xref: /freebsd/stand/i386/libi386/biosdisk.c (revision ca987d4641cdcd7f27e153db17c5bf064934faf5)
1 /*-
2  * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
3  * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 /*
32  * BIOS disk device handling.
33  *
34  * Ideas and algorithms from:
35  *
36  * - NetBSD libi386/biosdisk.c
37  * - FreeBSD biosboot/disk.c
38  *
39  */
40 
41 #include <sys/disk.h>
42 #include <sys/limits.h>
43 #include <stand.h>
44 #include <machine/bootinfo.h>
45 #include <stdarg.h>
46 
47 #include <bootstrap.h>
48 #include <btxv86.h>
49 #include <edd.h>
50 #include "disk.h"
51 #include "libi386.h"
52 
53 #ifdef LOADER_GELI_SUPPORT
54 #include "cons.h"
55 #include "drv.h"
56 #include "gpt.h"
57 #include "part.h"
58 #include <uuid.h>
59 struct pentry {
60 	struct ptable_entry	part;
61 	uint64_t		flags;
62 	union {
63 		uint8_t bsd;
64 		uint8_t	mbr;
65 		uuid_t	gpt;
66 		uint16_t vtoc8;
67 	} type;
68 	STAILQ_ENTRY(pentry)	entry;
69 };
70 struct ptable {
71 	enum ptable_type	type;
72 	uint16_t		sectorsize;
73 	uint64_t		sectors;
74 
75 	STAILQ_HEAD(, pentry)	entries;
76 };
77 
78 #include "geliboot.c"
79 #endif /* LOADER_GELI_SUPPORT */
80 
81 CTASSERT(sizeof(struct i386_devdesc) >= sizeof(struct disk_devdesc));
82 
83 #define BIOS_NUMDRIVES		0x475
84 #define BIOSDISK_SECSIZE	512
85 #define BUFSIZE			(1 * BIOSDISK_SECSIZE)
86 
87 #define DT_ATAPI		0x10		/* disk type for ATAPI floppies */
88 #define WDMAJOR			0		/* major numbers for devices we frontend for */
89 #define WFDMAJOR		1
90 #define FDMAJOR			2
91 #define DAMAJOR			4
92 
93 #ifdef DISK_DEBUG
94 # define DEBUG(fmt, args...)	printf("%s: " fmt "\n" , __func__ , ## args)
95 #else
96 # define DEBUG(fmt, args...)
97 #endif
98 
99 /*
100  * List of BIOS devices, translation from disk unit number to
101  * BIOS unit number.
102  */
103 static struct bdinfo
104 {
105 	int		bd_unit;	/* BIOS unit number */
106 	int		bd_cyl;		/* BIOS geometry */
107 	int		bd_hds;
108 	int		bd_sec;
109 	int		bd_flags;
110 #define	BD_MODEINT13	0x0000
111 #define	BD_MODEEDD1	0x0001
112 #define	BD_MODEEDD3	0x0002
113 #define	BD_MODEMASK	0x0003
114 #define	BD_FLOPPY	0x0004
115 	int		bd_type;	/* BIOS 'drive type' (floppy only) */
116 	uint16_t	bd_sectorsize;	/* Sector size */
117 	uint64_t	bd_sectors;	/* Disk size */
118 	int		bd_open;	/* reference counter */
119 	void		*bd_bcache;	/* buffer cache data */
120 } bdinfo [MAXBDDEV];
121 static int nbdinfo = 0;
122 
123 #define	BD(dev)		(bdinfo[(dev)->d_unit])
124 
125 static int bd_read(struct disk_devdesc *dev, daddr_t dblk, int blks,
126     caddr_t dest);
127 static int bd_write(struct disk_devdesc *dev, daddr_t dblk, int blks,
128     caddr_t dest);
129 static int bd_int13probe(struct bdinfo *bd);
130 
131 static int bd_init(void);
132 static int bd_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
133     char *buf, size_t *rsize);
134 static int bd_realstrategy(void *devdata, int flag, daddr_t dblk, size_t size,
135     char *buf, size_t *rsize);
136 static int bd_open(struct open_file *f, ...);
137 static int bd_close(struct open_file *f);
138 static int bd_ioctl(struct open_file *f, u_long cmd, void *data);
139 static int bd_print(int verbose);
140 
141 #ifdef LOADER_GELI_SUPPORT
142 enum isgeli {
143 	ISGELI_UNKNOWN,
144 	ISGELI_NO,
145 	ISGELI_YES
146 };
147 static enum isgeli geli_status[MAXBDDEV][MAXTBLENTS];
148 
149 int bios_read(void *vdev __unused, struct dsk *priv, off_t off, char *buf,
150     size_t bytes);
151 #endif /* LOADER_GELI_SUPPORT */
152 
153 struct devsw biosdisk = {
154 	"disk",
155 	DEVT_DISK,
156 	bd_init,
157 	bd_strategy,
158 	bd_open,
159 	bd_close,
160 	bd_ioctl,
161 	bd_print,
162 	NULL
163 };
164 
165 /*
166  * Translate between BIOS device numbers and our private unit numbers.
167  */
168 int
169 bd_bios2unit(int biosdev)
170 {
171 	int i;
172 
173 	DEBUG("looking for bios device 0x%x", biosdev);
174 	for (i = 0; i < nbdinfo; i++) {
175 		DEBUG("bd unit %d is BIOS device 0x%x", i, bdinfo[i].bd_unit);
176 		if (bdinfo[i].bd_unit == biosdev)
177 			return (i);
178 	}
179 	return (-1);
180 }
181 
182 int
183 bd_unit2bios(int unit)
184 {
185 
186 	if ((unit >= 0) && (unit < nbdinfo))
187 		return (bdinfo[unit].bd_unit);
188 	return (-1);
189 }
190 
191 /*
192  * Quiz the BIOS for disk devices, save a little info about them.
193  */
194 static int
195 bd_init(void)
196 {
197 	int base, unit, nfd = 0;
198 
199 #ifdef LOADER_GELI_SUPPORT
200 	geli_init();
201 #endif
202 	/* sequence 0, 0x80 */
203 	for (base = 0; base <= 0x80; base += 0x80) {
204 		for (unit = base; (nbdinfo < MAXBDDEV); unit++) {
205 #ifndef VIRTUALBOX
206 			/*
207 			 * Check the BIOS equipment list for number
208 			 * of fixed disks.
209 			 */
210 			if(base == 0x80 &&
211 			    (nfd >= *(unsigned char *)PTOV(BIOS_NUMDRIVES)))
212 				break;
213 #endif
214 			bdinfo[nbdinfo].bd_open = 0;
215 			bdinfo[nbdinfo].bd_bcache = NULL;
216 			bdinfo[nbdinfo].bd_unit = unit;
217 			bdinfo[nbdinfo].bd_flags = unit < 0x80 ? BD_FLOPPY: 0;
218 			if (!bd_int13probe(&bdinfo[nbdinfo]))
219 				break;
220 
221 			/* XXX we need "disk aliases" to make this simpler */
222 			printf("BIOS drive %c: is disk%d\n", (unit < 0x80) ?
223 			    ('A' + unit): ('C' + unit - 0x80), nbdinfo);
224 			nbdinfo++;
225 			if (base == 0x80)
226 				nfd++;
227 		}
228 	}
229 	bcache_add_dev(nbdinfo);
230 	return(0);
231 }
232 
233 /*
234  * Try to detect a device supported by the legacy int13 BIOS
235  */
236 static int
237 bd_int13probe(struct bdinfo *bd)
238 {
239 	struct edd_params params;
240 	int ret = 1;	/* assume success */
241 
242 	v86.ctl = V86_FLAGS;
243 	v86.addr = 0x13;
244 	v86.eax = 0x800;
245 	v86.edx = bd->bd_unit;
246 	v86int();
247 
248 	/* Don't error out if we get bad sector number, try EDD as well */
249 	if (V86_CY(v86.efl) ||	/* carry set */
250 	    (v86.edx & 0xff) <= (unsigned)(bd->bd_unit & 0x7f))	/* unit # bad */
251 		return (0);	/* skip device */
252 
253 	if ((v86.ecx & 0x3f) == 0) /* absurd sector number */
254 		ret = 0;	/* set error */
255 
256 	/* Convert max cyl # -> # of cylinders */
257 	bd->bd_cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
258 	/* Convert max head # -> # of heads */
259 	bd->bd_hds = ((v86.edx & 0xff00) >> 8) + 1;
260 	bd->bd_sec = v86.ecx & 0x3f;
261 	bd->bd_type = v86.ebx & 0xff;
262 	bd->bd_flags |= BD_MODEINT13;
263 
264 	/* Calculate sectors count from the geometry */
265 	bd->bd_sectors = bd->bd_cyl * bd->bd_hds * bd->bd_sec;
266 	bd->bd_sectorsize = BIOSDISK_SECSIZE;
267 	DEBUG("unit 0x%x geometry %d/%d/%d", bd->bd_unit, bd->bd_cyl,
268 	    bd->bd_hds, bd->bd_sec);
269 
270 	/* Determine if we can use EDD with this device. */
271 	v86.ctl = V86_FLAGS;
272 	v86.addr = 0x13;
273 	v86.eax = 0x4100;
274 	v86.edx = bd->bd_unit;
275 	v86.ebx = 0x55aa;
276 	v86int();
277 	if (V86_CY(v86.efl) ||	/* carry set */
278 	    (v86.ebx & 0xffff) != 0xaa55 || /* signature */
279 	    (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
280 		return (ret);	/* return code from int13 AH=08 */
281 
282 	/* EDD supported */
283 	bd->bd_flags |= BD_MODEEDD1;
284 	if ((v86.eax & 0xff00) >= 0x3000)
285 		bd->bd_flags |= BD_MODEEDD3;
286 	/* Get disk params */
287 	params.len = sizeof(struct edd_params);
288 	v86.ctl = V86_FLAGS;
289 	v86.addr = 0x13;
290 	v86.eax = 0x4800;
291 	v86.edx = bd->bd_unit;
292 	v86.ds = VTOPSEG(&params);
293 	v86.esi = VTOPOFF(&params);
294 	v86int();
295 	if (!V86_CY(v86.efl)) {
296 		uint64_t total;
297 
298 		/*
299 		 * Sector size must be a multiple of 512 bytes.
300 		 * An alternate test would be to check power of 2,
301 		 * powerof2(params.sector_size).
302 		 */
303 		if (params.sector_size % BIOSDISK_SECSIZE)
304 			bd->bd_sectorsize = BIOSDISK_SECSIZE;
305 		else
306 			bd->bd_sectorsize = params.sector_size;
307 
308 		total = bd->bd_sectorsize * params.sectors;
309 		if (params.sectors != 0) {
310 			/* Only update if we did not overflow. */
311 			if (total > params.sectors)
312 				bd->bd_sectors = params.sectors;
313 		}
314 
315 		total = (uint64_t)params.cylinders *
316 		    params.heads * params.sectors_per_track;
317 		if (bd->bd_sectors < total)
318 			bd->bd_sectors = total;
319 
320 		ret = 1;
321 	}
322 	DEBUG("unit 0x%x flags %x, sectors %llu, sectorsize %u",
323 	    bd->bd_unit, bd->bd_flags, bd->bd_sectors, bd->bd_sectorsize);
324 	return (ret);
325 }
326 
327 /*
328  * Print information about disks
329  */
330 static int
331 bd_print(int verbose)
332 {
333 	static char line[80];
334 	struct disk_devdesc dev;
335 	int i, ret = 0;
336 
337 	if (nbdinfo == 0)
338 		return (0);
339 
340 	printf("%s devices:", biosdisk.dv_name);
341 	if ((ret = pager_output("\n")) != 0)
342 		return (ret);
343 
344 	for (i = 0; i < nbdinfo; i++) {
345 		snprintf(line, sizeof(line),
346 		    "    disk%d:   BIOS drive %c (%ju X %u):\n", i,
347 		    (bdinfo[i].bd_unit < 0x80) ? ('A' + bdinfo[i].bd_unit):
348 		    ('C' + bdinfo[i].bd_unit - 0x80),
349 		    (uintmax_t)bdinfo[i].bd_sectors,
350 		    bdinfo[i].bd_sectorsize);
351 		if ((ret = pager_output(line)) != 0)
352 			break;
353 		dev.d_dev = &biosdisk;
354 		dev.d_unit = i;
355 		dev.d_slice = -1;
356 		dev.d_partition = -1;
357 		if (disk_open(&dev,
358 		    bdinfo[i].bd_sectorsize * bdinfo[i].bd_sectors,
359 		    bdinfo[i].bd_sectorsize) == 0) {
360 			snprintf(line, sizeof(line), "    disk%d", i);
361 			ret = disk_print(&dev, line, verbose);
362 			disk_close(&dev);
363 			if (ret != 0)
364 			    return (ret);
365 		}
366 	}
367 	return (ret);
368 }
369 
370 /*
371  * Attempt to open the disk described by (dev) for use by (f).
372  *
373  * Note that the philosophy here is "give them exactly what
374  * they ask for".  This is necessary because being too "smart"
375  * about what the user might want leads to complications.
376  * (eg. given no slice or partition value, with a disk that is
377  *  sliced - are they after the first BSD slice, or the DOS
378  *  slice before it?)
379  */
380 static int
381 bd_open(struct open_file *f, ...)
382 {
383 	struct disk_devdesc *dev, rdev;
384 	struct disk_devdesc disk;
385 	int err, g_err;
386 	va_list ap;
387 	uint64_t size;
388 
389 	va_start(ap, f);
390 	dev = va_arg(ap, struct disk_devdesc *);
391 	va_end(ap);
392 
393 	if (dev->d_unit < 0 || dev->d_unit >= nbdinfo)
394 		return (EIO);
395 	BD(dev).bd_open++;
396 	if (BD(dev).bd_bcache == NULL)
397 	    BD(dev).bd_bcache = bcache_allocate();
398 
399 	/*
400 	 * Read disk size from partition.
401 	 * This is needed to work around buggy BIOS systems returning
402 	 * wrong (truncated) disk media size.
403 	 * During bd_probe() we tested if the mulitplication of bd_sectors
404 	 * would overflow so it should be safe to perform here.
405 	 */
406 	disk.d_dev = dev->d_dev;
407 	disk.d_type = dev->d_type;
408 	disk.d_unit = dev->d_unit;
409 	disk.d_opendata = NULL;
410 	disk.d_slice = -1;
411 	disk.d_partition = -1;
412 	disk.d_offset = 0;
413 	if (disk_open(&disk, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
414 	    BD(dev).bd_sectorsize) == 0) {
415 
416 		if (disk_ioctl(&disk, DIOCGMEDIASIZE, &size) == 0) {
417 			size /= BD(dev).bd_sectorsize;
418 			if (size > BD(dev).bd_sectors)
419 				BD(dev).bd_sectors = size;
420 		}
421 		disk_close(&disk);
422 	}
423 
424 	err = disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
425 	    BD(dev).bd_sectorsize);
426 
427 #ifdef LOADER_GELI_SUPPORT
428 	static char gelipw[GELI_PW_MAXLEN];
429 	char *passphrase;
430 
431 	if (err)
432 		return (err);
433 
434 	/* if we already know there is no GELI, skip the rest */
435 	if (geli_status[dev->d_unit][dev->d_slice] != ISGELI_UNKNOWN)
436 		return (err);
437 
438 	struct dsk dskp;
439 	struct ptable *table = NULL;
440 	struct ptable_entry part;
441 	struct pentry *entry;
442 	int geli_part = 0;
443 
444 	dskp.drive = bd_unit2bios(dev->d_unit);
445 	dskp.type = dev->d_type;
446 	dskp.unit = dev->d_unit;
447 	dskp.slice = dev->d_slice;
448 	dskp.part = dev->d_partition;
449 	dskp.start = dev->d_offset;
450 
451 	memcpy(&rdev, dev, sizeof(rdev));
452 	/* to read the GPT table, we need to read the first sector */
453 	rdev.d_offset = 0;
454 	/* We need the LBA of the end of the partition */
455 	table = ptable_open(&rdev, BD(dev).bd_sectors,
456 	    BD(dev).bd_sectorsize, ptblread);
457 	if (table == NULL) {
458 		DEBUG("Can't read partition table");
459 		/* soft failure, return the exit status of disk_open */
460 		return (err);
461 	}
462 
463 	if (table->type == PTABLE_GPT)
464 		dskp.part = 255;
465 
466 	STAILQ_FOREACH(entry, &table->entries, entry) {
467 		dskp.slice = entry->part.index;
468 		dskp.start = entry->part.start;
469 		if (is_geli(&dskp) == 0) {
470 			geli_status[dev->d_unit][dskp.slice] = ISGELI_YES;
471 			return (0);
472 		}
473 		if (geli_taste(bios_read, &dskp,
474 		    entry->part.end - entry->part.start) == 0) {
475 			if (geli_havekey(&dskp) == 0) {
476 				geli_status[dev->d_unit][dskp.slice] = ISGELI_YES;
477 				geli_part++;
478 				continue;
479 			}
480 			if ((passphrase = getenv("kern.geom.eli.passphrase"))
481 			    != NULL) {
482 				/* Use the cached passphrase */
483 				bcopy(passphrase, &gelipw, GELI_PW_MAXLEN);
484 			}
485 			if (geli_passphrase(&gelipw, dskp.unit, 'p',
486 				    (dskp.slice > 0 ? dskp.slice : dskp.part),
487 				    &dskp) == 0) {
488 				setenv("kern.geom.eli.passphrase", &gelipw, 1);
489 				bzero(gelipw, sizeof(gelipw));
490 				geli_status[dev->d_unit][dskp.slice] = ISGELI_YES;
491 				geli_part++;
492 				continue;
493 			}
494 		} else
495 			geli_status[dev->d_unit][dskp.slice] = ISGELI_NO;
496 	}
497 
498 	/* none of the partitions on this disk have GELI */
499 	if (geli_part == 0) {
500 		/* found no GELI */
501 		geli_status[dev->d_unit][dev->d_slice] = ISGELI_NO;
502 	}
503 #endif /* LOADER_GELI_SUPPORT */
504 
505 	return (err);
506 }
507 
508 static int
509 bd_close(struct open_file *f)
510 {
511 	struct disk_devdesc *dev;
512 
513 	dev = (struct disk_devdesc *)f->f_devdata;
514 	BD(dev).bd_open--;
515 	if (BD(dev).bd_open == 0) {
516 	    bcache_free(BD(dev).bd_bcache);
517 	    BD(dev).bd_bcache = NULL;
518 	}
519 	return (disk_close(dev));
520 }
521 
522 static int
523 bd_ioctl(struct open_file *f, u_long cmd, void *data)
524 {
525 	struct disk_devdesc *dev;
526 	int rc;
527 
528 	dev = (struct disk_devdesc *)f->f_devdata;
529 
530 	rc = disk_ioctl(dev, cmd, data);
531 	if (rc != ENOTTY)
532 		return (rc);
533 
534 	switch (cmd) {
535 	case DIOCGSECTORSIZE:
536 		*(u_int *)data = BD(dev).bd_sectorsize;
537 		break;
538 	case DIOCGMEDIASIZE:
539 		*(uint64_t *)data = BD(dev).bd_sectors * BD(dev).bd_sectorsize;
540 		break;
541 	default:
542 		return (ENOTTY);
543 	}
544 	return (0);
545 }
546 
547 static int
548 bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
549     char *buf, size_t *rsize)
550 {
551 	struct bcache_devdata bcd;
552 	struct disk_devdesc *dev;
553 
554 	dev = (struct disk_devdesc *)devdata;
555 	bcd.dv_strategy = bd_realstrategy;
556 	bcd.dv_devdata = devdata;
557 	bcd.dv_cache = BD(dev).bd_bcache;
558 	return (bcache_strategy(&bcd, rw, dblk + dev->d_offset,
559 	    size, buf, rsize));
560 }
561 
562 static int
563 bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t size,
564     char *buf, size_t *rsize)
565 {
566     struct disk_devdesc *dev = (struct disk_devdesc *)devdata;
567     uint64_t		disk_blocks;
568     int			blks, rc;
569 #ifdef BD_SUPPORT_FRAGS /* XXX: sector size */
570     char		fragbuf[BIOSDISK_SECSIZE];
571     size_t		fragsize;
572 
573     fragsize = size % BIOSDISK_SECSIZE;
574 #else
575     if (size % BD(dev).bd_sectorsize)
576 	panic("bd_strategy: %d bytes I/O not multiple of block size", size);
577 #endif
578 
579     DEBUG("open_disk %p", dev);
580 
581     /*
582      * Check the value of the size argument. We do have quite small
583      * heap (64MB), but we do not know good upper limit, so we check against
584      * INT_MAX here. This will also protect us against possible overflows
585      * while translating block count to bytes.
586      */
587     if (size > INT_MAX) {
588 	DEBUG("too large read: %zu bytes", size);
589 	return (EIO);
590     }
591 
592     blks = size / BD(dev).bd_sectorsize;
593     if (dblk > dblk + blks)
594 	return (EIO);
595 
596     if (rsize)
597 	*rsize = 0;
598 
599     /* Get disk blocks, this value is either for whole disk or for partition */
600     if (disk_ioctl(dev, DIOCGMEDIASIZE, &disk_blocks)) {
601 	/* DIOCGMEDIASIZE does return bytes. */
602         disk_blocks /= BD(dev).bd_sectorsize;
603     } else {
604 	/* We should not get here. Just try to survive. */
605 	disk_blocks = BD(dev).bd_sectors - dev->d_offset;
606     }
607 
608     /* Validate source block address. */
609     if (dblk < dev->d_offset || dblk >= dev->d_offset + disk_blocks)
610 	return (EIO);
611 
612     /*
613      * Truncate if we are crossing disk or partition end.
614      */
615     if (dblk + blks >= dev->d_offset + disk_blocks) {
616 	blks = dev->d_offset + disk_blocks - dblk;
617 	size = blks * BD(dev).bd_sectorsize;
618 	DEBUG("short read %d", blks);
619     }
620 
621     switch (rw & F_MASK) {
622     case F_READ:
623 	DEBUG("read %d from %lld to %p", blks, dblk, buf);
624 
625 	if (blks && (rc = bd_read(dev, dblk, blks, buf))) {
626 	    /* Filter out floppy controller errors */
627 	    if (BD(dev).bd_flags != BD_FLOPPY || rc != 0x20) {
628 		printf("read %d from %lld to %p, error: 0x%x", blks, dblk,
629 		    buf, rc);
630 	    }
631 	    return (EIO);
632 	}
633 #ifdef BD_SUPPORT_FRAGS /* XXX: sector size */
634 	DEBUG("bd_strategy: frag read %d from %d+%d to %p",
635 	    fragsize, dblk, blks, buf + (blks * BIOSDISK_SECSIZE));
636 	if (fragsize && bd_read(od, dblk + blks, 1, fragsize)) {
637 	    DEBUG("frag read error");
638 	    return(EIO);
639 	}
640 	bcopy(fragbuf, buf + (blks * BIOSDISK_SECSIZE), fragsize);
641 #endif
642 	break;
643     case F_WRITE :
644 	DEBUG("write %d from %d to %p", blks, dblk, buf);
645 
646 	if (blks && bd_write(dev, dblk, blks, buf)) {
647 	    DEBUG("write error");
648 	    return (EIO);
649 	}
650 #ifdef BD_SUPPORT_FRAGS
651 	if(fragsize) {
652 	    DEBUG("Attempted to write a frag");
653 	    return (EIO);
654 	}
655 #endif
656 	break;
657     default:
658 	/* DO NOTHING */
659 	return (EROFS);
660     }
661 
662     if (rsize)
663 	*rsize = size;
664     return (0);
665 }
666 
667 static int
668 bd_edd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
669     int write)
670 {
671     static struct edd_packet packet;
672 
673     packet.len = sizeof(struct edd_packet);
674     packet.count = blks;
675     packet.off = VTOPOFF(dest);
676     packet.seg = VTOPSEG(dest);
677     packet.lba = dblk;
678     v86.ctl = V86_FLAGS;
679     v86.addr = 0x13;
680     if (write)
681 	/* Should we Write with verify ?? 0x4302 ? */
682 	v86.eax = 0x4300;
683     else
684 	v86.eax = 0x4200;
685     v86.edx = BD(dev).bd_unit;
686     v86.ds = VTOPSEG(&packet);
687     v86.esi = VTOPOFF(&packet);
688     v86int();
689     if (V86_CY(v86.efl))
690 	return (v86.eax >> 8);
691     return (0);
692 }
693 
694 static int
695 bd_chs_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest,
696     int write)
697 {
698     u_int	x, bpc, cyl, hd, sec;
699 
700     bpc = BD(dev).bd_sec * BD(dev).bd_hds;	/* blocks per cylinder */
701     x = dblk;
702     cyl = x / bpc;			/* block # / blocks per cylinder */
703     x %= bpc;				/* block offset into cylinder */
704     hd = x / BD(dev).bd_sec;		/* offset / blocks per track */
705     sec = x % BD(dev).bd_sec;		/* offset into track */
706 
707     /* correct sector number for 1-based BIOS numbering */
708     sec++;
709 
710     if (cyl > 1023)
711 	/* CHS doesn't support cylinders > 1023. */
712 	return (1);
713 
714     v86.ctl = V86_FLAGS;
715     v86.addr = 0x13;
716     if (write)
717 	v86.eax = 0x300 | blks;
718     else
719 	v86.eax = 0x200 | blks;
720     v86.ecx = ((cyl & 0xff) << 8) | ((cyl & 0x300) >> 2) | sec;
721     v86.edx = (hd << 8) | BD(dev).bd_unit;
722     v86.es = VTOPSEG(dest);
723     v86.ebx = VTOPOFF(dest);
724     v86int();
725     if (V86_CY(v86.efl))
726 	return (v86.eax >> 8);
727     return (0);
728 }
729 
730 static int
731 bd_io(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest, int write)
732 {
733     u_int	x, sec, result, resid, retry, maxfer;
734     caddr_t	p, xp, bbuf;
735 
736     /* Just in case some idiot actually tries to read/write -1 blocks... */
737     if (blks < 0)
738 	return (-1);
739 
740     resid = blks;
741     p = dest;
742 
743     /* Decide whether we have to bounce */
744     if (VTOP(dest) >> 20 != 0 || (BD(dev).bd_unit < 0x80 &&
745 	(VTOP(dest) >> 16) != (VTOP(dest +
746 	blks * BD(dev).bd_sectorsize) >> 16))) {
747 
748 	/*
749 	 * There is a 64k physical boundary somewhere in the
750 	 * destination buffer, or the destination buffer is above
751 	 * first 1MB of physical memory so we have to arrange a
752 	 * suitable bounce buffer.  Allocate a buffer twice as large
753 	 * as we need to.  Use the bottom half unless there is a break
754 	 * there, in which case we use the top half.
755 	 */
756 	x = V86_IO_BUFFER_SIZE / BD(dev).bd_sectorsize;
757 	x = min(x, (unsigned)blks);
758 	bbuf = PTOV(V86_IO_BUFFER);
759 	maxfer = x;		/* limit transfers to bounce region size */
760     } else {
761 	bbuf = NULL;
762 	maxfer = 0;
763     }
764 
765     while (resid > 0) {
766 	/*
767 	 * Play it safe and don't cross track boundaries.
768 	 * (XXX this is probably unnecessary)
769 	 */
770 	sec = dblk % BD(dev).bd_sec;	/* offset into track */
771 	x = min(BD(dev).bd_sec - sec, resid);
772 	if (maxfer > 0)
773 	    x = min(x, maxfer);		/* fit bounce buffer */
774 
775 	/* where do we transfer to? */
776 	xp = bbuf == NULL ? p : bbuf;
777 
778 	/*
779 	 * Put your Data In, Put your Data out,
780 	 * Put your Data In, and shake it all about
781 	 */
782 	if (write && bbuf != NULL)
783 	    bcopy(p, bbuf, x * BD(dev).bd_sectorsize);
784 
785 	/*
786 	 * Loop retrying the operation a couple of times.  The BIOS
787 	 * may also retry.
788 	 */
789 	for (retry = 0; retry < 3; retry++) {
790 	    /* if retrying, reset the drive */
791 	    if (retry > 0) {
792 		v86.ctl = V86_FLAGS;
793 		v86.addr = 0x13;
794 		v86.eax = 0;
795 		v86.edx = BD(dev).bd_unit;
796 		v86int();
797 	    }
798 
799 	    if (BD(dev).bd_flags & BD_MODEEDD1)
800 		result = bd_edd_io(dev, dblk, x, xp, write);
801 	    else
802 		result = bd_chs_io(dev, dblk, x, xp, write);
803 	    if (result == 0)
804 		break;
805 	}
806 
807 	if (write)
808 	    DEBUG("Write %d sector(s) from %p (0x%x) to %lld %s", x,
809 		p, VTOP(p), dblk, result ? "failed" : "ok");
810 	else
811 	    DEBUG("Read %d sector(s) from %lld to %p (0x%x) %s", x,
812 		dblk, p, VTOP(p), result ? "failed" : "ok");
813 	if (result) {
814 	    return (result);
815 	}
816 	if (!write && bbuf != NULL)
817 	    bcopy(bbuf, p, x * BD(dev).bd_sectorsize);
818 	p += (x * BD(dev).bd_sectorsize);
819 	dblk += x;
820 	resid -= x;
821     }
822 
823 /*    hexdump(dest, (blks * BD(dev).bd_sectorsize)); */
824     return(0);
825 }
826 
827 static int
828 bd_read(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest)
829 {
830 #ifdef LOADER_GELI_SUPPORT
831 	struct dsk dskp;
832 	off_t p_off, diff;
833 	daddr_t alignlba;
834 	int err, n, alignblks;
835 	char *tmpbuf;
836 
837 	/* if we already know there is no GELI, skip the rest */
838 	if (geli_status[dev->d_unit][dev->d_slice] != ISGELI_YES)
839 		return (bd_io(dev, dblk, blks, dest, 0));
840 
841 	if (geli_status[dev->d_unit][dev->d_slice] == ISGELI_YES) {
842 		/*
843 		 * Align reads to DEV_GELIBOOT_BSIZE bytes because partial
844 		 * sectors cannot be decrypted. Round the requested LBA down to
845 		 * nearest multiple of DEV_GELIBOOT_BSIZE bytes.
846 		 */
847 		alignlba = rounddown2(dblk * BD(dev).bd_sectorsize,
848 		    DEV_GELIBOOT_BSIZE) / BD(dev).bd_sectorsize;
849 		/*
850 		 * Round number of blocks to read up to nearest multiple of
851 		 * DEV_GELIBOOT_BSIZE
852 		 */
853 		diff = (dblk - alignlba) * BD(dev).bd_sectorsize;
854 		alignblks = roundup2(blks * BD(dev).bd_sectorsize + diff,
855 		    DEV_GELIBOOT_BSIZE) / BD(dev).bd_sectorsize;
856 
857 		/*
858 		 * If the read is rounded up to a larger size, use a temporary
859 		 * buffer here because the buffer provided by the caller may be
860 		 * too small.
861 		 */
862 		if (diff == 0) {
863 			tmpbuf = dest;
864 		} else {
865 			tmpbuf = malloc(alignblks * BD(dev).bd_sectorsize);
866 			if (tmpbuf == NULL) {
867 				return (-1);
868 			}
869 		}
870 
871 		err = bd_io(dev, alignlba, alignblks, tmpbuf, 0);
872 		if (err)
873 			return (err);
874 
875 		dskp.drive = bd_unit2bios(dev->d_unit);
876 		dskp.type = dev->d_type;
877 		dskp.unit = dev->d_unit;
878 		dskp.slice = dev->d_slice;
879 		dskp.part = dev->d_partition;
880 		dskp.start = dev->d_offset;
881 
882 		/* GELI needs the offset relative to the partition start */
883 		p_off = alignlba - dskp.start;
884 
885 		err = geli_read(&dskp, p_off * BD(dev).bd_sectorsize, tmpbuf,
886 		    alignblks * BD(dev).bd_sectorsize);
887 		if (err)
888 			return (err);
889 
890 		if (tmpbuf != dest) {
891 			bcopy(tmpbuf + diff, dest, blks * BD(dev).bd_sectorsize);
892 			free(tmpbuf);
893 		}
894 		return (0);
895 	}
896 #endif /* LOADER_GELI_SUPPORT */
897 
898 	return (bd_io(dev, dblk, blks, dest, 0));
899 }
900 
901 static int
902 bd_write(struct disk_devdesc *dev, daddr_t dblk, int blks, caddr_t dest)
903 {
904 
905 	return (bd_io(dev, dblk, blks, dest, 1));
906 }
907 
908 /*
909  * Return the BIOS geometry of a given "fixed drive" in a format
910  * suitable for the legacy bootinfo structure.  Since the kernel is
911  * expecting raw int 0x13/0x8 values for N_BIOS_GEOM drives, we
912  * prefer to get the information directly, rather than rely on being
913  * able to put it together from information already maintained for
914  * different purposes and for a probably different number of drives.
915  *
916  * For valid drives, the geometry is expected in the format (31..0)
917  * "000000cc cccccccc hhhhhhhh 00ssssss"; and invalid drives are
918  * indicated by returning the geometry of a "1.2M" PC-format floppy
919  * disk.  And, incidentally, what is returned is not the geometry as
920  * such but the highest valid cylinder, head, and sector numbers.
921  */
922 u_int32_t
923 bd_getbigeom(int bunit)
924 {
925 
926     v86.ctl = V86_FLAGS;
927     v86.addr = 0x13;
928     v86.eax = 0x800;
929     v86.edx = 0x80 + bunit;
930     v86int();
931     if (V86_CY(v86.efl))
932 	return 0x4f010f;
933     return ((v86.ecx & 0xc0) << 18) | ((v86.ecx & 0xff00) << 8) |
934 	   (v86.edx & 0xff00) | (v86.ecx & 0x3f);
935 }
936 
937 /*
938  * Return a suitable dev_t value for (dev).
939  *
940  * In the case where it looks like (dev) is a SCSI disk, we allow the number of
941  * IDE disks to be specified in $num_ide_disks.  There should be a Better Way.
942  */
943 int
944 bd_getdev(struct i386_devdesc *d)
945 {
946     struct disk_devdesc		*dev;
947     int				biosdev;
948     int 			major;
949     int				rootdev;
950     char			*nip, *cp;
951     int				i, unit;
952 
953     dev = (struct disk_devdesc *)d;
954     biosdev = bd_unit2bios(dev->d_unit);
955     DEBUG("unit %d BIOS device %d", dev->d_unit, biosdev);
956     if (biosdev == -1)				/* not a BIOS device */
957 	return(-1);
958     if (disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
959 	BD(dev).bd_sectorsize) != 0)		/* oops, not a viable device */
960 	    return (-1);
961     else
962 	disk_close(dev);
963 
964     if (biosdev < 0x80) {
965 	/* floppy (or emulated floppy) or ATAPI device */
966 	if (bdinfo[dev->d_unit].bd_type == DT_ATAPI) {
967 	    /* is an ATAPI disk */
968 	    major = WFDMAJOR;
969 	} else {
970 	    /* is a floppy disk */
971 	    major = FDMAJOR;
972 	}
973     } else {
974 	    /* assume an IDE disk */
975 	    major = WDMAJOR;
976     }
977     /* default root disk unit number */
978     unit = biosdev & 0x7f;
979 
980     /* XXX a better kludge to set the root disk unit number */
981     if ((nip = getenv("root_disk_unit")) != NULL) {
982 	i = strtol(nip, &cp, 0);
983 	/* check for parse error */
984 	if ((cp != nip) && (*cp == 0))
985 	    unit = i;
986     }
987 
988     rootdev = MAKEBOOTDEV(major, dev->d_slice + 1, unit, dev->d_partition);
989     DEBUG("dev is 0x%x\n", rootdev);
990     return(rootdev);
991 }
992 
993 #ifdef LOADER_GELI_SUPPORT
994 int
995 bios_read(void *vdev __unused, struct dsk *priv, off_t off, char *buf, size_t bytes)
996 {
997 	struct disk_devdesc dev;
998 
999 	dev.d_dev = &biosdisk;
1000 	dev.d_type = priv->type;
1001 	dev.d_unit = priv->unit;
1002 	dev.d_slice = priv->slice;
1003 	dev.d_partition = priv->part;
1004 	dev.d_offset = priv->start;
1005 
1006 	off = off / BD(&dev).bd_sectorsize;
1007 	/* GELI gives us the offset relative to the partition start */
1008 	off += dev.d_offset;
1009 	bytes = bytes / BD(&dev).bd_sectorsize;
1010 
1011 	return (bd_io(&dev, off, bytes, buf, 0));
1012 }
1013 #endif /* LOADER_GELI_SUPPORT */
1014