xref: /openbsd-src/sys/kern/subr_disk.c (revision f84b1df5a16cdd762c93854218de246e79975d3b)
1 /*	$OpenBSD: subr_disk.c,v 1.248 2022/01/02 17:26:14 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/cons.h>
67 
68 #include <lib/libz/zlib.h>
69 
70 #include "softraid.h"
71 
72 #ifdef DEBUG
73 #define DPRINTF(x...)	printf(x)
74 #else
75 #define DPRINTF(x...)
76 #endif
77 
78 /*
79  * A global list of all disks attached to the system.  May grow or
80  * shrink over time.
81  */
82 struct	disklist_head disklist;	/* TAILQ_HEAD */
83 int	disk_count;		/* number of drives in global disklist */
84 int	disk_change;		/* set if a disk has been attached/detached
85 				 * since last we looked at this variable. This
86 				 * is reset by hw_sysctl()
87 				 */
88 
89 #define DUID_SIZE 8
90 
91 u_char	bootduid[DUID_SIZE];	/* DUID of boot disk. */
92 u_char	rootduid[DUID_SIZE];	/* DUID of root disk. */
93 
94 /* softraid callback, do not use! */
95 void (*softraid_disk_attach)(struct disk *, int);
96 
97 void sr_map_root(void);
98 
99 struct disk_attach_task {
100 	struct task task;
101 	struct disk *dk;
102 };
103 
104 void disk_attach_callback(void *);
105 
106 int spoofgptlabel(struct buf *, void (*)(struct buf *), struct disklabel *);
107 
108 int gpt_chk_mbr(struct dos_partition *, uint64_t);
109 int gpt_chk_hdr(struct gpt_header *, struct disklabel *);
110 int gpt_chk_parts(struct gpt_header *, struct gpt_partition *);
111 int gpt_get_fstype(struct uuid *);
112 
113 int duid_equal(u_char *, u_char *);
114 
115 /*
116  * Compute checksum for disk label.
117  */
118 u_int
119 dkcksum(struct disklabel *lp)
120 {
121 	u_int16_t *start, *end;
122 	u_int16_t sum = 0;
123 
124 	start = (u_int16_t *)lp;
125 	end = (u_int16_t *)&lp->d_partitions[lp->d_npartitions];
126 	while (start < end)
127 		sum ^= *start++;
128 	return (sum);
129 }
130 
131 int
132 initdisklabel(struct disklabel *lp)
133 {
134 	int i;
135 
136 	/* minimal requirements for archetypal disk label */
137 	if (lp->d_secsize < DEV_BSIZE)
138 		lp->d_secsize = DEV_BSIZE;
139 	if (DL_GETDSIZE(lp) == 0)
140 		DL_SETDSIZE(lp, MAXDISKSIZE);
141 	if (lp->d_secpercyl == 0)
142 		return (ERANGE);
143 	lp->d_npartitions = MAXPARTITIONS;
144 	for (i = 0; i < RAW_PART; i++) {
145 		DL_SETPSIZE(&lp->d_partitions[i], 0);
146 		DL_SETPOFFSET(&lp->d_partitions[i], 0);
147 	}
148 	if (DL_GETPSIZE(&lp->d_partitions[RAW_PART]) == 0)
149 		DL_SETPSIZE(&lp->d_partitions[RAW_PART], DL_GETDSIZE(lp));
150 	DL_SETPOFFSET(&lp->d_partitions[RAW_PART], 0);
151 	DL_SETBSTART(lp, 0);
152 	DL_SETBEND(lp, DL_GETDSIZE(lp));
153 	lp->d_version = 1;
154 	lp->d_bbsize = 8192;
155 	lp->d_sbsize = 64*1024;			/* XXX ? */
156 	return (0);
157 }
158 
159 /*
160  * Check an incoming block to make sure it is a disklabel, convert it to
161  * a newer version if needed, etc etc.
162  */
163 int
164 checkdisklabel(void *rlp, struct disklabel *lp, u_int64_t boundstart,
165     u_int64_t boundend)
166 {
167 	struct disklabel *dlp = rlp;
168 	struct __partitionv0 *v0pp;
169 	struct partition *pp;
170 	u_int64_t disksize;
171 	int error = 0;
172 	int i;
173 
174 	if (dlp->d_magic != DISKMAGIC || dlp->d_magic2 != DISKMAGIC)
175 		error = ENOENT;	/* no disk label */
176 	else if (dlp->d_npartitions > MAXPARTITIONS)
177 		error = E2BIG;	/* too many partitions */
178 	else if (dlp->d_secpercyl == 0)
179 		error = EINVAL;	/* invalid label */
180 	else if (dlp->d_secsize == 0)
181 		error = ENOSPC;	/* disk too small */
182 	else if (dkcksum(dlp) != 0)
183 		error = EINVAL;	/* incorrect checksum */
184 
185 	if (error) {
186 		u_int16_t *start, *end, sum = 0;
187 
188 		/* If it is byte-swapped, attempt to convert it */
189 		if (swap32(dlp->d_magic) != DISKMAGIC ||
190 		    swap32(dlp->d_magic2) != DISKMAGIC ||
191 		    swap16(dlp->d_npartitions) > MAXPARTITIONS)
192 			return (error);
193 
194 		/*
195 		 * Need a byte-swap aware dkcksum variant
196 		 * inlined, because dkcksum uses a sub-field
197 		 */
198 		start = (u_int16_t *)dlp;
199 		end = (u_int16_t *)&dlp->d_partitions[
200 		    swap16(dlp->d_npartitions)];
201 		while (start < end)
202 			sum ^= *start++;
203 		if (sum != 0)
204 			return (error);
205 
206 		dlp->d_magic = swap32(dlp->d_magic);
207 		dlp->d_type = swap16(dlp->d_type);
208 
209 		/* d_typename and d_packname are strings */
210 
211 		dlp->d_secsize = swap32(dlp->d_secsize);
212 		dlp->d_nsectors = swap32(dlp->d_nsectors);
213 		dlp->d_ntracks = swap32(dlp->d_ntracks);
214 		dlp->d_ncylinders = swap32(dlp->d_ncylinders);
215 		dlp->d_secpercyl = swap32(dlp->d_secpercyl);
216 		dlp->d_secperunit = swap32(dlp->d_secperunit);
217 
218 		/* d_uid is a string */
219 
220 		dlp->d_acylinders = swap32(dlp->d_acylinders);
221 
222 		dlp->d_flags = swap32(dlp->d_flags);
223 
224 		for (i = 0; i < NDDATA; i++)
225 			dlp->d_drivedata[i] = swap32(dlp->d_drivedata[i]);
226 
227 		dlp->d_secperunith = swap16(dlp->d_secperunith);
228 		dlp->d_version = swap16(dlp->d_version);
229 
230 		for (i = 0; i < NSPARE; i++)
231 			dlp->d_spare[i] = swap32(dlp->d_spare[i]);
232 
233 		dlp->d_magic2 = swap32(dlp->d_magic2);
234 
235 		dlp->d_npartitions = swap16(dlp->d_npartitions);
236 		dlp->d_bbsize = swap32(dlp->d_bbsize);
237 		dlp->d_sbsize = swap32(dlp->d_sbsize);
238 
239 		for (i = 0; i < MAXPARTITIONS; i++) {
240 			pp = &dlp->d_partitions[i];
241 			pp->p_size = swap32(pp->p_size);
242 			pp->p_offset = swap32(pp->p_offset);
243 			if (dlp->d_version == 0) {
244 				v0pp = (struct __partitionv0 *)pp;
245 				v0pp->p_fsize = swap32(v0pp->p_fsize);
246 			} else {
247 				pp->p_offseth = swap16(pp->p_offseth);
248 				pp->p_sizeh = swap16(pp->p_sizeh);
249 			}
250 			pp->p_cpg = swap16(pp->p_cpg);
251 		}
252 
253 		dlp->d_checksum = 0;
254 		dlp->d_checksum = dkcksum(dlp);
255 		error = 0;
256 	}
257 
258 	/* XXX should verify lots of other fields and whine a lot */
259 
260 	/* Initial passed in lp contains the real disk size. */
261 	disksize = DL_GETDSIZE(lp);
262 
263 	if (lp != dlp)
264 		*lp = *dlp;
265 
266 	if (lp->d_version == 0) {
267 		lp->d_version = 1;
268 		lp->d_secperunith = 0;
269 
270 		v0pp = (struct __partitionv0 *)lp->d_partitions;
271 		pp = lp->d_partitions;
272 		for (i = 0; i < lp->d_npartitions; i++, pp++, v0pp++) {
273 			pp->p_fragblock = DISKLABELV1_FFS_FRAGBLOCK(v0pp->
274 			    p_fsize, v0pp->p_frag);
275 			pp->p_offseth = 0;
276 			pp->p_sizeh = 0;
277 		}
278 	}
279 
280 #ifdef DEBUG
281 	if (DL_GETDSIZE(lp) != disksize)
282 		printf("on-disk disklabel has incorrect disksize (%llu)\n",
283 		    DL_GETDSIZE(lp));
284 	if (DL_GETPSIZE(&lp->d_partitions[RAW_PART]) != disksize)
285 		printf("on-disk disklabel RAW_PART has incorrect size (%llu)\n",
286 		    DL_GETPSIZE(&lp->d_partitions[RAW_PART]));
287 	if (DL_GETPOFFSET(&lp->d_partitions[RAW_PART]) != 0)
288 		printf("on-disk disklabel RAW_PART offset != 0 (%llu)\n",
289 		    DL_GETPOFFSET(&lp->d_partitions[RAW_PART]));
290 #endif
291 	DL_SETDSIZE(lp, disksize);
292 	DL_SETPSIZE(&lp->d_partitions[RAW_PART], disksize);
293 	DL_SETPOFFSET(&lp->d_partitions[RAW_PART], 0);
294 	DL_SETBSTART(lp, boundstart);
295 	DL_SETBEND(lp, boundend < DL_GETDSIZE(lp) ? boundend : DL_GETDSIZE(lp));
296 
297 	lp->d_checksum = 0;
298 	lp->d_checksum = dkcksum(lp);
299 	return (0);
300 }
301 
302 /*
303  * Read a disk sector.
304  */
305 int
306 readdisksector(struct buf *bp, void (*strat)(struct buf *),
307     struct disklabel *lp, u_int64_t sector)
308 {
309 	bp->b_blkno = DL_SECTOBLK(lp, sector);
310 	bp->b_bcount = lp->d_secsize;
311 	bp->b_error = 0;
312 	CLR(bp->b_flags, B_READ | B_WRITE | B_DONE | B_ERROR);
313 	SET(bp->b_flags, B_BUSY | B_READ | B_RAW);
314 
315 	(*strat)(bp);
316 
317 	return (biowait(bp));
318 }
319 
320 /*
321  * If dos partition table requested, attempt to load it and
322  * find disklabel inside a DOS partition. Return buffer
323  * for use in signalling errors if requested.
324  *
325  * We would like to check if each MBR has a valid BOOT_MAGIC, but
326  * we cannot because it doesn't always exist. So.. we assume the
327  * MBR is valid.
328  */
329 int
330 readdoslabel(struct buf *bp, void (*strat)(struct buf *),
331     struct disklabel *lp, daddr_t *partoffp, int spoofonly)
332 {
333 	struct dos_partition dp[NDOSPART], *dp2;
334 	struct disklabel *gptlp;
335 	u_int64_t dospartoff = 0, dospartend = DL_GETBEND(lp);
336 	u_int64_t sector = DOSBBSECTOR;
337 	u_int32_t extoff = 0;
338 	int ourpart = -1, wander = 1, n = 0, loop = 0;
339 	int efi, error, i, offset;
340 
341 	if (lp->d_secpercyl == 0)
342 		return (EINVAL);	/* invalid label */
343 	if (lp->d_secsize == 0)
344 		return (ENOSPC);	/* disk too small */
345 
346 	/* do DOS partitions in the process of getting disklabel? */
347 
348 	/*
349 	 * Read dos partition table, follow extended partitions.
350 	 * Map the partitions to disklabel entries i-p
351 	 */
352 	while (wander && loop < DOS_MAXEBR) {
353 		loop++;
354 		wander = 0;
355 		if (sector < extoff)
356 			sector = extoff;
357 
358 		/* read MBR/EBR */
359 		error = readdisksector(bp, strat, lp, sector);
360 		if (error) {
361 /*wrong*/		if (partoffp)
362 /*wrong*/			*partoffp = -1;
363 			return (error);
364 		}
365 
366 		bcopy(bp->b_data + DOSPARTOFF, dp, sizeof(dp));
367 
368 		if (n == 0 && sector == DOSBBSECTOR) {
369 			u_int16_t mbrtest;
370 
371 			/* Check the end of sector marker. */
372 			mbrtest = ((bp->b_data[510] << 8) & 0xff00) |
373 			    (bp->b_data[511] & 0xff);
374 			if (mbrtest != 0x55aa)
375 				goto notmbr;
376 
377 			efi = gpt_chk_mbr(dp, DL_GETDSIZE(lp));
378 			if (efi == -1)
379 				goto notgpt;
380 
381 			gptlp = malloc(sizeof(struct disklabel), M_DEVBUF,
382 			    M_NOWAIT);
383 			if (gptlp == NULL)
384 				return (ENOMEM);
385 			*gptlp = *lp;
386 			error = spoofgptlabel(bp, strat, gptlp);
387 			if (error == 0) {
388 				dospartoff = DL_GETBSTART(gptlp);
389 				dospartend = DL_GETBEND(gptlp);
390 				if (partoffp == NULL)
391 					*lp = *gptlp;
392 				free(gptlp, M_DEVBUF,
393 				    sizeof(struct disklabel));
394 				if (partoffp && dospartoff == 0)
395 					return (ENXIO);
396 				goto notfat;
397 			} else {
398 				free(gptlp, M_DEVBUF,
399 				    sizeof(struct disklabel));
400 				goto notmbr;
401 			}
402 		}
403 
404 notgpt:
405 		if (ourpart == -1) {
406 			/* Search for our MBR partition */
407 			for (dp2=dp, i=0; i < NDOSPART && ourpart == -1;
408 			    i++, dp2++)
409 				if (letoh32(dp2->dp_size) &&
410 				    dp2->dp_typ == DOSPTYP_OPENBSD)
411 					ourpart = i;
412 			if (ourpart == -1)
413 				goto donot;
414 			/*
415 			 * This is our MBR partition. need sector
416 			 * address for SCSI/IDE, cylinder for
417 			 * ESDI/ST506/RLL
418 			 */
419 			dp2 = &dp[ourpart];
420 			dospartoff = letoh32(dp2->dp_start) + sector;
421 			dospartend = dospartoff + letoh32(dp2->dp_size);
422 
423 			/*
424 			 * Record the OpenBSD partition's placement (in
425 			 * 512-byte blocks!) for the caller. No need to
426 			 * finish spoofing.
427 			 */
428 			if (partoffp) {
429 				*partoffp = DL_SECTOBLK(lp, dospartoff);
430 				return (0);
431 			}
432 
433 			if (lp->d_ntracks == 0)
434 				lp->d_ntracks = dp2->dp_ehd + 1;
435 			if (lp->d_nsectors == 0)
436 				lp->d_nsectors = DPSECT(dp2->dp_esect);
437 			if (lp->d_secpercyl == 0)
438 				lp->d_secpercyl = lp->d_ntracks *
439 				    lp->d_nsectors;
440 		}
441 donot:
442 		/*
443 		 * In case the disklabel read below fails, we want to
444 		 * provide a fake label in i-p.
445 		 */
446 		for (dp2=dp, i=0; i < NDOSPART; i++, dp2++) {
447 			struct partition *pp;
448 			u_int8_t fstype;
449 
450 			if (dp2->dp_typ == DOSPTYP_OPENBSD ||
451 			    dp2->dp_typ == DOSPTYP_EFI)
452 				continue;
453 			if (letoh32(dp2->dp_size) > DL_GETDSIZE(lp))
454 				continue;
455 			if (letoh32(dp2->dp_start) > DL_GETDSIZE(lp))
456 				continue;
457 			if (letoh32(dp2->dp_size) == 0)
458 				continue;
459 
460 			switch (dp2->dp_typ) {
461 			case DOSPTYP_UNUSED:
462 				fstype = FS_UNUSED;
463 				break;
464 
465 			case DOSPTYP_LINUX:
466 				fstype = FS_EXT2FS;
467 				break;
468 
469 			case DOSPTYP_NTFS:
470 				fstype = FS_NTFS;
471 				break;
472 
473 			case DOSPTYP_EFISYS:
474 			case DOSPTYP_FAT12:
475 			case DOSPTYP_FAT16S:
476 			case DOSPTYP_FAT16B:
477 			case DOSPTYP_FAT16L:
478 			case DOSPTYP_FAT32:
479 			case DOSPTYP_FAT32L:
480 				fstype = FS_MSDOS;
481 				break;
482 			case DOSPTYP_EXTEND:
483 			case DOSPTYP_EXTENDL:
484 				sector = letoh32(dp2->dp_start) + extoff;
485 				if (!extoff) {
486 					extoff = letoh32(dp2->dp_start);
487 					sector = 0;
488 				}
489 				wander = 1;
490 				continue;
491 				break;
492 			default:
493 				fstype = FS_OTHER;
494 				break;
495 			}
496 
497 			/*
498 			 * Don't set fstype/offset/size when just looking for
499 			 * the offset of the OpenBSD partition. It would
500 			 * invalidate the disklabel checksum!
501 			 *
502 			 * Don't try to spoof more than 8 partitions, i.e.
503 			 * 'i' -'p'.
504 			 */
505 			if (partoffp || n >= 8)
506 				continue;
507 
508 			pp = &lp->d_partitions[8+n];
509 			n++;
510 			pp->p_fstype = fstype;
511 			if (letoh32(dp2->dp_start))
512 				DL_SETPOFFSET(pp,
513 				    letoh32(dp2->dp_start) + sector);
514 			DL_SETPSIZE(pp, letoh32(dp2->dp_size));
515 		}
516 	}
517 
518 notmbr:
519 	if (n == 0 && sector == DOSBBSECTOR && ourpart == -1) {
520 		u_int16_t fattest;
521 
522 		/* Check for a valid initial jmp instruction. */
523 		switch ((u_int8_t)bp->b_data[0]) {
524 		case 0xeb:
525 			/*
526 			 * Two-byte jmp instruction. The 2nd byte is the number
527 			 * of bytes to jmp and the 3rd byte must be a NOP.
528 			 */
529 			if ((u_int8_t)bp->b_data[2] != 0x90)
530 				goto notfat;
531 			break;
532 		case 0xe9:
533 			/*
534 			 * Three-byte jmp instruction. The next two bytes are a
535 			 * little-endian 16 bit value.
536 			 */
537 			break;
538 		default:
539 			goto notfat;
540 			break;
541 		}
542 
543 		/* Check for a valid bytes per sector value. */
544 		fattest = ((bp->b_data[12] << 8) & 0xff00) |
545 		    (bp->b_data[11] & 0xff);
546 		if (fattest < 512 || fattest > 4096 || (fattest % 512 != 0))
547 			goto notfat;
548 
549 		if (partoffp)
550 			return (ENXIO);	/* No place for disklabel on FAT! */
551 
552 		DL_SETPSIZE(&lp->d_partitions['i' - 'a'],
553 		    DL_GETPSIZE(&lp->d_partitions[RAW_PART]));
554 		DL_SETPOFFSET(&lp->d_partitions['i' - 'a'], 0);
555 		lp->d_partitions['i' - 'a'].p_fstype = FS_MSDOS;
556 
557 		spoofonly = 1;	/* No disklabel to read from disk. */
558 	}
559 
560 notfat:
561 	/* record the OpenBSD partition's placement for the caller */
562 	if (partoffp)
563 		*partoffp = DL_SECTOBLK(lp, dospartoff);
564 	else {
565 		DL_SETBSTART(lp, dospartoff);
566 		DL_SETBEND(lp, (dospartend < DL_GETDSIZE(lp)) ? dospartend :
567 		    DL_GETDSIZE(lp));
568 	}
569 
570 	/* don't read the on-disk label if we are in spoofed-only mode */
571 	if (spoofonly)
572 		return (0);
573 
574 	error = readdisksector(bp, strat, lp, dospartoff +
575 	    DL_BLKTOSEC(lp, DOS_LABELSECTOR));
576 	if (error)
577 		return (bp->b_error);
578 
579 	offset = DL_BLKOFFSET(lp, DOS_LABELSECTOR);
580 	error = checkdisklabel(bp->b_data + offset, lp,
581 	    DL_GETBSTART((struct disklabel*)(bp->b_data+offset)),
582 	    DL_GETBEND((struct disklabel *)(bp->b_data+offset)));
583 
584 	return (error);
585 }
586 
587 /*
588  * Return the index into dp[] of the EFI GPT (0xEE) partition, or -1 if no such
589  * partition exists.
590  *
591  * Copied into sbin/fdisk/mbr.c.
592  */
593 int
594 gpt_chk_mbr(struct dos_partition *dp, uint64_t dsize)
595 {
596 	struct dos_partition *dp2;
597 	int efi, eficnt, found, i;
598 	uint32_t psize;
599 
600 	found = efi = eficnt = 0;
601 	for (dp2 = dp, i = 0; i < NDOSPART; i++, dp2++) {
602 		if (dp2->dp_typ == DOSPTYP_UNUSED)
603 			continue;
604 		found++;
605 		if (dp2->dp_typ != DOSPTYP_EFI)
606 			continue;
607 		if (letoh32(dp2->dp_start) != GPTSECTOR)
608 			continue;
609 		psize = letoh32(dp2->dp_size);
610 		if (psize <= (dsize - GPTSECTOR) || psize == UINT32_MAX) {
611 			efi = i;
612 			eficnt++;
613 		}
614 	}
615 	if (found == 1 && eficnt == 1)
616 		return (efi);
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 gh_csum;
627 	uint32_t ghsize, ghpartsize, ghpartspersec;
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 	ghpartlba = letoh64(gh->gh_part_lba);
639 	ghlbaend = letoh64(gh->gh_lba_end);
640 	ghlbastart = letoh64(gh->gh_lba_start);
641 
642 	if (ghsize < GPTMINHDRSIZE || ghsize > sizeof(struct gpt_header))
643 		return (EINVAL);
644 
645 	gh_csum = gh->gh_csum;
646 	gh->gh_csum = 0;
647 	gh->gh_csum = htole32(crc32(0, (unsigned char *)gh, ghsize));
648 
649 	if (gh_csum != gh->gh_csum)
650 		return (EINVAL);
651 
652 	if (ghlbastart >= DL_GETDSIZE(lp) ||
653 	    ghpartlba >= DL_GETDSIZE(lp))
654 		return (EINVAL);
655 
656 	/*
657 	* Size per partition entry shall be 128*(2**n) with n >= 0.
658 	* We don't support partition entries larger than block size.
659 	*/
660 	if (ghpartsize % GPTMINPARTSIZE || ghpartsize > lp->d_secsize
661 	    || ghpartspersec == 0) {
662 		DPRINTF("invalid partition size\n");
663 		return (EINVAL);
664 	}
665 
666 	/* XXX: we don't support multiples of GPTMINPARTSIZE yet */
667 	if (ghpartsize != GPTMINPARTSIZE) {
668 		DPRINTF("partition sizes larger than %d bytes are not "
669 		    "supported", GPTMINPARTSIZE);
670 		return (EINVAL);
671 	}
672 
673 	return 0;
674 }
675 
676 int
677 gpt_chk_parts(struct gpt_header *gh, struct gpt_partition *gp)
678 {
679 	u_int32_t gh_part_csum;
680 
681 	gh_part_csum = htole32(crc32(0, (unsigned char *)gp,
682 	    letoh32(gh->gh_part_num) * letoh32(gh->gh_part_size)));
683 
684 	if (gh_part_csum != gh->gh_part_csum)
685 		return (EINVAL);
686 
687 	return 0;
688 }
689 
690 int
691 gpt_get_fstype(struct uuid *uuid_part)
692 {
693 	static int init = 0;
694 	static struct uuid uuid_openbsd, uuid_msdos, uuid_chromefs,
695 	    uuid_linux, uuid_hfs, uuid_unused, uuid_efi_system, uuid_bios_boot;
696 	static const uint8_t gpt_uuid_openbsd[] = GPT_UUID_OPENBSD;
697 	static const uint8_t gpt_uuid_msdos[] = GPT_UUID_MSDOS;
698 	static const uint8_t gpt_uuid_chromerootfs[] = GPT_UUID_CHROMEROOTFS;
699 	static const uint8_t gpt_uuid_linux[] = GPT_UUID_LINUX;
700 	static const uint8_t gpt_uuid_hfs[] = GPT_UUID_APPLE_HFS;
701 	static const uint8_t gpt_uuid_unused[] = GPT_UUID_UNUSED;
702 	static const uint8_t gpt_uuid_efi_system[] = GPT_UUID_EFI_SYSTEM;
703 	static const uint8_t gpt_uuid_bios_boot[] = GPT_UUID_BIOS_BOOT;
704 
705 	if (init == 0) {
706 		uuid_dec_be(gpt_uuid_openbsd, &uuid_openbsd);
707 		uuid_dec_be(gpt_uuid_msdos, &uuid_msdos);
708 		uuid_dec_be(gpt_uuid_chromerootfs, &uuid_chromefs);
709 		uuid_dec_be(gpt_uuid_linux, &uuid_linux);
710 		uuid_dec_be(gpt_uuid_hfs, &uuid_hfs);
711 		uuid_dec_be(gpt_uuid_unused, &uuid_unused);
712 		uuid_dec_be(gpt_uuid_efi_system, &uuid_efi_system);
713 		uuid_dec_be(gpt_uuid_bios_boot, &uuid_bios_boot);
714 		init = 1;
715 	}
716 
717 	if (!memcmp(uuid_part, &uuid_unused, sizeof(struct uuid)))
718 		return FS_UNUSED;
719 	else if (!memcmp(uuid_part, &uuid_openbsd, sizeof(struct uuid)))
720 		return FS_BSDFFS;
721 	else if (!memcmp(uuid_part, &uuid_msdos, sizeof(struct uuid)))
722 		return FS_MSDOS;
723 	else if (!memcmp(uuid_part, &uuid_chromefs, sizeof(struct uuid)))
724 		return FS_EXT2FS;
725 	else if (!memcmp(uuid_part, &uuid_linux, sizeof(struct uuid)))
726 		return FS_EXT2FS;
727 	else if (!memcmp(uuid_part, &uuid_hfs, sizeof(struct uuid)))
728 		return FS_HFS;
729 	else if (!memcmp(uuid_part, &uuid_efi_system, sizeof(struct uuid)))
730 		return FS_MSDOS;
731 	else if (!memcmp(uuid_part, &uuid_bios_boot, sizeof(struct uuid)))
732 		return FS_BOOT;
733 	else
734 		return FS_OTHER;
735 }
736 
737 /*
738  * Spoof a disklabel based on the GPT information on the disk.
739  */
740 int
741 spoofgptlabel(struct buf *bp, void (*strat)(struct buf *),
742     struct disklabel *lp)
743 {
744 	static const u_int8_t gpt_uuid_openbsd[] = GPT_UUID_OPENBSD;
745 	struct gpt_header gh;
746 	struct uuid uuid_part, uuid_openbsd;
747 	struct gpt_partition *gp;
748 	struct partition *pp;
749 	size_t gpsz;
750 	u_int64_t ghlbaend, ghlbastart, sector;
751 	u_int64_t start, end;
752 	int i, error, found, n;
753 	uint32_t ghpartnum;
754 
755 	uuid_dec_be(gpt_uuid_openbsd, &uuid_openbsd);
756 
757 	for (sector = GPTSECTOR; ; sector = DL_GETDSIZE(lp) - 1) {
758 		uint64_t ghpartlba;
759 		uint32_t ghpartsize;
760 		uint32_t ghpartspersec;
761 
762 		error = readdisksector(bp, strat, lp, sector);
763 		if (error) {
764 			DPRINTF("error reading from disk\n");
765 			return (error);
766 		}
767 
768 		bcopy(bp->b_data, &gh, sizeof(gh));
769 
770 		if (gpt_chk_hdr(&gh, lp)) {
771 			if (sector != GPTSECTOR) {
772 				DPRINTF("alternate header also broken\n");
773 				return (EINVAL);
774 			}
775 			continue;
776 		}
777 
778 		ghpartsize = letoh32(gh.gh_part_size);
779 		ghpartspersec = lp->d_secsize / ghpartsize;
780 		ghpartnum = letoh32(gh.gh_part_num);
781 		ghpartlba = letoh64(gh.gh_part_lba);
782 		ghlbaend = letoh64(gh.gh_lba_end);
783 		ghlbastart = letoh64(gh.gh_lba_start);
784 
785 		/* read GPT partition entry array */
786 		gp = mallocarray(ghpartnum, sizeof(struct gpt_partition),
787 		    M_DEVBUF, M_NOWAIT|M_ZERO);
788 		if (gp == NULL)
789 			return (ENOMEM);
790 		gpsz = ghpartnum * sizeof(struct gpt_partition);
791 
792 		/*
793 		* XXX:	Fails if # of partition entries is not a multiple of
794 		*	ghpartspersec.
795 		*/
796 		for (i = 0; i < ghpartnum / ghpartspersec; i++) {
797 			error = readdisksector(bp, strat, lp, ghpartlba + i);
798 			if (error) {
799 				free(gp, M_DEVBUF, gpsz);
800 				return (error);
801 			}
802 
803 			bcopy(bp->b_data, gp + i * ghpartspersec,
804 			    ghpartspersec * sizeof(struct gpt_partition));
805 		}
806 
807 		if (gpt_chk_parts(&gh, gp)) {
808 			free(gp, M_DEVBUF, gpsz);
809 			if (letoh64(gh.gh_lba_self) != GPTSECTOR) {
810 				DPRINTF("alternate partition entries are also "
811 				    "broken\n");
812 				return (EINVAL);
813 			}
814 			continue;
815 		}
816 		break;
817 	}
818 
819 	/* Find OpenBSD partition and spoof others along the way. */
820 	DL_SETBSTART(lp, ghlbastart);
821 	DL_SETBEND(lp, ghlbaend + 1);
822 	found = 0;
823 	n = 'i' - 'a';	/* Start spoofing at 'i', a.k.a. 8. */
824 	for (i = 0; i < ghpartnum; i++) {
825 		start = letoh64(gp[i].gp_lba_start);
826 		end = letoh64(gp[i].gp_lba_end);
827 		if (start > end || start < ghlbastart || end > ghlbaend)
828 			continue; /* entry invalid */
829 
830 		uuid_dec_le(&gp[i].gp_type, &uuid_part);
831 		if (memcmp(&uuid_part, &uuid_openbsd, sizeof(struct uuid)) == 0) {
832 			if (found == 0) {
833 				found = 1;
834 				DL_SETBSTART(lp, start);
835 				DL_SETBEND(lp, end + 1);
836 			}
837 		} else if (n < MAXPARTITIONS) {
838 			pp = &lp->d_partitions[n];
839 			n++;
840 			pp->p_fstype = gpt_get_fstype(&uuid_part);
841 			DL_SETPOFFSET(pp, start);
842 			DL_SETPSIZE(pp, end - start + 1);
843 		}
844 	}
845 
846 	free(gp, M_DEVBUF, gpsz);
847 
848 	return (0);
849 }
850 
851 /*
852  * Check new disk label for sensibility before setting it.
853  */
854 int
855 setdisklabel(struct disklabel *olp, struct disklabel *nlp, u_int openmask)
856 {
857 	struct partition *opp, *npp;
858 	struct disk *dk;
859 	int i;
860 
861 	/* sanity clause */
862 	if (nlp->d_secpercyl == 0 || nlp->d_secsize == 0 ||
863 	    (nlp->d_secsize % DEV_BSIZE) != 0)
864 		return (EINVAL);
865 
866 	/* special case to allow disklabel to be invalidated */
867 	if (nlp->d_magic == 0xffffffff) {
868 		*olp = *nlp;
869 		return (0);
870 	}
871 
872 	if (nlp->d_magic != DISKMAGIC || nlp->d_magic2 != DISKMAGIC ||
873 	    dkcksum(nlp) != 0)
874 		return (EINVAL);
875 
876 	/* XXX missing check if other dos partitions will be overwritten */
877 
878 	for (i = 0; i < MAXPARTITIONS; i++) {
879 		opp = &olp->d_partitions[i];
880 		npp = &nlp->d_partitions[i];
881 		if ((openmask & (1 << i)) &&
882 		    (DL_GETPOFFSET(npp) != DL_GETPOFFSET(opp) ||
883 		    DL_GETPSIZE(npp) < DL_GETPSIZE(opp)))
884 			return (EBUSY);
885 		/*
886 		 * Copy internally-set partition information
887 		 * if new label doesn't include it.		XXX
888 		 */
889 		if (npp->p_fstype == FS_UNUSED && opp->p_fstype != FS_UNUSED) {
890 			npp->p_fragblock = opp->p_fragblock;
891 			npp->p_cpg = opp->p_cpg;
892 		}
893 	}
894 
895 	/* Generate a UID if the disklabel does not already have one. */
896 	if (duid_iszero(nlp->d_uid)) {
897 		do {
898 			arc4random_buf(nlp->d_uid, sizeof(nlp->d_uid));
899 			TAILQ_FOREACH(dk, &disklist, dk_link)
900 				if (dk->dk_label &&
901 				    duid_equal(dk->dk_label->d_uid, nlp->d_uid))
902 					break;
903 		} while (dk != NULL || duid_iszero(nlp->d_uid));
904 	}
905 
906 	/* Preserve the disk size and RAW_PART values. */
907 	DL_SETDSIZE(nlp, DL_GETDSIZE(olp));
908 	npp = &nlp->d_partitions[RAW_PART];
909 	DL_SETPOFFSET(npp, 0);
910 	DL_SETPSIZE(npp, DL_GETDSIZE(nlp));
911 
912 	nlp->d_checksum = 0;
913 	nlp->d_checksum = dkcksum(nlp);
914 	*olp = *nlp;
915 
916 	disk_change = 1;
917 
918 	return (0);
919 }
920 
921 /*
922  * Determine the size of the transfer, and make sure it is within the
923  * boundaries of the partition. Adjust transfer if needed, and signal errors or
924  * early completion.
925  */
926 int
927 bounds_check_with_label(struct buf *bp, struct disklabel *lp)
928 {
929 	struct partition *p = &lp->d_partitions[DISKPART(bp->b_dev)];
930 	daddr_t partblocks, sz;
931 
932 	/* Avoid division by zero, negative offsets, and negative sizes. */
933 	if (lp->d_secpercyl == 0 || bp->b_blkno < 0 || bp->b_bcount < 0)
934 		goto bad;
935 
936 	/* Ensure transfer is a whole number of aligned sectors. */
937 	if ((bp->b_blkno % DL_BLKSPERSEC(lp)) != 0 ||
938 	    (bp->b_bcount % lp->d_secsize) != 0)
939 		goto bad;
940 
941 	/* Ensure transfer starts within partition boundary. */
942 	partblocks = DL_SECTOBLK(lp, DL_GETPSIZE(p));
943 	if (bp->b_blkno > partblocks)
944 		goto bad;
945 
946 	/* If exactly at end of partition or null transfer, return EOF. */
947 	if (bp->b_blkno == partblocks || bp->b_bcount == 0)
948 		goto done;
949 
950 	/* Truncate request if it extends past the end of the partition. */
951 	sz = bp->b_bcount >> DEV_BSHIFT;
952 	if (sz > partblocks - bp->b_blkno) {
953 		sz = partblocks - bp->b_blkno;
954 		bp->b_bcount = sz << DEV_BSHIFT;
955 	}
956 
957 	return (0);
958 
959  bad:
960 	bp->b_error = EINVAL;
961 	bp->b_flags |= B_ERROR;
962  done:
963 	bp->b_resid = bp->b_bcount;
964 	return (-1);
965 }
966 
967 /*
968  * Disk error is the preface to plaintive error messages
969  * about failing disk transfers.  It prints messages of the form
970 
971 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
972 
973  * if the offset of the error in the transfer and a disk label
974  * are both available.  blkdone should be -1 if the position of the error
975  * is unknown; the disklabel pointer may be null from drivers that have not
976  * been converted to use them.  The message is printed with printf
977  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
978  * The message should be completed (with at least a newline) with printf
979  * or addlog, respectively.  There is no trailing space.
980  */
981 void
982 diskerr(struct buf *bp, char *dname, char *what, int pri, int blkdone,
983     struct disklabel *lp)
984 {
985 	int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
986 	int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2)));
987 	char partname = 'a' + part;
988 	daddr_t sn;
989 
990 	if (pri != LOG_PRINTF) {
991 		log(pri, "%s", "");
992 		pr = addlog;
993 	} else
994 		pr = printf;
995 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
996 	    bp->b_flags & B_READ ? "read" : "writ");
997 	sn = bp->b_blkno;
998 	if (bp->b_bcount <= DEV_BSIZE)
999 		(*pr)("%lld", (long long)sn);
1000 	else {
1001 		if (blkdone >= 0) {
1002 			sn += blkdone;
1003 			(*pr)("%lld of ", (long long)sn);
1004 		}
1005 		(*pr)("%lld-%lld", (long long)bp->b_blkno,
1006 		    (long long)(bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE));
1007 	}
1008 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
1009 		sn += DL_SECTOBLK(lp, DL_GETPOFFSET(&lp->d_partitions[part]));
1010 		(*pr)(" (%s%d bn %lld; cn %lld", dname, unit, (long long)sn,
1011 		    (long long)(sn / DL_SECTOBLK(lp, lp->d_secpercyl)));
1012 		sn %= DL_SECTOBLK(lp, lp->d_secpercyl);
1013 		(*pr)(" tn %lld sn %lld)",
1014 		    (long long)(sn / DL_SECTOBLK(lp, lp->d_nsectors)),
1015 		    (long long)(sn % DL_SECTOBLK(lp, lp->d_nsectors)));
1016 	}
1017 }
1018 
1019 /*
1020  * Initialize the disklist.  Called by main() before autoconfiguration.
1021  */
1022 void
1023 disk_init(void)
1024 {
1025 
1026 	TAILQ_INIT(&disklist);
1027 	disk_count = disk_change = 0;
1028 }
1029 
1030 int
1031 disk_construct(struct disk *diskp)
1032 {
1033 	rw_init_flags(&diskp->dk_lock, "dklk", RWL_IS_VNODE);
1034 	mtx_init(&diskp->dk_mtx, IPL_BIO);
1035 
1036 	diskp->dk_flags |= DKF_CONSTRUCTED;
1037 
1038 	return (0);
1039 }
1040 
1041 /*
1042  * Attach a disk.
1043  */
1044 void
1045 disk_attach(struct device *dv, struct disk *diskp)
1046 {
1047 	int majdev;
1048 
1049 	KERNEL_ASSERT_LOCKED();
1050 
1051 	if (!ISSET(diskp->dk_flags, DKF_CONSTRUCTED))
1052 		disk_construct(diskp);
1053 
1054 	/*
1055 	 * Allocate and initialize the disklabel structures.  Note that
1056 	 * it's not safe to sleep here, since we're probably going to be
1057 	 * called during autoconfiguration.
1058 	 */
1059 	diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF,
1060 	    M_NOWAIT|M_ZERO);
1061 	if (diskp->dk_label == NULL)
1062 		panic("disk_attach: can't allocate storage for disklabel");
1063 
1064 	/*
1065 	 * Set the attached timestamp.
1066 	 */
1067 	microuptime(&diskp->dk_attachtime);
1068 
1069 	/*
1070 	 * Link into the disklist.
1071 	 */
1072 	TAILQ_INSERT_TAIL(&disklist, diskp, dk_link);
1073 	++disk_count;
1074 	disk_change = 1;
1075 
1076 	/*
1077 	 * Store device structure and number for later use.
1078 	 */
1079 	diskp->dk_device = dv;
1080 	diskp->dk_devno = NODEV;
1081 	if (dv != NULL) {
1082 		majdev = findblkmajor(dv);
1083 		if (majdev >= 0)
1084 			diskp->dk_devno =
1085 			    MAKEDISKDEV(majdev, dv->dv_unit, RAW_PART);
1086 
1087 		if (diskp->dk_devno != NODEV) {
1088 			struct disk_attach_task *dat;
1089 
1090 			dat = malloc(sizeof(*dat), M_TEMP, M_WAITOK);
1091 
1092 			/* XXX: Assumes dk is part of the device softc. */
1093 			device_ref(dv);
1094 			dat->dk = diskp;
1095 
1096 			task_set(&dat->task, disk_attach_callback, dat);
1097 			task_add(systq, &dat->task);
1098 		}
1099 	}
1100 
1101 	if (softraid_disk_attach)
1102 		softraid_disk_attach(diskp, 1);
1103 }
1104 
1105 void
1106 disk_attach_callback(void *xdat)
1107 {
1108 	struct disk_attach_task *dat = xdat;
1109 	struct disk *dk = dat->dk;
1110 	struct disklabel dl;
1111 	char errbuf[100];
1112 
1113 	free(dat, M_TEMP, sizeof(*dat));
1114 
1115 	if (dk->dk_flags & (DKF_OPENED | DKF_NOLABELREAD))
1116 		goto done;
1117 
1118 	/* Read disklabel. */
1119 	if (disk_readlabel(&dl, dk->dk_devno, errbuf, sizeof(errbuf)) == NULL) {
1120 		enqueue_randomness(dl.d_checksum);
1121 		dk->dk_flags |= DKF_LABELVALID;
1122 	}
1123 
1124 done:
1125 	dk->dk_flags |= DKF_OPENED;
1126 	device_unref(dk->dk_device);
1127 	wakeup(dk);
1128 }
1129 
1130 /*
1131  * Detach a disk.
1132  */
1133 void
1134 disk_detach(struct disk *diskp)
1135 {
1136 	KERNEL_ASSERT_LOCKED();
1137 
1138 	if (softraid_disk_attach)
1139 		softraid_disk_attach(diskp, -1);
1140 
1141 	/*
1142 	 * Free the space used by the disklabel structures.
1143 	 */
1144 	free(diskp->dk_label, M_DEVBUF, sizeof(*diskp->dk_label));
1145 
1146 	/*
1147 	 * Remove from the disklist.
1148 	 */
1149 	TAILQ_REMOVE(&disklist, diskp, dk_link);
1150 	disk_change = 1;
1151 	if (--disk_count < 0)
1152 		panic("disk_detach: disk_count < 0");
1153 }
1154 
1155 int
1156 disk_openpart(struct disk *dk, int part, int fmt, int haslabel)
1157 {
1158 	KASSERT(part >= 0 && part < MAXPARTITIONS);
1159 
1160 	/* Unless opening the raw partition, check that the partition exists. */
1161 	if (part != RAW_PART && (!haslabel ||
1162 	    part >= dk->dk_label->d_npartitions ||
1163 	    dk->dk_label->d_partitions[part].p_fstype == FS_UNUSED))
1164 		return (ENXIO);
1165 
1166 	/* Ensure the partition doesn't get changed under our feet. */
1167 	switch (fmt) {
1168 	case S_IFCHR:
1169 		dk->dk_copenmask |= (1 << part);
1170 		break;
1171 	case S_IFBLK:
1172 		dk->dk_bopenmask |= (1 << part);
1173 		break;
1174 	}
1175 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
1176 
1177 	return (0);
1178 }
1179 
1180 void
1181 disk_closepart(struct disk *dk, int part, int fmt)
1182 {
1183 	KASSERT(part >= 0 && part < MAXPARTITIONS);
1184 
1185 	switch (fmt) {
1186 	case S_IFCHR:
1187 		dk->dk_copenmask &= ~(1 << part);
1188 		break;
1189 	case S_IFBLK:
1190 		dk->dk_bopenmask &= ~(1 << part);
1191 		break;
1192 	}
1193 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
1194 }
1195 
1196 void
1197 disk_gone(int (*open)(dev_t, int, int, struct proc *), int unit)
1198 {
1199 	int bmaj, cmaj, mn;
1200 
1201 	/* Locate the lowest minor number to be detached. */
1202 	mn = DISKMINOR(unit, 0);
1203 
1204 	for (bmaj = 0; bmaj < nblkdev; bmaj++)
1205 		if (bdevsw[bmaj].d_open == open)
1206 			vdevgone(bmaj, mn, mn + MAXPARTITIONS - 1, VBLK);
1207 	for (cmaj = 0; cmaj < nchrdev; cmaj++)
1208 		if (cdevsw[cmaj].d_open == open)
1209 			vdevgone(cmaj, mn, mn + MAXPARTITIONS - 1, VCHR);
1210 }
1211 
1212 /*
1213  * Increment a disk's busy counter.  If the counter is going from
1214  * 0 to 1, set the timestamp.
1215  */
1216 void
1217 disk_busy(struct disk *diskp)
1218 {
1219 
1220 	/*
1221 	 * XXX We'd like to use something as accurate as microtime(),
1222 	 * but that doesn't depend on the system TOD clock.
1223 	 */
1224 	mtx_enter(&diskp->dk_mtx);
1225 	if (diskp->dk_busy++ == 0)
1226 		microuptime(&diskp->dk_timestamp);
1227 	mtx_leave(&diskp->dk_mtx);
1228 }
1229 
1230 /*
1231  * Decrement a disk's busy counter, increment the byte count, total busy
1232  * time, and reset the timestamp.
1233  */
1234 void
1235 disk_unbusy(struct disk *diskp, long bcount, daddr_t blkno, int read)
1236 {
1237 	struct timeval dv_time, diff_time;
1238 
1239 	mtx_enter(&diskp->dk_mtx);
1240 
1241 	if (diskp->dk_busy-- == 0)
1242 		printf("disk_unbusy: %s: dk_busy < 0\n", diskp->dk_name);
1243 
1244 	microuptime(&dv_time);
1245 
1246 	timersub(&dv_time, &diskp->dk_timestamp, &diff_time);
1247 	timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time);
1248 
1249 	diskp->dk_timestamp = dv_time;
1250 	if (bcount > 0) {
1251 		if (read) {
1252 			diskp->dk_rbytes += bcount;
1253 			diskp->dk_rxfer++;
1254 		} else {
1255 			diskp->dk_wbytes += bcount;
1256 			diskp->dk_wxfer++;
1257 		}
1258 	} else
1259 		diskp->dk_seek++;
1260 
1261 	mtx_leave(&diskp->dk_mtx);
1262 
1263 	enqueue_randomness(bcount ^ diff_time.tv_usec ^
1264 	    (blkno >> 32) ^ (blkno & 0xffffffff));
1265 }
1266 
1267 int
1268 disk_lock(struct disk *dk)
1269 {
1270 	return (rw_enter(&dk->dk_lock, RW_WRITE|RW_INTR));
1271 }
1272 
1273 void
1274 disk_lock_nointr(struct disk *dk)
1275 {
1276 	rw_enter_write(&dk->dk_lock);
1277 }
1278 
1279 void
1280 disk_unlock(struct disk *dk)
1281 {
1282 	rw_exit_write(&dk->dk_lock);
1283 }
1284 
1285 int
1286 dk_mountroot(void)
1287 {
1288 	char errbuf[100];
1289 	int part = DISKPART(rootdev);
1290 	int (*mountrootfn)(void);
1291 	struct disklabel dl;
1292 	char *error;
1293 
1294 	error = disk_readlabel(&dl, rootdev, errbuf, sizeof(errbuf));
1295 	if (error)
1296 		panic("%s", error);
1297 
1298 	if (DL_GETPSIZE(&dl.d_partitions[part]) == 0)
1299 		panic("root filesystem has size 0");
1300 	switch (dl.d_partitions[part].p_fstype) {
1301 #ifdef EXT2FS
1302 	case FS_EXT2FS:
1303 		{
1304 		extern int ext2fs_mountroot(void);
1305 		mountrootfn = ext2fs_mountroot;
1306 		}
1307 		break;
1308 #endif
1309 #ifdef FFS
1310 	case FS_BSDFFS:
1311 		{
1312 		extern int ffs_mountroot(void);
1313 		mountrootfn = ffs_mountroot;
1314 		}
1315 		break;
1316 #endif
1317 #ifdef CD9660
1318 	case FS_ISO9660:
1319 		{
1320 		extern int cd9660_mountroot(void);
1321 		mountrootfn = cd9660_mountroot;
1322 		}
1323 		break;
1324 #endif
1325 	default:
1326 #ifdef FFS
1327 		{
1328 		extern int ffs_mountroot(void);
1329 
1330 		printf("filesystem type %d not known.. assuming ffs\n",
1331 		    dl.d_partitions[part].p_fstype);
1332 		mountrootfn = ffs_mountroot;
1333 		}
1334 #else
1335 		panic("disk 0x%x filesystem type %d not known",
1336 		    rootdev, dl.d_partitions[part].p_fstype);
1337 #endif
1338 	}
1339 	return (*mountrootfn)();
1340 }
1341 
1342 struct device *
1343 getdisk(char *str, int len, int defpart, dev_t *devp)
1344 {
1345 	struct device *dv;
1346 
1347 	if ((dv = parsedisk(str, len, defpart, devp)) == NULL) {
1348 		printf("use one of: exit");
1349 		TAILQ_FOREACH(dv, &alldevs, dv_list) {
1350 			if (dv->dv_class == DV_DISK)
1351 				printf(" %s[a-p]", dv->dv_xname);
1352 #if defined(NFSCLIENT)
1353 			if (dv->dv_class == DV_IFNET)
1354 				printf(" %s", dv->dv_xname);
1355 #endif
1356 		}
1357 		printf("\n");
1358 	}
1359 	return (dv);
1360 }
1361 
1362 struct device *
1363 parsedisk(char *str, int len, int defpart, dev_t *devp)
1364 {
1365 	struct device *dv;
1366 	int majdev, part = defpart;
1367 	char c;
1368 
1369 	if (len == 0)
1370 		return (NULL);
1371 	c = str[len-1];
1372 	if (c >= 'a' && (c - 'a') < MAXPARTITIONS) {
1373 		part = c - 'a';
1374 		len -= 1;
1375 	}
1376 
1377 	TAILQ_FOREACH(dv, &alldevs, dv_list) {
1378 		if (dv->dv_class == DV_DISK &&
1379 		    strncmp(str, dv->dv_xname, len) == 0 &&
1380 		    dv->dv_xname[len] == '\0') {
1381 			majdev = findblkmajor(dv);
1382 			if (majdev < 0)
1383 				return NULL;
1384 			*devp = MAKEDISKDEV(majdev, dv->dv_unit, part);
1385 			break;
1386 		}
1387 #if defined(NFSCLIENT)
1388 		if (dv->dv_class == DV_IFNET &&
1389 		    strncmp(str, dv->dv_xname, len) == 0 &&
1390 		    dv->dv_xname[len] == '\0') {
1391 			*devp = NODEV;
1392 			break;
1393 		}
1394 #endif
1395 	}
1396 
1397 	return (dv);
1398 }
1399 
1400 void
1401 setroot(struct device *bootdv, int part, int exitflags)
1402 {
1403 	int majdev, unit, len, s, slept = 0;
1404 	struct swdevt *swp;
1405 	struct device *rootdv, *dv;
1406 	dev_t nrootdev, nswapdev = NODEV, temp = NODEV;
1407 	struct ifnet *ifp = NULL;
1408 	struct disk *dk;
1409 	char buf[128];
1410 #if defined(NFSCLIENT)
1411 	extern char *nfsbootdevname;
1412 #endif
1413 
1414 	/* Ensure that all disk attach callbacks have completed. */
1415 	do {
1416 		TAILQ_FOREACH(dk, &disklist, dk_link) {
1417 			if (dk->dk_devno != NODEV &&
1418 			    (dk->dk_flags & DKF_OPENED) == 0) {
1419 				tsleep_nsec(dk, 0, "dkopen", SEC_TO_NSEC(1));
1420 				slept++;
1421 				break;
1422 			}
1423 		}
1424 	} while (dk != NULL && slept < 5);
1425 
1426 	if (slept == 5) {
1427 		printf("disklabels not read:");
1428 		TAILQ_FOREACH(dk, &disklist, dk_link)
1429 			if (dk->dk_devno != NODEV &&
1430 			    (dk->dk_flags & DKF_OPENED) == 0)
1431 				printf(" %s", dk->dk_name);
1432 		printf("\n");
1433 	}
1434 
1435 	if (duid_iszero(bootduid)) {
1436 		/* Locate DUID for boot disk since it was not provided. */
1437 		TAILQ_FOREACH(dk, &disklist, dk_link)
1438 			if (dk->dk_device == bootdv)
1439 				break;
1440 		if (dk && (dk->dk_flags & DKF_LABELVALID))
1441 			bcopy(dk->dk_label->d_uid, bootduid, sizeof(bootduid));
1442 	} else if (bootdv == NULL) {
1443 		/* Locate boot disk based on the provided DUID. */
1444 		TAILQ_FOREACH(dk, &disklist, dk_link)
1445 			if (duid_equal(dk->dk_label->d_uid, bootduid))
1446 				break;
1447 		if (dk && (dk->dk_flags & DKF_LABELVALID))
1448 			bootdv = dk->dk_device;
1449 	}
1450 	bcopy(bootduid, rootduid, sizeof(rootduid));
1451 
1452 #if NSOFTRAID > 0
1453 	sr_map_root();
1454 #endif
1455 
1456 	/*
1457 	 * If `swap generic' and we couldn't determine boot device,
1458 	 * ask the user.
1459 	 */
1460 	dk = NULL;
1461 	if (mountroot == NULL && bootdv == NULL)
1462 		boothowto |= RB_ASKNAME;
1463 	if (boothowto & RB_ASKNAME) {
1464 		while (1) {
1465 			printf("root device");
1466 			if (bootdv != NULL) {
1467 				printf(" (default %s", bootdv->dv_xname);
1468 				if (bootdv->dv_class == DV_DISK)
1469 					printf("%c", 'a' + part);
1470 				printf(")");
1471 			}
1472 			printf(": ");
1473 			s = splhigh();
1474 			cnpollc(1);
1475 			len = getsn(buf, sizeof(buf));
1476 			cnpollc(0);
1477 			splx(s);
1478 			if (strcmp(buf, "exit") == 0)
1479 				reboot(exitflags);
1480 			if (len == 0 && bootdv != NULL) {
1481 				strlcpy(buf, bootdv->dv_xname, sizeof buf);
1482 				len = strlen(buf);
1483 			}
1484 			if (len > 0 && buf[len - 1] == '*') {
1485 				buf[--len] = '\0';
1486 				dv = getdisk(buf, len, part, &nrootdev);
1487 				if (dv != NULL) {
1488 					rootdv = dv;
1489 					nswapdev = nrootdev;
1490 					goto gotswap;
1491 				}
1492 			}
1493 			dv = getdisk(buf, len, part, &nrootdev);
1494 			if (dv != NULL) {
1495 				rootdv = dv;
1496 				break;
1497 			}
1498 		}
1499 
1500 		if (rootdv->dv_class == DV_IFNET)
1501 			goto gotswap;
1502 
1503 		/* try to build swap device out of new root device */
1504 		while (1) {
1505 			printf("swap device");
1506 			if (rootdv != NULL)
1507 				printf(" (default %s%s)", rootdv->dv_xname,
1508 				    rootdv->dv_class == DV_DISK ? "b" : "");
1509 			printf(": ");
1510 			s = splhigh();
1511 			cnpollc(1);
1512 			len = getsn(buf, sizeof(buf));
1513 			cnpollc(0);
1514 			splx(s);
1515 			if (strcmp(buf, "exit") == 0)
1516 				reboot(exitflags);
1517 			if (len == 0 && rootdv != NULL) {
1518 				switch (rootdv->dv_class) {
1519 				case DV_IFNET:
1520 					nswapdev = NODEV;
1521 					break;
1522 				case DV_DISK:
1523 					nswapdev = MAKEDISKDEV(major(nrootdev),
1524 					    DISKUNIT(nrootdev), 1);
1525 					if (nswapdev == nrootdev)
1526 						continue;
1527 					break;
1528 				default:
1529 					break;
1530 				}
1531 				break;
1532 			}
1533 			dv = getdisk(buf, len, 1, &nswapdev);
1534 			if (dv) {
1535 				if (dv->dv_class == DV_IFNET)
1536 					nswapdev = NODEV;
1537 				if (nswapdev == nrootdev)
1538 					continue;
1539 				break;
1540 			}
1541 		}
1542 gotswap:
1543 		rootdev = nrootdev;
1544 		dumpdev = nswapdev;
1545 		swdevt[0].sw_dev = nswapdev;
1546 		swdevt[1].sw_dev = NODEV;
1547 #if defined(NFSCLIENT)
1548 	} else if (mountroot == nfs_mountroot) {
1549 		rootdv = bootdv;
1550 		rootdev = dumpdev = swapdev = NODEV;
1551 #endif
1552 	} else if (mountroot == NULL && rootdev == NODEV) {
1553 		/*
1554 		 * `swap generic'
1555 		 */
1556 		rootdv = bootdv;
1557 
1558 		if (bootdv->dv_class == DV_DISK) {
1559 			if (!duid_iszero(rootduid)) {
1560 				TAILQ_FOREACH(dk, &disklist, dk_link)
1561 					if ((dk->dk_flags & DKF_LABELVALID) &&
1562 					    dk->dk_label && duid_equal(
1563 					    dk->dk_label->d_uid, rootduid))
1564 						break;
1565 				if (dk == NULL)
1566 					panic("root device (%s) not found",
1567 					    duid_format(rootduid));
1568 				rootdv = dk->dk_device;
1569 			}
1570 		}
1571 
1572 		majdev = findblkmajor(rootdv);
1573 		if (majdev >= 0) {
1574 			/*
1575 			 * Root and swap are on the disk.
1576 			 * Assume swap is on partition b.
1577 			 */
1578 			rootdev = MAKEDISKDEV(majdev, rootdv->dv_unit, part);
1579 			nswapdev = MAKEDISKDEV(majdev, rootdv->dv_unit, 1);
1580 		} else {
1581 			/*
1582 			 * Root and swap are on a net.
1583 			 */
1584 			nswapdev = NODEV;
1585 		}
1586 		dumpdev = nswapdev;
1587 		swdevt[0].sw_dev = nswapdev;
1588 		/* swdevt[1].sw_dev = NODEV; */
1589 	} else {
1590 		/* Completely pre-configured, but we want rootdv .. */
1591 		majdev = major(rootdev);
1592 		if (findblkname(majdev) == NULL)
1593 			return;
1594 		unit = DISKUNIT(rootdev);
1595 		part = DISKPART(rootdev);
1596 		snprintf(buf, sizeof buf, "%s%d%c",
1597 		    findblkname(majdev), unit, 'a' + part);
1598 		rootdv = parsedisk(buf, strlen(buf), 0, &nrootdev);
1599 		if (rootdv == NULL)
1600 			panic("root device (%s) not found", buf);
1601 	}
1602 
1603 	if (bootdv != NULL && bootdv->dv_class == DV_IFNET)
1604 		ifp = if_unit(bootdv->dv_xname);
1605 
1606 	if (ifp) {
1607 		if_addgroup(ifp, "netboot");
1608 		if_put(ifp);
1609 	}
1610 
1611 	switch (rootdv->dv_class) {
1612 #if defined(NFSCLIENT)
1613 	case DV_IFNET:
1614 		mountroot = nfs_mountroot;
1615 		nfsbootdevname = rootdv->dv_xname;
1616 		return;
1617 #endif
1618 	case DV_DISK:
1619 		mountroot = dk_mountroot;
1620 		part = DISKPART(rootdev);
1621 		break;
1622 	default:
1623 		printf("can't figure root, hope your kernel is right\n");
1624 		return;
1625 	}
1626 
1627 	printf("root on %s%c", rootdv->dv_xname, 'a' + part);
1628 
1629 	if (dk && dk->dk_device == rootdv)
1630 		printf(" (%s.%c)", duid_format(rootduid), 'a' + part);
1631 
1632 	/*
1633 	 * Make the swap partition on the root drive the primary swap.
1634 	 */
1635 	for (swp = swdevt; swp->sw_dev != NODEV; swp++) {
1636 		if (major(rootdev) == major(swp->sw_dev) &&
1637 		    DISKUNIT(rootdev) == DISKUNIT(swp->sw_dev)) {
1638 			temp = swdevt[0].sw_dev;
1639 			swdevt[0].sw_dev = swp->sw_dev;
1640 			swp->sw_dev = temp;
1641 			break;
1642 		}
1643 	}
1644 	if (swp->sw_dev != NODEV) {
1645 		/*
1646 		 * If dumpdev was the same as the old primary swap device,
1647 		 * move it to the new primary swap device.
1648 		 */
1649 		if (temp == dumpdev)
1650 			dumpdev = swdevt[0].sw_dev;
1651 	}
1652 	if (swdevt[0].sw_dev != NODEV)
1653 		printf(" swap on %s%d%c", findblkname(major(swdevt[0].sw_dev)),
1654 		    DISKUNIT(swdevt[0].sw_dev),
1655 		    'a' + DISKPART(swdevt[0].sw_dev));
1656 	if (dumpdev != NODEV)
1657 		printf(" dump on %s%d%c", findblkname(major(dumpdev)),
1658 		    DISKUNIT(dumpdev), 'a' + DISKPART(dumpdev));
1659 	printf("\n");
1660 }
1661 
1662 extern struct nam2blk nam2blk[];
1663 
1664 int
1665 findblkmajor(struct device *dv)
1666 {
1667 	char buf[16], *p;
1668 	int i;
1669 
1670 	if (strlcpy(buf, dv->dv_xname, sizeof buf) >= sizeof buf)
1671 		return (-1);
1672 	for (p = buf; *p; p++)
1673 		if (*p >= '0' && *p <= '9')
1674 			*p = '\0';
1675 
1676 	for (i = 0; nam2blk[i].name; i++)
1677 		if (!strcmp(buf, nam2blk[i].name))
1678 			return (nam2blk[i].maj);
1679 	return (-1);
1680 }
1681 
1682 char *
1683 findblkname(int maj)
1684 {
1685 	int i;
1686 
1687 	for (i = 0; nam2blk[i].name; i++)
1688 		if (nam2blk[i].maj == maj)
1689 			return (nam2blk[i].name);
1690 	return (NULL);
1691 }
1692 
1693 char *
1694 disk_readlabel(struct disklabel *dl, dev_t dev, char *errbuf, size_t errsize)
1695 {
1696 	struct vnode *vn;
1697 	dev_t chrdev, rawdev;
1698 	int error;
1699 
1700 	chrdev = blktochr(dev);
1701 	rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(chrdev), RAW_PART);
1702 
1703 #ifdef DEBUG
1704 	printf("dev=0x%x chrdev=0x%x rawdev=0x%x\n", dev, chrdev, rawdev);
1705 #endif
1706 
1707 	if (cdevvp(rawdev, &vn)) {
1708 		snprintf(errbuf, errsize,
1709 		    "cannot obtain vnode for 0x%x/0x%x", dev, rawdev);
1710 		return (errbuf);
1711 	}
1712 
1713 	error = VOP_OPEN(vn, FREAD, NOCRED, curproc);
1714 	if (error) {
1715 		snprintf(errbuf, errsize,
1716 		    "cannot open disk, 0x%x/0x%x, error %d",
1717 		    dev, rawdev, error);
1718 		goto done;
1719 	}
1720 
1721 	error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)dl, FREAD, NOCRED, curproc);
1722 	if (error) {
1723 		snprintf(errbuf, errsize,
1724 		    "cannot read disk label, 0x%x/0x%x, error %d",
1725 		    dev, rawdev, error);
1726 	}
1727 done:
1728 	VOP_CLOSE(vn, FREAD, NOCRED, curproc);
1729 	vput(vn);
1730 	if (error)
1731 		return (errbuf);
1732 	return (NULL);
1733 }
1734 
1735 int
1736 disk_map(char *path, char *mappath, int size, int flags)
1737 {
1738 	struct disk *dk, *mdk;
1739 	u_char uid[8];
1740 	char c, part;
1741 	int i;
1742 
1743 	/*
1744 	 * Attempt to map a request for a disklabel UID to the correct device.
1745 	 * We should be supplied with a disklabel UID which has the following
1746 	 * format:
1747 	 *
1748 	 * [disklabel uid] . [partition]
1749 	 *
1750 	 * Alternatively, if the DM_OPENPART flag is set the disklabel UID can
1751 	 * based passed on its own.
1752 	 */
1753 
1754 	if (strchr(path, '/') != NULL)
1755 		return -1;
1756 
1757 	/* Verify that the device name is properly formed. */
1758 	if (!((strlen(path) == 16 && (flags & DM_OPENPART)) ||
1759 	    (strlen(path) == 18 && path[16] == '.')))
1760 		return -1;
1761 
1762 	/* Get partition. */
1763 	if (flags & DM_OPENPART)
1764 		part = 'a' + RAW_PART;
1765 	else
1766 		part = path[17];
1767 
1768 	if (part < 'a' || part >= 'a' + MAXPARTITIONS)
1769 		return -1;
1770 
1771 	/* Derive label UID. */
1772 	memset(uid, 0, sizeof(uid));
1773 	for (i = 0; i < 16; i++) {
1774 		c = path[i];
1775 		if (c >= '0' && c <= '9')
1776 			c -= '0';
1777 		else if (c >= 'a' && c <= 'f')
1778 			c -= ('a' - 10);
1779 		else
1780 			return -1;
1781 
1782 		uid[i / 2] <<= 4;
1783 		uid[i / 2] |= c & 0xf;
1784 	}
1785 
1786 	mdk = NULL;
1787 	TAILQ_FOREACH(dk, &disklist, dk_link) {
1788 		if ((dk->dk_flags & DKF_LABELVALID) && dk->dk_label &&
1789 		    memcmp(dk->dk_label->d_uid, uid,
1790 		    sizeof(dk->dk_label->d_uid)) == 0) {
1791 			/* Fail if there are duplicate UIDs! */
1792 			if (mdk != NULL)
1793 				return -1;
1794 			mdk = dk;
1795 		}
1796 	}
1797 
1798 	if (mdk == NULL || mdk->dk_name == NULL)
1799 		return -1;
1800 
1801 	snprintf(mappath, size, "/dev/%s%s%c",
1802 	    (flags & DM_OPENBLCK) ? "" : "r", mdk->dk_name, part);
1803 
1804 	return 0;
1805 }
1806 
1807 /*
1808  * Lookup a disk device and verify that it has completed attaching.
1809  */
1810 struct device *
1811 disk_lookup(struct cfdriver *cd, int unit)
1812 {
1813 	struct device *dv;
1814 	struct disk *dk;
1815 
1816 	dv = device_lookup(cd, unit);
1817 	if (dv == NULL)
1818 		return (NULL);
1819 
1820 	TAILQ_FOREACH(dk, &disklist, dk_link)
1821 		if (dk->dk_device == dv)
1822 			break;
1823 
1824 	if (dk == NULL) {
1825 		device_unref(dv);
1826 		return (NULL);
1827 	}
1828 
1829 	return (dv);
1830 }
1831 
1832 int
1833 duid_equal(u_char *duid1, u_char *duid2)
1834 {
1835 	return (memcmp(duid1, duid2, DUID_SIZE) == 0);
1836 }
1837 
1838 int
1839 duid_iszero(u_char *duid)
1840 {
1841 	u_char zeroduid[DUID_SIZE];
1842 
1843 	memset(zeroduid, 0, sizeof(zeroduid));
1844 
1845 	return (duid_equal(duid, zeroduid));
1846 }
1847 
1848 const char *
1849 duid_format(u_char *duid)
1850 {
1851 	static char duid_str[17];
1852 
1853 	KERNEL_ASSERT_LOCKED();
1854 
1855 	snprintf(duid_str, sizeof(duid_str),
1856 	    "%02x%02x%02x%02x%02x%02x%02x%02x",
1857 	    duid[0], duid[1], duid[2], duid[3],
1858 	    duid[4], duid[5], duid[6], duid[7]);
1859 
1860 	return (duid_str);
1861 }
1862