x55aa)
372 				goto notmbr;
373 
374 			if (gpt_chk_mbr(dp, DL_GETDSIZE(lp)) != 0)
375 				goto notgpt;
376 
377 			gptlp = malloc(sizeof(struct disklabel), M_DEVBUF,
378 			    M_NOWAIT);
379 			if (gptlp == NULL)
380 				return (ENOMEM);
381 			*gptlp = *lp;
382 			error = spoofgptlabel(bp, strat, gptlp);
383 			if (error == 0) {
384 				dospartoff = DL_GETBSTART(gptlp);
385 				dospartend = DL_GETBEND(gptlp);
386 				if (partoffp) {
387 					if (dospartoff == 0)
388 						return (ENXIO);
389 					else
390 						goto notfat;
391 				}
392 				*lp = *gptlp;
393 				free(gptlp, M_DEVBUF,
394 				    sizeof(struct disklabel));
395 				goto notfat;
396 			} else {
397 				free(gptlp, M_DEVBUF,
398 				    sizeof(struct disklabel));
399 				goto notmbr;
400 			}
401 		}
402 
403 notgpt:
404 		if (ourpart == -1) {
405 			/* Search for our MBR partition */
406 			for (dp2=dp, i=0; i < NDOSPART && ourpart == -1;
407 			    i++, dp2++)
408 				if (letoh32(dp2->dp_size) &&
409 				    dp2->dp_typ == DOSPTYP_OPENBSD)
410 					ourpart = i;
411 			if (ourpart == -1)
412 				goto donot;
413 			/*
414 			 * This is our MBR partition. need sector
415 			 * address for SCSI/IDE, cylinder for
416 			 * ESDI/ST506/RLL
417 			 */
418 			dp2 = &dp[ourpart];
419 			dospartoff = letoh32(dp2->dp_start) + sector;
420 			dospartend = dospartoff + letoh32(dp2->dp_size);
421 
422 			/*
423 			 * Record the OpenBSD partition's placement (in
424 			 * 512-byte blocks!) for the caller. No need to
425 			 * finish spoofing.
426 			 */
427 			if (partoffp) {
428 				*partoffp = DL_SECTOBLK(lp, dospartoff);
429 				return (0);
430 			}
431 
432 			if (lp->d_ntracks == 0)
433 				lp->d_ntracks = dp2->dp_ehd + 1;
434 			if (lp->d_nsectors == 0)
435 				lp->d_nsectors = DPSECT(dp2->dp_esect);
436 			if (lp->d_secpercyl == 0)
437 				lp->d_secpercyl = lp->d_ntracks *
438 				    lp->d_nsectors;
439 		}
440 donot:
441 		/*
442 		 * In case the disklabel read below fails, we want to
443 		 * provide a fake label in i-p.
444 		 */
445 		for (dp2=dp, i=0; i < NDOSPART; i++, dp2++) {
446 			struct partition *pp;
447 			u_int8_t fstype;
448 
449 			if (dp2->dp_typ == DOSPTYP_OPENBSD ||
450 			    dp2->dp_typ == DOSPTYP_EFI)
451 				continue;
452 			if (letoh32(dp2->dp_size) > DL_GETDSIZE(lp))
453 				continue;
454 			if (letoh32(dp2->dp_start) > DL_GETDSIZE(lp))
455 				continue;
456 			if (letoh32(dp2->dp_size) == 0)
457 				continue;
458 
459 			switch (dp2->dp_typ) {
460 			case DOSPTYP_UNUSED:
461 				fstype = FS_UNUSED;
462 				break;
463 
464 			case DOSPTYP_LINUX:
465 				fstype = FS_EXT2FS;
466 				break;
467 
468 			case DOSPTYP_NTFS:
469 				fstype = FS_NTFS;
470 				break;
471 
472 			case DOSPTYP_EFISYS:
473 			case DOSPTYP_FAT12:
474 			case DOSPTYP_FAT16S:
475 			case DOSPTYP_FAT16B:
476 			case DOSPTYP_FAT16L:
477 			case DOSPTYP_FAT32:
478 			case DOSPTYP_FAT32L:
479 				fstype = FS_MSDOS;
480 				break;
481 			case DOSPTYP_EXTEND:
482 			case DOSPTYP_EXTENDL:
483 				sector = letoh32(dp2->dp_start) + extoff;
484 				if (!extoff) {
485 					extoff = letoh32(dp2->dp_start);
486 					sector = 0;
487 				}
488 				wander = 1;
489 				continue;
490 				break;
491 			default:
492 				fstype = FS_OTHER;
493 				break;
494 			}
495 
496 			/*
497 			 * Don't set fstype/offset/size when just looking for
498 			 * the offset of the OpenBSD partition. It would
499 			 * invalidate the disklabel checksum!
500 			 *
501 			 * Don't try to spoof more than 8 partitions, i.e.
502 			 * 'i' -'p'.
503 			 */
504 			if (partoffp || n >= 8)
505 				continue;
506 
507 			pp = &lp->d_partitions[8+n];
508 			n++;
509 			pp->p_fstype = fstype;
510 			if (letoh32(dp2->dp_start))
511 				DL_SETPOFFSET(pp,
512 				    letoh32(dp2->dp_start) + sector);
513 			DL_SETPSIZE(pp, letoh32(dp2->dp_size));
514 		}
515 	}
516 
517 notmbr:
518 	if (n == 0 && sector == DOSBBSECTOR && ourpart == -1) {
519 		u_int16_t fattest;
520 
521 		/* Check for a valid initial jmp instruction. */
522 		switch ((u_int8_t)bp->b_data[0]) {
523 		case 0xeb:
524 			/*
525 			 * Two-byte jmp instruction. The 2nd byte is the number
526 			 * of bytes to jmp and the 3rd byte must be a NOP.
527 			 */
528 			if ((u_int8_t)bp->b_data[2] != 0x90)
529 				goto notfat;
530 			break;
531 		case 0xe9:
532 			/*
533 			 * Three-byte jmp instruction. The next two bytes are a
534 			 * little-endian 16 bit value.
535 			 */
536 			break;
537 		default:
538 			goto notfat;
539 			break;
540 		}
541 
542 		/* Check for a valid bytes per sector value. */
543 		fattest = ((bp->b_data[12] << 8) & 0xff00) |
544 		    (bp->b_data[11] & 0xff);
545 		if (fattest < 512 || fattest > 4096 || (fattest % 512 != 0))
546 			goto notfat;
547 
548 		if (partoffp)
549 			return (ENXIO);	/* No place for disklabel on FAT! */
550 
551 		DL_SETPSIZE(&lp->d_partitions['i' - 'a'],
552 		    DL_GETPSIZE(&lp->d_partitions[RAW_PART]));
553 		DL_SETPOFFSET(&lp->d_partitions['i' - 'a'], 0);
554 		lp->d_partitions['i' - 'a'].p_fstype = FS_MSDOS;
555 
556 		spoofonly = 1;	/* No disklabel to read from disk. */
557 	}
558 
559 notfat:
560 	/* record the OpenBSD partition's placement for the caller */
561 	if (partoffp)
562 		*partoffp = DL_SECTOBLK(lp, dospartoff);
563 	else {
564 		DL_SETBSTART(lp, dospartoff);
565 		DL_SETBEND(lp, (dospartend < DL_GETDSIZE(lp)) ? dospartend :
566 		    DL_GETDSIZE(lp));
567 	}
568 
569 	/* don't read the on-disk label if we are in spoofed-only mode */
570 	if (spoofonly)
571 		return (0);
572 
573 	error = readdisksector(bp, strat, lp, dospartoff +
574 	    DL_BLKTOSEC(lp, DOS_LABELSECTOR));
575 	if (error)
576 		return (bp->b_error);
577 
578 	offset = DL_BLKOFFSET(lp, DOS_LABELSECTOR);
579 	error = checkdisklabel(bp->b_data + offset, lp,
580 	    DL_GETBSTART((struct disklabel*)(bp->b_data+offset)),
581 	    DL_GETBEND((struct disklabel *)(bp->b_data+offset)));
582 
583 	return (error);
584 }
585 
586 /*
587  * Returns 0 if the MBR with the provided partition array is a GPT protective
588  * MBR, and returns 1 otherwise. A GPT protective MBR would have one and only
589  * one MBR partition, an EFI partition that either covers the whole disk or as
590  * much of it as is possible with a 32bit size field.
591  *
592  * NOTE: MS always uses a size of UINT32_MAX for the EFI partition!**
593  */
594 int
595 gpt_chk_mbr(struct dos_partition *dp, u_int64_t dsize)
596 {
597 	struct dos_partition *dp2;
598 	int efi, found, i;
599 	u_int32_t psize;
600 
601 	found = efi = 0;
602 	for (dp2=dp, i=0; i < NDOSPART; i++, dp2++) {
603 		if (dp2->dp_typ == DOSPTYP_UNUSED)
604 			continue;
605 		found++;
606 		if (dp2->dp_typ != DOSPTYP_EFI)
607 			continue;
608 		psize = letoh32(dp2->dp_size);
609 		if (psize == (dsize - 1) ||
610 		    psize == UINT32_MAX) {
611 			if (letoh32(dp2->dp_start) == 1)
612 				efi++;
613 		}
614 	}
615 	if (found == 1 && efi == 1)
616 		return (0);
617 
618 	return (1);
619 }
620 
621 int
622 gpt_chk_hdr(struct gpt_header *gh, struct disklabel *lp)
623 {
624 	uint64_t ghpartlba;
625 	uint64_t ghlbaend, ghlbastart;
626 	uint32_t orig_gh_csum;
627 	uint32_t ghsize, ghpartsize, ghpartspersec, ghpartnum;
628 
629 	if (letoh64(gh->gh_sig) != GPTSIGNATURE)
630 		return (EINVAL);
631 
632 	if (letoh32(gh->gh_rev) != GPTREVISION)
633 		return (EINVAL);
634 
635 	ghsize = letoh32(gh->gh_size);
636 	ghpartsize = letoh32(gh->gh_part_size);
637 	ghpartspersec = lp->d_secsize / ghpartsize;
638 	ghpartnum = letoh32(gh->gh_part_num);
639 	ghpartlba = letoh64(gh->gh_part_lba);
640 	ghlbaend = letoh64(gh->gh_lba_end);
641 	ghlbastart = letoh64(gh->gh_lba_start);
642 
643 	if (ghsize < GPTMINHDRSIZE || ghsize > sizeof(struct gpt_header))
644 		return (EINVAL);
645 
646 	orig_gh_csum = gh->gh_csum;
647 	gh->gh_csum = 0;
648 	gh->gh_csum = crc32(0, (unsigned char *)gh, ghsize);
649 
650 	if (orig_gh_csum != gh->gh_csum)
651 		return (EINVAL);
652 
653 	if (ghlbastart >= DL_GETDSIZE(lp) ||
654 	    ghlbaend >= DL_GETDSIZE(lp) ||
655 	    ghpartlba >= DL_GETDSIZE(lp))
656 		return (EINVAL);
657 
658 	/*
659 	* Size per partition entry shall be 128*(2**n) with n >= 0.
660 	* We don't support partition entries larger than block size.
661 	*/
662 	if (ghpartsize % GPTMINPARTSIZE || ghpartsize > lp->d_secsize
663 	    || ghpartspersec == 0) {
664 		DPRINTF("invalid partition size\n");
665 		return (EINVAL);
666 	}
667 
668 	/* XXX: we don't support multiples of GPTMINPARTSIZE yet */
669 	if (ghpartsize != GPTMINPARTSIZE) {
670 		DPRINTF("partition sizes larger than %d bytes are not "
671 		    "supported", GPTMINPARTSIZE);
672 		return (EINVAL);
673 	}
674 
675 	if (letoh64(gh->gh_lba_alt) >= DL_GETDSIZE(lp)) {
676 		DPRINTF("alternate header's position is bogus\n");
677 		return (EINVAL);
678 	}
679 
680 	return 0;
681 }
682 
683 int
684 gpt_chk_parts(struct gpt_header *gh, struct gpt_partition *gp)
685 {
686 	u_int32_t checksum;
687 	checksum = crc32(0, (unsigned char *)gp,
688 	    letoh32(gh->gh_part_num) * letoh32(gh->gh_part_size));
689 
690 	if (checksum != gh->gh_part_csum)
691 		return (EINVAL);
692 
693 	return 0;
694 }
695 
696 int
697 gpt_get_fstype(struct uuid *uuid_part)
698 {
699 	static int init = 0;
700 	static struct uuid uuid_openbsd, uuid_msdos, uuid_chromefs,
701 	    uuid_linux, uuid_hfs, uuid_unused, uuid_efi_system;
702 	static const uint8_t gpt_uuid_openbsd[] = GPT_UUID_OPENBSD;
703 	static const uint8_t gpt_uuid_msdos[] = GPT_UUID_MSDOS;
704 	static const uint8_t gpt_uuid_chromerootfs[] = GPT_UUID_CHROMEROOTFS;
705 	static const uint8_t gpt_uuid_linux[] = GPT_UUID_LINUX;
706 	static const uint8_t gpt_uuid_hfs[] = GPT_UUID_APPLE_HFS;
707 	static const uint8_t gpt_uuid_unused[] = GPT_UUID_UNUSED;
708 	static const uint8_t gpt_uuid_efi_system[] = GPT_UUID_EFI_SYSTEM;
709 
710 	if (init == 0) {
711 		uuid_dec_be(gpt_uuid_openbsd, &uuid_openbsd);
712 		uuid_dec_be(gpt_uuid_msdos, &uuid_msdos);
713 		uuid_dec_be(gpt_uuid_chromerootfs, &uuid_chromefs);
714 		uuid_dec_be(gpt_uuid_linux, &uuid_linux);
715 		uuid_dec_be(gpt_uuid_hfs, &uuid_hfs);
716 		uuid_dec_be(gpt_uuid_unused, &uuid_unused);
717 		uuid_dec_be(gpt_uuid_efi_system, &uuid_efi_system);
718 		init = 1;
719 	}
720 
721 	if (!memcmp(uuid_part, &uuid_unused, sizeof(struct uuid)))
722 		return FS_UNUSED;
723 	else if (!memcmp(uuid_part, &uuid_openbsd, sizeof(struct uuid)))
724 		return FS_BSDFFS;
725 	else if (!memcmp(uuid_part, &uuid_msdos, sizeof(struct uuid)))
726 		return FS_MSDOS;
727 	else if (!memcmp(uuid_part, &uuid_chromefs, sizeof(struct uuid)))
728 		return FS_EXT2FS;
729 	else if (!memcmp(uuid_part, &uuid_linux, sizeof(struct uuid)))
730 		return FS_EXT2FS;
731 	else if (!memcmp(uuid_part, &uuid_hfs, sizeof(struct uuid)))
732 		return FS_HFS;
733 	else if (!memcmp(uuid_part, &uuid_efi_system, sizeof(struct uuid)))
734 		return FS_MSDOS;
735 	else
736 		return FS_OTHER;
737 }
738 
739 /*
740  * Spoof a disklabel based on the GPT information on the disk.
741  */
742 int
743 spoofgptlabel(struct buf *bp, void (*strat)(struct buf *),
744     struct disklabel *lp)
745 {
746 	static const u_int8_t gpt_uuid_openbsd[] = GPT_UUID_OPENBSD;
747 	struct gpt_header gh;
748 	struct uuid uuid_part, uuid_openbsd;
749 	struct gpt_partition *gp, *gp_tmp;
750 	struct partition *pp;
751 	size_t gpsz;
752 	u_int64_t ghlbaend, ghlbastart, gptpartoff, gptpartend, sector;
753 	u_int64_t start, end;
754 	int i, altheader = 0, error, n;
755 	uint32_t ghpartnum;
756 
757 	uuid_dec_be(gpt_uuid_openbsd, &uuid_openbsd);
758 
759 	for (sector = GPTSECTOR; ; sector = DL_GETDSIZE(lp)-1, altheader = 1) {
760 		uint64_t ghpartlba;
761 		uint32_t ghpartsize;
762 		uint32_t ghpartspersec;
763 
764 		error = readdisksector(bp, strat, lp, sector);
765 		if (error) {
766 			DPRINTF("error reading from disk\n");
767 			return (error);
768 		}
769 
770 		bcopy(bp->b_data, &gh, sizeof(gh));
771 
772 		if (gpt_chk_hdr(&gh, lp)) {
773 			if (altheader) {
774 				DPRINTF("alternate header also broken\n");
775 				return (EINVAL);
776 			}
777 			continue;
778 		}
779 
780 		ghpartsize = letoh32(gh.gh_part_size);
781 		ghpartspersec = lp->d_secsize / ghpartsize;
782 		ghpartnum = letoh32(gh.gh_part_num);
783 		ghpartlba = letoh64(gh.gh_part_lba);
784 		ghlbaend = letoh64(gh.gh_lba_end);
785 		ghlbastart = letoh64(gh.gh_lba_start);
786 
787 		/* read GPT partition entry array */
788 		gp = mallocarray(ghpartnum, sizeof(struct gpt_partition),
789 		    M_DEVBUF, M_NOWAIT|M_ZERO);
790 		if (gp == NULL)
791 			return (ENOMEM);
792 		gpsz = ghpartnum * sizeof(struct gpt_partition);
793 
794 		/*
795 		* XXX:	Fails if # of partition entries is not a multiple of
796 		*	ghpartspersec.
797 		*/
798 		sector = ghpartlba;
799 		for (i = 0; i < ghpartnum / ghpartspersec; i++, sector++) {
800 			error = readdisksector(bp, strat, lp, sector);
801 			if (error) {
802 				free(gp, M_DEVBUF, gpsz);
803 				return (error);
804 			}
805 
806 			bcopy(bp->b_data, gp + i * ghpartspersec,
807 			    ghpartspersec * sizeof(struct gpt_partition));
808 		}
809 
810 		if (gpt_chk_parts(&gh, gp)) {
811 			free(gp, M_DEVBUF, gpsz);
812 			if (altheader) {
813 				DPRINTF("alternate partition entries are also "
814 				    "broken\n");
815 				return (EINVAL);
816 			}
817 			continue;
818 		}
819 		break;
820 	}
821 
822 	/* Find OpenBSD partition and spoof others along the way. */
823 	n = 0;
824 	gptpartoff = 0;
825 	gptpartend = DL_GETBEND(lp);
826 	for (gp_tmp = gp, i = 0; i < ghpartnum; gp_tmp++, i++) {
827 		start = letoh64(gp_tmp->gp_lba_start);
828 		end = letoh64(gp_tmp->gp_lba_end);
829 		if (start > end || start < ghlbastart || end > ghlbaend)
830 			continue; /* entry invalid */
831 
832 		uuid_dec_le(&gp_tmp->gp_type, &uuid_part);
833 		if (!memcmp(&uuid_part, &uuid_openbsd, sizeof(struct uuid))) {
834 			if (gptpartoff == 0) {
835 				gptpartoff = start;
836 				gptpartend = end + 1;
837 			}
838 			continue; /* Do *NOT* spoof OpenBSD partitions! */
839 		}
840 
841 		 /*
842 		 * Don't try to spoof more than 8 partitions, i.e.
843 		 * 'i' -'p'.
844 		 */
845 		if (n >= 8)
846 			continue;
847 
848 		pp = &lp->d_partitions[8+n];
849 		n++;
850 		pp->p_fstype = gpt_get_fstype(&uuid_part);
851 		DL_SETPOFFSET(pp, start);
852 		DL_SETPSIZE(pp, end - start + 1);
853 	}
854 
855 	free(gp, M_DEVBUF, gpsz);
856 
857 	DL_SETBSTART(lp, gptpartoff);
858 	DL_SETBEND(lp, (gptpartend < DL_GETDSIZE(lp)) ? gptpartend :
859 	    DL_GETDSIZE(lp));
860 
861 	return (0);
862 }
863 
864 /*
865  * Check new disk label for sensibility before setting it.
866  */
867 int
868 setdisklabel(struct disklabel *olp, struct disklabel *nlp, u_int openmask)
869 {
870 	struct partition *opp, *npp;
871 	struct disk *dk;
872 	u_int64_t uid;
873 	int i;
874 
875 	/* sanity clause */
876 	if (nlp->d_secpercyl == 0 || nlp->d_secsize == 0 ||
877 	    (nlp->d_secsize % DEV_BSIZE) != 0)
878 		return (EINVAL);
879 
880 	/* special case to allow disklabel to be invalidated */
881 	if (nlp->d_magic == 0xffffffff) {
882 		*olp = *nlp;
883 		return (0);
884 	}
885 
886 	if (nlp->d_magic != DISKMAGIC || nlp->d_magic2 != DISKMAGIC ||
887 	    dkcksum(nlp) != 0)
888 		return (EINVAL);
889 
890 	/* XXX missing check if other dos partitions will be overwritten */
891 
892 	for (i = 0; i < MAXPARTITIONS; i++) {
893 		opp = &olp->d_partitions[i];
894 		npp = &nlp->d_partitions[i];
895 		if ((openmask & (1 << i)) &&
896 		    (DL_GETPOFFSET(npp) != DL_GETPOFFSET(opp) ||
897 		    DL_GETPSIZE(npp) < DL_GETPSIZE(opp)))
898 			return (EBUSY);
899 		/*
900 		 * Copy internally-set partition information
901 		 * if new label doesn't include it.		XXX
902 		 */
903 		if (npp->p_fstype == FS_UNUSED && opp->p_fstype != FS_UNUSED) {
904 			npp->p_fragblock = opp->p_fragblock;
905 			npp->p_cpg = opp->p_cpg;
906 		}
907 	}
908 
909 	/* Generate a UID if the disklabel does not already have one. */
910 	uid = 0;
911 	if (memcmp(nlp->d_uid, &uid, sizeof(nlp->d_uid)) == 0) {
912 		do {
913 			arc4random_buf(nlp->d_uid, sizeof(nlp->d_uid));
914 			TAILQ_FOREACH(dk, &disklist, dk_link)
915 				if (dk->dk_label && memcmp(dk->dk_label->d_uid,
916 				    nlp->d_uid, sizeof(nlp->d_uid)) == 0)
917 					break;
918 		} while (dk != NULL &&
919 		    memcmp(nlp->d_uid, &uid, sizeof(nlp->d_uid)) == 0);
920 	}
921 
922 	/* Preserve the disk size and RAW_PART values. */
923 	DL_SETDSIZE(nlp, DL_GETDSIZE(olp));
924 	npp = &nlp->d_partitions[RAW_PART];
925 	DL_SETPOFFSET(npp, 0);
926 	DL_SETPSIZE(npp, DL_GETDSIZE(nlp));
927 
928 	nlp->d_checksum = 0;
929 	nlp->d_checksum = dkcksum(nlp);
930 	*olp = *nlp;
931 
932 	disk_change = 1;
933 
934 	return (0);
935 }
936 
937 /*
938  * Determine the size of the transfer, and make sure it is within the
939  * boundaries of the partition. Adjust transfer if needed, and signal errors or
940  * early completion.
941  */
942 int
943 bounds_check_with_label(struct buf *bp, struct disklabel *lp)
944 {
945 	struct partition *p = &lp->d_partitions[DISKPART(bp->b_dev)];
946 	daddr_t partblocks, sz;
947 
948 	/* Avoid division by zero, negative offsets, and negative sizes. */
949 	if (lp->d_secpercyl == 0 || bp->b_blkno < 0 || bp->b_bcount < 0)
950 		goto bad;
951 
952 	/* Ensure transfer is a whole number of aligned sectors. */
953 	if ((bp->b_blkno % DL_BLKSPERSEC(lp)) != 0 ||
954 	    (bp->b_bcount % lp->d_secsize) != 0)
955 		goto bad;
956 
957 	/* Ensure transfer starts within partition boundary. */
958 	partblocks = DL_SECTOBLK(lp, DL_GETPSIZE(p));
959 	if (bp->b_blkno > partblocks)
960 		goto bad;
961 
962 	/* If exactly at end of partition or null transfer, return EOF. */
963 	if (bp->b_blkno == partblocks || bp->b_bcount == 0)
964 		goto done;
965 
966 	/* Truncate request if it extends past the end of the partition. */
967 	sz = bp->b_bcount >> DEV_BSHIFT;
968 	if (sz > partblocks - bp->b_blkno) {
969 		sz = partblocks - bp->b_blkno;
970 		bp->b_bcount = sz << DEV_BSHIFT;
971 	}
972 
973 	return (0);
974 
975  bad:
976 	bp->b_error = EINVAL;
977 	bp->b_flags |= B_ERROR;
978  done:
979 	bp->b_resid = bp->b_bcount;
980 	return (-1);
981 }
982 
983 /*
984  * Disk error is the preface to plaintive error messages
985  * about failing disk transfers.  It prints messages of the form
986 
987 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
988 
989  * if the offset of the error in the transfer and a disk label
990  * are both available.  blkdone should be -1 if the position of the error
991  * is unknown; the disklabel pointer may be null from drivers that have not
992  * been converted to use them.  The message is printed with printf
993  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
994  * The message should be completed (with at least a newline) with printf
995  * or addlog, respectively.  There is no trailing space.
996  */
997 void
998 diskerr(struct buf *bp, char *dname, char *what, int pri, int blkdone,
999     struct disklabel *lp)
1000 {
1001 	int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
1002 	int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2)));
1003 	char partname = 'a' + part;
1004 	daddr_t sn;
1005 
1006 	if (pri != LOG_PRINTF) {
1007 		log(pri, "%s", "");
1008 		pr = addlog;
1009 	} else
1010 		pr = printf;
1011 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
1012 	    bp->b_flags & B_READ ? "read" : "writ");
1013 	sn = bp->b_blkno;
1014 	if (bp->b_bcount <= DEV_BSIZE)
1015 		(*pr)("%lld", (long long)sn);
1016 	else {
1017 		if (blkdone >= 0) {
1018 			sn += blkdone;
1019 			(*pr)("%lld of ", (long long)sn);
1020 		}
1021 		(*pr)("%lld-%lld", (long long)bp->b_blkno,
1022 		    (long long)(bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE));
1023 	}
1024 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
1025 		sn += DL_SECTOBLK(lp, DL_GETPOFFSET(&lp->d_partitions[part]));
1026 		(*pr)(" (%s%d bn %lld; cn %lld", dname, unit, (long long)sn,
1027 		    (long long)(sn / DL_SECTOBLK(lp, lp->d_secpercyl)));
1028 		sn %= DL_SECTOBLK(lp, lp->d_secpercyl);
1029 		(*pr)(" tn %lld sn %lld)",
1030 		    (long long)(sn / DL_SECTOBLK(lp, lp->d_nsectors)),
1031 		    (long long)(sn % DL_SECTOBLK(lp, lp->d_nsectors)));
1032 	}
1033 }
1034 
1035 /*
1036  * Initialize the disklist.  Called by main() before autoconfiguration.
1037  */
1038 void
1039 disk_init(void)
1040 {
1041 
1042 	TAILQ_INIT(&disklist);
1043 	disk_count = disk_change = 0;
1044 }
1045 
1046 int
1047 disk_construct(struct disk *diskp)
1048 {
1049 	rw_init(&diskp->dk_lock, "dklk");
1050 	mtx_init(&diskp->dk_mtx, IPL_BIO);
1051 
1052 	diskp->dk_flags |= DKF_CONSTRUCTED;
1053 
1054 	return (0);
1055 }
1056 
1057 /*
1058  * Attach a disk.
1059  */
1060 void
1061 disk_attach(struct device *dv, struct disk *diskp)
1062 {
1063 	int majdev;
1064 
1065 	if (!ISSET(diskp->dk_flags, DKF_CONSTRUCTED))
1066 		disk_construct(diskp);
1067 
1068 	/*
1069 	 * Allocate and initialize the disklabel structures.  Note that
1070 	 * it's not safe to sleep here, since we're probably going to be
1071 	 * called during autoconfiguration.
1072 	 */
1073 	diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF,
1074 	    M_NOWAIT|M_ZERO);
1075 	if (diskp->dk_label == NULL)
1076 		panic("disk_attach: can't allocate storage for disklabel");
1077 
1078 	/*
1079 	 * Set the attached timestamp.
1080 	 */
1081 	microuptime(&diskp->dk_attachtime);
1082 
1083 	/*
1084 	 * Link into the disklist.
1085 	 */
1086 	TAILQ_INSERT_TAIL(&disklist, diskp, dk_link);
1087 	++disk_count;
1088 	disk_change = 1;
1089 
1090 	/*
1091 	 * Store device structure and number for later use.
1092 	 */
1093 	diskp->dk_device = dv;
1094 	diskp->dk_devno = NODEV;
1095 	if (dv != NULL) {
1096 		majdev = findblkmajor(dv);
1097 		if (majdev >= 0)
1098 			diskp->dk_devno =
1099 			    MAKEDISKDEV(majdev, dv->dv_unit, RAW_PART);
1100 
1101 		if (diskp->dk_devno != NODEV) {
1102 			struct disk_attach_task *dat;
1103 
1104 			dat = malloc(sizeof(*dat), M_TEMP, M_WAITOK);
1105 
1106 			/* XXX: Assumes dk is part of the device softc. */
1107 			device_ref(dv);
1108 			dat->dk = diskp;
1109 
1110 			task_set(&dat->task, disk_attach_callback, dat);
1111 			task_add(systq, &dat->task);
1112 		}
1113 	}
1114 
1115 	if (softraid_disk_attach)
1116 		softraid_disk_attach(diskp, 1);
1117 }
1118 
1119 void
1120 disk_attach_callback(void *xdat)
1121 {
1122 	struct disk_attach_task *dat = xdat;
1123 	struct disk *dk = dat->dk;
1124 	char errbuf[100];
1125 	struct disklabel dl;
1126 	dev_t dev;
1127 
1128 	free(dat, M_TEMP, sizeof(*dat));
1129 
1130 	if (dk->dk_flags & (DKF_OPENED | DKF_NOLABELREAD))
1131 		goto done;
1132 
1133 	/* Read disklabel. */
1134 	dev = dk->dk_devno;
1135 	if (disk_readlabel(&dl, dk->dk_devno, errbuf, sizeof(errbuf)) == NULL) {
1136 		add_timer_randomness(dl.d_checksum);
1137 		dk->dk_flags |= DKF_LABELVALID;
1138 	}
1139 
1140 done:
1141 	dk->dk_flags |= DKF_OPENED;
1142 	device_unref(dk->dk_device);
1143 	wakeup(dk);
1144 }
1145 
1146 /*
1147  * Detach a disk.
1148  */
1149 void
1150 disk_detach(struct disk *diskp)
1151 {
1152 
1153 	if (softraid_disk_attach)
1154 		softraid_disk_attach(diskp, -1);
1155 
1156 	/*
1157 	 * Free the space used by the disklabel structures.
1158 	 */
1159 	free(diskp->dk_label, M_DEVBUF, sizeof(*diskp->dk_label));
1160 
1161 	/*
1162 	 * Remove from the disklist.
1163 	 */
1164 	TAILQ_REMOVE(&disklist, diskp, dk_link);
1165 	disk_change = 1;
1166 	if (--disk_count < 0)
1167 		panic("disk_detach: disk_count < 0");
1168 }
1169 
1170 int
1171 disk_openpart(struct disk *dk, int part, int fmt, int haslabel)
1172 {
1173 	KASSERT(part >= 0 && part < MAXPARTITIONS);
1174 
1175 	/* Unless opening the raw partition, check that the partition exists. */
1176 	if (part != RAW_PART && (!haslabel ||
1177 	    part >= dk->dk_label->d_npartitions ||
1178 	    dk->dk_label->d_partitions[part].p_fstype == FS_UNUSED))
1179 		return (ENXIO);
1180 
1181 	/* Ensure the partition doesn't get changed under our feet. */
1182 	switch (fmt) {
1183 	case S_IFCHR:
1184 		dk->dk_copenmask |= (1 << part);
1185 		break;
1186 	case S_IFBLK:
1187 		dk->dk_bopenmask |= (1 << part);
1188 		break;
1189 	}
1190 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
1191 
1192 	return (0);
1193 }
1194 
1195 void
1196 disk_closepart(struct disk *dk, int part, int fmt)
1197 {
1198 	KASSERT(part >= 0 && part < MAXPARTITIONS);
1199 
1200 	switch (fmt) {
1201 	case S_IFCHR:
1202 		dk->dk_copenmask &= ~(1 << part);
1203 		break;
1204 	case S_IFBLK:
1205 		dk->dk_bopenmask &= ~(1 << part);
1206 		break;
1207 	}
1208 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
1209 }
1210 
1211 void
1212 disk_gone(int (*open)(dev_t, int, int, struct proc *), int unit)
1213 {
1214 	int bmaj, cmaj, mn;
1215 
1216 	/* Locate the lowest minor number to be detached. */
1217 	mn = DISKMINOR(unit, 0);
1218 
1219 	for (bmaj = 0; bmaj < nblkdev; bmaj++)
1220 		if (bdevsw[bmaj].d_open == open)
1221 			vdevgone(bmaj, mn, mn + MAXPARTITIONS - 1, VBLK);
1222 	for (cmaj = 0; cmaj < nchrdev; cmaj++)
1223 		if (cdevsw[cmaj].d_open == open)
1224 			vdevgone(cmaj, mn, mn + MAXPARTITIONS - 1, VCHR);
1225 }
1226 
1227 /*
1228  * Increment a disk's busy counter.  If the counter is going from
1229  * 0 to 1, set the timestamp.
1230  */
1231 void
1232 disk_busy(struct disk *diskp)
1233 {
1234 
1235 	/*
1236 	 * XXX We'd like to use something as accurate as microtime(),
1237 	 * but that doesn't depend on the system TOD clock.
1238 	 */
1239 	mtx_enter(&diskp->dk_mtx);
1240 	if (diskp->dk_busy++ == 0)
1241 		microuptime(&diskp->dk_timestamp);
1242 	mtx_leave(&diskp->dk_mtx);
1243 }
1244 
1245 /*
1246  * Decrement a disk's busy counter, increment the byte count, total busy
1247  * time, and reset the timestamp.
1248  */
1249 void
1250 disk_unbusy(struct disk *diskp, long bcount, int read)
1251 {
1252 	struct timeval dv_time, diff_time;
1253 
1254 	mtx_enter(&diskp->dk_mtx);
1255 
1256 	if (diskp->dk_busy-- == 0)
1257 		printf("disk_unbusy: %s: dk_busy < 0\n", diskp->dk_name);
1258 
1259 	microuptime(&dv_time);
1260 
1261 	timersub(&dv_time, &diskp->dk_timestamp, &diff_time);
1262 	timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time);
1263 
1264 	diskp->dk_timestamp = dv_time;
1265 	if (bcount > 0) {
1266 		if (read) {
1267 			diskp->dk_rbytes += bcount;
1268 			diskp->dk_rxfer++;
1269 		} else {
1270 			diskp->dk_wbytes += bcount;
1271 			diskp->dk_wxfer++;
1272 		}
1273 	} else
1274 		diskp->dk_seek++;
1275 
1276 	mtx_leave(&diskp->dk_mtx);
1277 
1278 	add_disk_randomness(bcount ^ diff_time.tv_usec);
1279 }
1280 
1281 int
1282 disk_lock(struct disk *dk)
1283 {
1284 	return (rw_enter(&dk->dk_lock, RW_WRITE|RW_INTR));
1285 }
1286 
1287 void
1288 disk_lock_nointr(struct disk *dk)
1289 {
1290 	rw_enter_write(&dk->dk_lock);
1291 }
1292 
1293 void
1294 disk_unlock(struct disk *dk)
1295 {
1296 	rw_exit_write(&dk->dk_lock);
1297 }
1298 
1299 int
1300 dk_mountroot(void)
1301 {
1302 	char errbuf[100];
1303 	int part = DISKPART(rootdev);
1304 	int (*mountrootfn)(void);
1305 	struct disklabel dl;
1306 	char *error;
1307 
1308 	error = disk_readlabel(&dl, rootdev, errbuf, sizeof(errbuf));
1309 	if (error)
1310 		panic("%s", error);
1311 
1312 	if (DL_GETPSIZE(&dl.d_partitions[part]) == 0)
1313 		panic("root filesystem has size 0");
1314 	switch (dl.d_partitions[part].p_fstype) {
1315 #ifdef EXT2FS
1316 	case FS_EXT2FS:
1317 		{
1318 		extern int ext2fs_mountroot(void);
1319 		mountrootfn = ext2fs_mountroot;
1320 		}
1321 		break;
1322 #endif
1323 #ifdef FFS
1324 	case FS_BSDFFS:
1325 		{
1326 		extern int ffs_mountroot(void);
1327 		mountrootfn = ffs_mountroot;
1328 		}
1329 		break;
1330 #endif
1331 #ifdef CD9660
1332 	case FS_ISO9660:
1333 		{
1334 		extern int cd9660_mountroot(void);
1335 		mountrootfn = cd9660_mountroot;
1336 		}
1337 		break;
1338 #endif
1339 	default:
1340 #ifdef FFS
1341 		{
1342 		extern int ffs_mountroot(void);
1343 
1344 		printf("filesystem type %d not known.. assuming ffs\n",
1345 		    dl.d_partitions[part].p_fstype);
1346 		mountrootfn = ffs_mountroot;
1347 		}
1348 #else
1349 		panic("disk 0x%x filesystem type %d not known",
1350 		    rootdev, dl.d_partitions[part].p_fstype);
1351 #endif
1352 	}
1353 	return (*mountrootfn)();
1354 }
1355 
1356 struct device *
1357 getdisk(char *str, int len, int defpart, dev_t *devp)
1358 {
1359 	struct device *dv;
1360 
1361 	if ((dv = parsedisk(str, len, defpart, devp)) == NULL) {
1362 		printf("use one of: exit");
1363 		TAILQ_FOREACH(dv, &alldevs, dv_list) {
1364 			if (dv->dv_class == DV_DISK)
1365 				printf(" %s[a-p]", dv->dv_xname);
1366 #if defined(NFSCLIENT)
1367 			if (dv->dv_class == DV_IFNET)
1368 				printf(" %s", dv->dv_xname);
1369 #endif
1370 		}
1371 		printf("\n");
1372 	}
1373 	return (dv);
1374 }
1375 
1376 struct device *
1377 parsedisk(char *str, int len, int defpart, dev_t *devp)
1378 {
1379 	struct device *dv;
1380 	int majdev, part = defpart;
1381 	char c;
1382 
1383 	if (len == 0)
1384 		return (NULL);
1385 	c = str[len-1];
1386 	if (c >= 'a' && (c - 'a') < MAXPARTITIONS) {
1387 		part = c - 'a';
1388 		len -= 1;
1389 	}
1390 
1391 	TAILQ_FOREACH(dv, &alldevs, dv_list) {
1392 		if (dv->dv_class == DV_DISK &&
1393 		    strncmp(str, dv->dv_xname, len) == 0 &&
1394 		    dv->dv_xname[len] == '\0') {
1395 			majdev = findblkmajor(dv);
1396 			if (majdev < 0)
1397 				return NULL;
1398 			*devp = MAKEDISKDEV(majdev, dv->dv_unit, part);
1399 			break;
1400 		}
1401 #if defined(NFSCLIENT)
1402 		if (dv->dv_class == DV_IFNET &&
1403 		    strncmp(str, dv->dv_xname, len) == 0 &&
1404 		    dv->dv_xname[len] == '\0') {
1405 			*devp = NODEV;
1406 			break;
1407 		}
1408 #endif
1409 	}
1410 
1411 	return (dv);
1412 }
1413 
1414 void
1415 setroot(struct device *bootdv, int part, int exitflags)
1416 {
1417 	int majdev, unit, len, s, slept = 0;
1418 	struct swdevt *swp;
1419 	struct device *rootdv, *dv;
1420 	dev_t nrootdev, nswapdev = NODEV, temp = NODEV;
1421 	struct ifnet *ifp = NULL;
1422 	struct disk *dk;
1423 	u_char duid[8];
1424 	char buf[128];
1425 #if defined(NFSCLIENT)
1426 	extern char *nfsbootdevname;
1427 #endif
1428 
1429 	/* Ensure that all disk attach callbacks have completed. */
1430 	do {
1431 		TAILQ_FOREACH(dk, &disklist, dk_link) {
1432 			if (dk->dk_devno != NODEV &&
1433 			    (dk->dk_flags & DKF_OPENED) == 0) {
1434 				tsleep(dk, 0, "dkopen", hz);
1435 				slept++;
1436 				break;
1437 			}
1438 		}
1439 	} while (dk != NULL && slept < 5);
1440 
1441 	if (slept == 5) {
1442 		printf("disklabels not read:");
1443 		TAILQ_FOREACH(dk, &disklist, dk_link)
1444 			if (dk->dk_devno != NODEV &&
1445 			    (dk->dk_flags & DKF_OPENED) == 0)
1446 				printf(" %s", dk->dk_name);
1447 		printf("\n");
1448 	}
1449 
1450 	/* Locate DUID for boot disk if not already provided. */
1451 	memset(duid, 0, sizeof(duid));
1452 	if (memcmp(bootduid, duid, sizeof(bootduid)) == 0) {
1453 		TAILQ_FOREACH(dk, &disklist, dk_link)
1454 			if (dk->dk_device == bootdv)
1455 				break;
1456 		if (dk && (dk->dk_flags & DKF_LABELVALID))
1457 			bcopy(dk->dk_label->d_uid, bootduid, sizeof(bootduid));
1458 	}
1459 	bcopy(bootduid, rootduid, sizeof(rootduid));
1460 
1461 #if NSOFTRAID > 0
1462 	sr_map_root();
1463 #endif
1464 
1465 	/*
1466 	 * If `swap generic' and we couldn't determine boot device,
1467 	 * ask the user.
1468 	 */
1469 	dk = NULL;
1470 	if (mountroot == NULL && bootdv == NULL)
1471 		boothowto |= RB_ASKNAME;
1472 	if (boothowto & RB_ASKNAME) {
1473 		while (1) {
1474 			printf("root device");
1475 			if (bootdv != NULL) {
1476 				printf(" (default %s", bootdv->dv_xname);
1477 				if (bootdv->dv_class == DV_DISK)
1478 					printf("%c", 'a' + part);
1479 				printf(")");
1480 			}
1481 			printf(": ");
1482 			s = splhigh();
1483 			cnpollc(1);
1484 			len = getsn(buf, sizeof(buf));
1485 			cnpollc(0);
1486 			splx(s);
1487 			if (strcmp(buf, "exit") == 0)
1488 				reboot(exitflags);
1489 			if (len == 0 && bootdv != NULL) {
1490 				strlcpy(buf, bootdv->dv_xname, sizeof buf);
1491 				len = strlen(buf);
1492 			}
1493 			if (len > 0 && buf[len - 1] == '*') {
1494 				buf[--len] = '\0';
1495 				dv = getdisk(buf, len, part, &nrootdev);
1496 				if (dv != NULL) {
1497 					rootdv = dv;
1498 					nswapdev = nrootdev;
1499 					goto gotswap;
1500 				}
1501 			}
1502 			dv = getdisk(buf, len, part, &nrootdev);
1503 			if (dv != NULL) {
1504 				rootdv = dv;
1505 				break;
1506 			}
1507 		}
1508 
1509 		if (rootdv->dv_class == DV_IFNET)
1510 			goto gotswap;
1511 
1512 		/* try to build swap device out of new root device */
1513 		while (1) {
1514 			printf("swap device");
1515 			if (rootdv != NULL)
1516 				printf(" (default %s%s)", rootdv->dv_xname,
1517 				    rootdv->dv_class == DV_DISK ? "b" : "");
1518 			printf(": ");
1519 			s = splhigh();
1520 			cnpollc(1);
1521 			len = getsn(buf, sizeof(buf));
1522 			cnpollc(0);
1523 			splx(s);
1524 			if (strcmp(buf, "exit") == 0)
1525 				reboot(exitflags);
1526 			if (len == 0 && rootdv != NULL) {
1527 				switch (rootdv->dv_class) {
1528 				case DV_IFNET:
1529 					nswapdev = NODEV;
1530 					break;
1531 				case DV_DISK:
1532 					nswapdev = MAKEDISKDEV(major(nrootdev),
1533 					    DISKUNIT(nrootdev), 1);
1534 					if (nswapdev == nrootdev)
1535 						continue;
1536 					break;
1537 				default:
1538 					break;
1539 				}
1540 				break;
1541 			}
1542 			dv = getdisk(buf, len, 1, &nswapdev);
1543 			if (dv) {
1544 				if (dv->dv_class == DV_IFNET)
1545 					nswapdev = NODEV;
1546 				if (nswapdev == nrootdev)
1547 					continue;
1548 				break;
1549 			}
1550 		}
1551 gotswap:
1552 		rootdev = nrootdev;
1553 		dumpdev = nswapdev;
1554 		swdevt[0].sw_dev = nswapdev;
1555 		swdevt[1].sw_dev = NODEV;
1556 #if defined(NFSCLIENT)
1557 	} else if (mountroot == nfs_mountroot) {
1558 		rootdv = bootdv;
1559 		rootdev = dumpdev = swapdev = NODEV;
1560 #endif
1561 	} else if (mountroot == NULL && rootdev == NODEV) {
1562 		/*
1563 		 * `swap generic'
1564 		 */
1565 		rootdv = bootdv;
1566 
1567 		if (bootdv->dv_class == DV_DISK) {
1568 			memset(&duid, 0, sizeof(duid));
1569 			if (memcmp(rootduid, &duid, sizeof(rootduid)) != 0) {
1570 				TAILQ_FOREACH(dk, &disklist, dk_link)
1571 					if ((dk->dk_flags & DKF_LABELVALID) &&
1572 					    dk->dk_label && memcmp(dk->dk_label->d_uid,
1573 					    &rootduid, sizeof(rootduid)) == 0)
1574 						break;
1575 				if (dk == NULL)
1576 					panic("root device (%02hx%02hx%02hx%02hx"
1577 					    "%02hx%02hx%02hx%02hx) not found",
1578 					    rootduid[0], rootduid[1], rootduid[2],
1579 					    rootduid[3], rootduid[4], rootduid[5],
1580 					    rootduid[6], rootduid[7]);
1581 				rootdv = dk->dk_device;
1582 			}
1583 		}
1584 
1585 		majdev = findblkmajor(rootdv);
1586 		if (majdev >= 0) {
1587 			/*
1588 			 * Root and swap are on the disk.
1589 			 * Assume swap is on partition b.
1590 			 */
1591 			rootdev = MAKEDISKDEV(majdev, rootdv->dv_unit, part);
1592 			nswapdev = MAKEDISKDEV(majdev, rootdv->dv_unit, 1);
1593 		} else {
1594 			/*
1595 			 * Root and swap are on a net.
1596 			 */
1597 			nswapdev = NODEV;
1598 		}
1599 		dumpdev = nswapdev;
1600 		swdevt[0].sw_dev = nswapdev;
1601 		/* swdevt[1].sw_dev = NODEV; */
1602 	} else {
1603 		/* Completely pre-configured, but we want rootdv .. */
1604 		majdev = major(rootdev);
1605 		if (findblkname(majdev) == NULL)
1606 			return;
1607 		unit = DISKUNIT(rootdev);
1608 		part = DISKPART(rootdev);
1609 		snprintf(buf, sizeof buf, "%s%d%c",
1610 		    findblkname(majdev), unit, 'a' + part);
1611 		rootdv = parsedisk(buf, strlen(buf), 0, &nrootdev);
1612 		if (rootdv == NULL)
1613 			panic("root device (%s) not found", buf);
1614 	}
1615 
1616 	if (rootdv && rootdv == bootdv && rootdv->dv_class == DV_IFNET)
1617 		ifp = ifunit(rootdv->dv_xname);
1618 	else if (bootdv && bootdv->dv_class == DV_IFNET)
1619 		ifp = ifunit(bootdv->dv_xname);
1620 
1621 	if (ifp)
1622 		if_addgroup(ifp, "netboot");
1623 
1624 	switch (rootdv->dv_class) {
1625 #if defined(NFSCLIENT)
1626 	case DV_IFNET:
1627 		mountroot = nfs_mountroot;
1628 		nfsbootdevname = rootdv->dv_xname;
1629 		return;
1630 #endif
1631 	case DV_DISK:
1632 		mountroot = dk_mountroot;
1633 		part = DISKPART(rootdev);
1634 		break;
1635 	default:
1636 		printf("can't figure root, hope your kernel is right\n");
1637 		return;
1638 	}
1639 
1640 	printf("root on %s%c", rootdv->dv_xname, 'a' + part);
1641 
1642 	if (dk && dk->dk_device == rootdv)
1643 		printf(" (%02hx%02hx%02hx%02hx%02hx%02hx%02hx%02hx.%c)",
1644 		    rootduid[0], rootduid[1], rootduid[2], rootduid[3],
1645 		    rootduid[4], rootduid[5], rootduid[6], rootduid[7],
1646 		    'a' + part);
1647 
1648 	/*
1649 	 * Make the swap partition on the root drive the primary swap.
1650 	 */
1651 	for (swp = swdevt; swp->sw_dev != NODEV; swp++) {
1652 		if (major(rootdev) == major(swp->sw_dev) &&
1653 		    DISKUNIT(rootdev) == DISKUNIT(swp->sw_dev)) {
1654 			temp = swdevt[0].sw_dev;
1655 			swdevt[0].sw_dev = swp->sw_dev;
1656 			swp->sw_dev = temp;
1657 			break;
1658 		}
1659 	}
1660 	if (swp->sw_dev != NODEV) {
1661 		/*
1662 		 * If dumpdev was the same as the old primary swap device,
1663 		 * move it to the new primary swap device.
1664 		 */
1665 		if (temp == dumpdev)
1666 			dumpdev = swdevt[0].sw_dev;
1667 	}
1668 	if (swdevt[0].sw_dev != NODEV)
1669 		printf(" swap on %s%d%c", findblkname(major(swdevt[0].sw_dev)),
1670 		    DISKUNIT(swdevt[0].sw_dev),
1671 		    'a' + DISKPART(swdevt[0].sw_dev));
1672 	if (dumpdev != NODEV)
1673 		printf(" dump on %s%d%c", findblkname(major(dumpdev)),
1674 		    DISKUNIT(dumpdev), 'a' + DISKPART(dumpdev));
1675 	printf("\n");
1676 }
1677 
1678 extern struct nam2blk nam2blk[];
1679 
1680 int
1681 findblkmajor(struct device *dv)
1682 {
1683 	char buf[16], *p;
1684 	int i;
1685 
1686 	if (strlcpy(buf, dv->dv_xname, sizeof buf) >= sizeof buf)
1687 		return (-1);
1688 	for (p = buf; *p; p++)
1689 		if (*p >= '0' && *p <= '9')
1690 			*p = '\0';
1691 
1692 	for (i = 0; nam2blk[i].name; i++)
1693 		if (!strcmp(buf, nam2blk[i].name))
1694 			return (nam2blk[i].maj);
1695 	return (-1);
1696 }
1697 
1698 char *
1699 findblkname(int maj)
1700 {
1701 	int i;
1702 
1703 	for (i = 0; nam2blk[i].name; i++)
1704 		if (nam2blk[i].maj == maj)
1705 			return (nam2blk[i].name);
1706 	return (NULL);
1707 }
1708 
1709 char *
1710 disk_readlabel(struct disklabel *dl, dev_t dev, char *errbuf, size_t errsize)
1711 {
1712 	struct vnode *vn;
1713 	dev_t chrdev, rawdev;
1714 	int error;
1715 
1716 	chrdev = blktochr(dev);
1717 	rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(chrdev), RAW_PART);
1718 
1719 #ifdef DEBUG
1720 	printf("dev=0x%x chrdev=0x%x rawdev=0x%x\n", dev, chrdev, rawdev);
1721 #endif
1722 
1723 	if (cdevvp(rawdev, &vn)) {
1724 		snprintf(errbuf, errsize,
1725 		    "cannot obtain vnode for 0x%x/0x%x", dev, rawdev);
1726 		return (errbuf);
1727 	}
1728 
1729 	error = VOP_OPEN(vn, FREAD, NOCRED, curproc);
1730 	if (error) {
1731 		snprintf(errbuf, errsize,
1732 		    "cannot open disk, 0x%x/0x%x, error %d",
1733 		    dev, rawdev, error);
1734 		goto done;
1735 	}
1736 
1737 	error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)dl, FREAD, NOCRED, curproc);
1738 	if (error) {
1739 		snprintf(errbuf, errsize,
1740 		    "cannot read disk label, 0x%x/0x%x, error %d",
1741 		    dev, rawdev, error);
1742 	}
1743 done:
1744 	VOP_CLOSE(vn, FREAD, NOCRED, curproc);
1745 	vput(vn);
1746 	if (error)
1747 		return (errbuf);
1748 	return (NULL);
1749 }
1750 
1751 int
1752 disk_map(char *path, char *mappath, int size, int flags)
1753 {
1754 	struct disk *dk, *mdk;
1755 	u_char uid[8];
1756 	char c, part;
1757 	int i;
1758 
1759 	/*
1760 	 * Attempt to map a request for a disklabel UID to the correct device.
1761 	 * We should be supplied with a disklabel UID which has the following
1762 	 * format:
1763 	 *
1764 	 * [disklabel uid] . [partition]
1765 	 *
1766 	 * Alternatively, if the DM_OPENPART flag is set the disklabel UID can
1767 	 * based passed on its own.
1768 	 */
1769 
1770 	if (strchr(path, '/') != NULL)
1771 		return -1;
1772 
1773 	/* Verify that the device name is properly formed. */
1774 	if (!((strlen(path) == 16 && (flags & DM_OPENPART)) ||
1775 	    (strlen(path) == 18 && path[16] == '.')))
1776 		return -1;
1777 
1778 	/* Get partition. */
1779 	if (flags & DM_OPENPART)
1780 		part = 'a' + RAW_PART;
1781 	else
1782 		part = path[17];
1783 
1784 	if (part < 'a' || part >= 'a' + MAXPARTITIONS)
1785 		return -1;
1786 
1787 	/* Derive label UID. */
1788 	memset(uid, 0, sizeof(uid));
1789 	for (i = 0; i < 16; i++) {
1790 		c = path[i];
1791 		if (c >= '0' && c <= '9')
1792 			c -= '0';
1793 		else if (c >= 'a' && c <= 'f')
1794 			c -= ('a' - 10);
1795 		else
1796 			return -1;
1797 
1798 		uid[i / 2] <<= 4;
1799 		uid[i / 2] |= c & 0xf;
1800 	}
1801 
1802 	mdk = NULL;
1803 	TAILQ_FOREACH(dk, &disklist, dk_link) {
1804 		if ((dk->dk_flags & DKF_LABELVALID) && dk->dk_label &&
1805 		    memcmp(dk->dk_label->d_uid, uid,
1806 		    sizeof(dk->dk_label->d_uid)) == 0) {
1807 			/* Fail if there are duplicate UIDs! */
1808 			if (mdk != NULL)
1809 				return -1;
1810 			mdk = dk;
1811 		}
1812 	}
1813 
1814 	if (mdk == NULL || mdk->dk_name == NULL)
1815 		return -1;
1816 
1817 	snprintf(mappath, size, "/dev/%s%s%c",
1818 	    (flags & DM_OPENBLCK) ? "" : "r", mdk->dk_name, part);
1819 
1820 	return 0;
1821 }
1822 
1823 /*
1824  * Lookup a disk device and verify that it has completed attaching.
1825  */
1826 struct device *
1827 disk_lookup(struct cfdriver *cd, int unit)
1828 {
1829 	struct device *dv;
1830 	struct disk *dk;
1831 
1832 	dv = device_lookup(cd, unit);
1833 	if (dv == NULL)
1834 		return (NULL);
1835 
1836 	TAILQ_FOREACH(dk, &disklist, dk_link)
1837 		if (dk->dk_device == dv)
1838 			break;
1839 
1840 	if (dk == NULL) {
1841 		device_unref(dv);
1842 		return (NULL);
1843 	}
1844 
1845 	return (dv);
1846 }
1847