xref: /freebsd/stand/i386/zfsboot/zfsboot.c (revision 8a0a413e)
1 /*-
2  * Copyright (c) 1998 Robert Nordier
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms are freely
6  * permitted provided that the above copyright notice and this
7  * paragraph and the following disclaimer are duplicated in all
8  * such forms.
9  *
10  * This software is provided "AS IS" and without any express or
11  * implied warranties, including, without limitation, the implied
12  * warranties of merchantability and fitness for a particular
13  * purpose.
14  */
15 
16 #include <sys/cdefs.h>
17 __FBSDID("$FreeBSD$");
18 
19 #include <sys/param.h>
20 #include <sys/errno.h>
21 #include <sys/diskmbr.h>
22 #ifdef GPT
23 #include <sys/gpt.h>
24 #endif
25 #include <sys/reboot.h>
26 #include <sys/queue.h>
27 
28 #include <machine/bootinfo.h>
29 #include <machine/elf.h>
30 #include <machine/pc/bios.h>
31 
32 #include <stdarg.h>
33 #include <stddef.h>
34 
35 #include <a.out.h>
36 
37 #include <btxv86.h>
38 
39 #include "lib.h"
40 #include "rbx.h"
41 #include "drv.h"
42 #include "edd.h"
43 #include "util.h"
44 #include "cons.h"
45 #include "bootargs.h"
46 #include "paths.h"
47 
48 #include "libzfs.h"
49 
50 #define ARGS			0x900
51 #define NOPT			14
52 #define NDEV			3
53 
54 #define BIOS_NUMDRIVES		0x475
55 #define DRV_HARD		0x80
56 #define DRV_MASK		0x7f
57 
58 #define TYPE_AD			0
59 #define TYPE_DA			1
60 #define TYPE_MAXHARD		TYPE_DA
61 #define TYPE_FD			2
62 
63 #define DEV_GELIBOOT_BSIZE	4096
64 
65 extern uint32_t _end;
66 
67 #ifdef GPT
68 static const uuid_t freebsd_zfs_uuid = GPT_ENT_TYPE_FREEBSD_ZFS;
69 #endif
70 static const char optstr[NOPT] = "DhaCcdgmnpqrsv"; /* Also 'P', 'S' */
71 static const unsigned char flags[NOPT] = {
72     RBX_DUAL,
73     RBX_SERIAL,
74     RBX_ASKNAME,
75     RBX_CDROM,
76     RBX_CONFIG,
77     RBX_KDB,
78     RBX_GDB,
79     RBX_MUTE,
80     RBX_NOINTR,
81     RBX_PAUSE,
82     RBX_QUIET,
83     RBX_DFLTROOT,
84     RBX_SINGLE,
85     RBX_VERBOSE
86 };
87 uint32_t opts;
88 
89 static const unsigned char dev_maj[NDEV] = {30, 4, 2};
90 
91 static char cmd[512];
92 static char cmddup[512];
93 static char kname[1024];
94 static char rootname[256];
95 static int comspeed = SIOSPD;
96 static struct bootinfo bootinfo;
97 static uint32_t bootdev;
98 static struct zfs_boot_args zfsargs;
99 static struct zfsmount zfsmount;
100 
101 vm_offset_t	high_heap_base;
102 uint32_t	bios_basemem, bios_extmem, high_heap_size;
103 
104 static struct bios_smap smap;
105 
106 /*
107  * The minimum amount of memory to reserve in bios_extmem for the heap.
108  */
109 #define	HEAP_MIN		(64 * 1024 * 1024)
110 
111 static char *heap_next;
112 static char *heap_end;
113 
114 /* Buffers that must not span a 64k boundary. */
115 #define READ_BUF_SIZE		8192
116 struct dmadat {
117 	char rdbuf[READ_BUF_SIZE];	/* for reading large things */
118 	char secbuf[READ_BUF_SIZE];	/* for MBR/disklabel */
119 };
120 static struct dmadat *dmadat;
121 
122 void exit(int);
123 void reboot(void);
124 static void load(void);
125 static int parse_cmd(void);
126 static void bios_getmem(void);
127 void *malloc(size_t n);
128 void free(void *ptr);
129 int main(void);
130 
131 void *
132 malloc(size_t n)
133 {
134 	char *p = heap_next;
135 	if (p + n > heap_end) {
136 		printf("malloc failure\n");
137 		for (;;)
138 		    ;
139 		/* NOTREACHED */
140 		return (0);
141 	}
142 	heap_next += n;
143 	return (p);
144 }
145 
146 void
147 free(void *ptr)
148 {
149 
150 	return;
151 }
152 
153 static char *
154 strdup(const char *s)
155 {
156 	char *p = malloc(strlen(s) + 1);
157 	strcpy(p, s);
158 	return (p);
159 }
160 
161 #ifdef LOADER_GELI_SUPPORT
162 #include "geliboot.c"
163 static char gelipw[GELI_PW_MAXLEN];
164 static struct keybuf *gelibuf;
165 #endif
166 
167 #include "zfsimpl.c"
168 
169 /*
170  * Read from a dnode (which must be from a ZPL filesystem).
171  */
172 static int
173 zfs_read(spa_t *spa, const dnode_phys_t *dnode, off_t *offp, void *start, size_t size)
174 {
175 	const znode_phys_t *zp = (const znode_phys_t *) dnode->dn_bonus;
176 	size_t n;
177 	int rc;
178 
179 	n = size;
180 	if (*offp + n > zp->zp_size)
181 		n = zp->zp_size - *offp;
182 
183 	rc = dnode_read(spa, dnode, *offp, start, n);
184 	if (rc)
185 		return (-1);
186 	*offp += n;
187 
188 	return (n);
189 }
190 
191 /*
192  * Current ZFS pool
193  */
194 static spa_t *spa;
195 static spa_t *primary_spa;
196 static vdev_t *primary_vdev;
197 
198 /*
199  * A wrapper for dskread that doesn't have to worry about whether the
200  * buffer pointer crosses a 64k boundary.
201  */
202 static int
203 vdev_read(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
204 {
205 	char *p;
206 	daddr_t lba, alignlba;
207 	off_t diff;
208 	unsigned int nb, alignnb;
209 	struct dsk *dsk = (struct dsk *) priv;
210 
211 	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
212 		return -1;
213 
214 	p = buf;
215 	lba = off / DEV_BSIZE;
216 	lba += dsk->start;
217 	/*
218 	 * Align reads to 4k else 4k sector GELIs will not decrypt.
219 	 * Round LBA down to nearest multiple of DEV_GELIBOOT_BSIZE bytes.
220 	 */
221 	alignlba = rounddown2(off, DEV_GELIBOOT_BSIZE) / DEV_BSIZE;
222 	/*
223 	 * The read must be aligned to DEV_GELIBOOT_BSIZE bytes relative to the
224 	 * start of the GELI partition, not the start of the actual disk.
225 	 */
226 	alignlba += dsk->start;
227 	diff = (lba - alignlba) * DEV_BSIZE;
228 
229 	while (bytes > 0) {
230 		nb = bytes / DEV_BSIZE;
231 		/*
232 		 * Ensure that the read size plus the leading offset does not
233 		 * exceed the size of the read buffer.
234 		 */
235 		if (nb > (READ_BUF_SIZE - diff) / DEV_BSIZE)
236 			nb = (READ_BUF_SIZE - diff) / DEV_BSIZE;
237 		/*
238 		 * Round the number of blocks to read up to the nearest multiple
239 		 * of DEV_GELIBOOT_BSIZE.
240 		 */
241 		alignnb = roundup2(nb * DEV_BSIZE + diff, DEV_GELIBOOT_BSIZE)
242 		    / DEV_BSIZE;
243 
244 		if (drvread(dsk, dmadat->rdbuf, alignlba, alignnb))
245 			return -1;
246 #ifdef LOADER_GELI_SUPPORT
247 		/* decrypt */
248 		if (is_geli(dsk) == 0) {
249 			if (geli_read(dsk, ((alignlba - dsk->start) *
250 			    DEV_BSIZE), dmadat->rdbuf, alignnb * DEV_BSIZE))
251 				return (-1);
252 		}
253 #endif
254 		memcpy(p, dmadat->rdbuf + diff, nb * DEV_BSIZE);
255 		p += nb * DEV_BSIZE;
256 		lba += nb;
257 		alignlba += alignnb;
258 		bytes -= nb * DEV_BSIZE;
259 		/* Don't need the leading offset after the first block. */
260 		diff = 0;
261 	}
262 
263 	return 0;
264 }
265 
266 static int
267 vdev_write(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
268 {
269 	char *p;
270 	daddr_t lba;
271 	unsigned int nb;
272 	struct dsk *dsk = (struct dsk *) priv;
273 
274 	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
275 		return -1;
276 
277 	p = buf;
278 	lba = off / DEV_BSIZE;
279 	lba += dsk->start;
280 	while (bytes > 0) {
281 		nb = bytes / DEV_BSIZE;
282 		if (nb > READ_BUF_SIZE / DEV_BSIZE)
283 			nb = READ_BUF_SIZE / DEV_BSIZE;
284 		memcpy(dmadat->rdbuf, p, nb * DEV_BSIZE);
285 		if (drvwrite(dsk, dmadat->rdbuf, lba, nb))
286 			return -1;
287 		p += nb * DEV_BSIZE;
288 		lba += nb;
289 		bytes -= nb * DEV_BSIZE;
290 	}
291 
292 	return 0;
293 }
294 
295 static int
296 xfsread(const dnode_phys_t *dnode, off_t *offp, void *buf, size_t nbyte)
297 {
298     if ((size_t)zfs_read(spa, dnode, offp, buf, nbyte) != nbyte) {
299 	printf("Invalid format\n");
300 	return -1;
301     }
302     return 0;
303 }
304 
305 /*
306  * Read Pad2 (formerly "Boot Block Header") area of the first
307  * vdev label of the given vdev.
308  */
309 static int
310 vdev_read_pad2(vdev_t *vdev, char *buf, size_t size)
311 {
312 	blkptr_t bp;
313 	char *tmp = zap_scratch;
314 	off_t off = offsetof(vdev_label_t, vl_pad2);
315 
316 	if (size > VDEV_PAD_SIZE)
317 		size = VDEV_PAD_SIZE;
318 
319 	BP_ZERO(&bp);
320 	BP_SET_LSIZE(&bp, VDEV_PAD_SIZE);
321 	BP_SET_PSIZE(&bp, VDEV_PAD_SIZE);
322 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
323 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
324 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
325 	if (vdev_read_phys(vdev, &bp, tmp, off, 0))
326 		return (EIO);
327 	memcpy(buf, tmp, size);
328 	return (0);
329 }
330 
331 static int
332 vdev_clear_pad2(vdev_t *vdev)
333 {
334 	char *zeroes = zap_scratch;
335 	uint64_t *end;
336 	off_t off = offsetof(vdev_label_t, vl_pad2);
337 
338 	memset(zeroes, 0, VDEV_PAD_SIZE);
339 	end = (uint64_t *)(zeroes + VDEV_PAD_SIZE);
340 	/* ZIO_CHECKSUM_LABEL magic and pre-calcualted checksum for all zeros */
341 	end[-5] = 0x0210da7ab10c7a11;
342 	end[-4] = 0x97f48f807f6e2a3f;
343 	end[-3] = 0xaf909f1658aacefc;
344 	end[-2] = 0xcbd1ea57ff6db48b;
345 	end[-1] = 0x6ec692db0d465fab;
346 	if (vdev_write(vdev, vdev->v_read_priv, off, zeroes, VDEV_PAD_SIZE))
347 		return (EIO);
348 	return (0);
349 }
350 
351 static void
352 bios_getmem(void)
353 {
354     uint64_t size;
355 
356     /* Parse system memory map */
357     v86.ebx = 0;
358     do {
359 	v86.ctl = V86_FLAGS;
360 	v86.addr = 0x15;		/* int 0x15 function 0xe820*/
361 	v86.eax = 0xe820;
362 	v86.ecx = sizeof(struct bios_smap);
363 	v86.edx = SMAP_SIG;
364 	v86.es = VTOPSEG(&smap);
365 	v86.edi = VTOPOFF(&smap);
366 	v86int();
367 	if (V86_CY(v86.efl) || (v86.eax != SMAP_SIG))
368 	    break;
369 	/* look for a low-memory segment that's large enough */
370 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0) &&
371 	    (smap.length >= (512 * 1024)))
372 	    bios_basemem = smap.length;
373 	/* look for the first segment in 'extended' memory */
374 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0x100000)) {
375 	    bios_extmem = smap.length;
376 	}
377 
378 	/*
379 	 * Look for the largest segment in 'extended' memory beyond
380 	 * 1MB but below 4GB.
381 	 */
382 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base > 0x100000) &&
383 	    (smap.base < 0x100000000ull)) {
384 	    size = smap.length;
385 
386 	    /*
387 	     * If this segment crosses the 4GB boundary, truncate it.
388 	     */
389 	    if (smap.base + size > 0x100000000ull)
390 		size = 0x100000000ull - smap.base;
391 
392 	    if (size > high_heap_size) {
393 		high_heap_size = size;
394 		high_heap_base = smap.base;
395 	    }
396 	}
397     } while (v86.ebx != 0);
398 
399     /* Fall back to the old compatibility function for base memory */
400     if (bios_basemem == 0) {
401 	v86.ctl = 0;
402 	v86.addr = 0x12;		/* int 0x12 */
403 	v86int();
404 
405 	bios_basemem = (v86.eax & 0xffff) * 1024;
406     }
407 
408     /* Fall back through several compatibility functions for extended memory */
409     if (bios_extmem == 0) {
410 	v86.ctl = V86_FLAGS;
411 	v86.addr = 0x15;		/* int 0x15 function 0xe801*/
412 	v86.eax = 0xe801;
413 	v86int();
414 	if (!V86_CY(v86.efl)) {
415 	    bios_extmem = ((v86.ecx & 0xffff) + ((v86.edx & 0xffff) * 64)) * 1024;
416 	}
417     }
418     if (bios_extmem == 0) {
419 	v86.ctl = 0;
420 	v86.addr = 0x15;		/* int 0x15 function 0x88*/
421 	v86.eax = 0x8800;
422 	v86int();
423 	bios_extmem = (v86.eax & 0xffff) * 1024;
424     }
425 
426     /*
427      * If we have extended memory and did not find a suitable heap
428      * region in the SMAP, use the last 3MB of 'extended' memory as a
429      * high heap candidate.
430      */
431     if (bios_extmem >= HEAP_MIN && high_heap_size < HEAP_MIN) {
432 	high_heap_size = HEAP_MIN;
433 	high_heap_base = bios_extmem + 0x100000 - HEAP_MIN;
434     }
435 }
436 
437 /*
438  * Try to detect a device supported by the legacy int13 BIOS
439  */
440 static int
441 int13probe(int drive)
442 {
443     v86.ctl = V86_FLAGS;
444     v86.addr = 0x13;
445     v86.eax = 0x800;
446     v86.edx = drive;
447     v86int();
448 
449     if (!V86_CY(v86.efl) &&				/* carry clear */
450 	((v86.edx & 0xff) != (drive & DRV_MASK))) {	/* unit # OK */
451 	if ((v86.ecx & 0x3f) == 0) {			/* absurd sector size */
452 		return(0);				/* skip device */
453 	}
454 	return (1);
455     }
456     return(0);
457 }
458 
459 /*
460  * We call this when we find a ZFS vdev - ZFS consumes the dsk
461  * structure so we must make a new one.
462  */
463 static struct dsk *
464 copy_dsk(struct dsk *dsk)
465 {
466     struct dsk *newdsk;
467 
468     newdsk = malloc(sizeof(struct dsk));
469     *newdsk = *dsk;
470     return (newdsk);
471 }
472 
473 /*
474  * Get disk size from eax=0x800 and 0x4800. We need to probe both
475  * because 0x4800 may not be available and we would like to get more
476  * or less correct disk size - if it is possible at all.
477  * Note we do not really want to touch drv.c because that code is shared
478  * with boot2 and we can not afford to grow that code.
479  */
480 static uint64_t
481 drvsize_ext(struct dsk *dskp)
482 {
483 	uint64_t size, tmp;
484 	int cyl, hds, sec;
485 
486 	v86.ctl = V86_FLAGS;
487 	v86.addr = 0x13;
488 	v86.eax = 0x800;
489 	v86.edx = dskp->drive;
490 	v86int();
491 
492 	/* Don't error out if we get bad sector number, try EDD as well */
493 	if (V86_CY(v86.efl) ||	/* carry set */
494 	    (v86.edx & 0xff) <= (unsigned)(dskp->drive & 0x7f)) /* unit # bad */
495 		return (0);
496 
497 	cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
498 	/* Convert max head # -> # of heads */
499 	hds = ((v86.edx & 0xff00) >> 8) + 1;
500 	sec = v86.ecx & 0x3f;
501 
502 	size = (uint64_t)cyl * hds * sec;
503 
504 	/* Determine if we can use EDD with this device. */
505 	v86.ctl = V86_FLAGS;
506 	v86.addr = 0x13;
507 	v86.eax = 0x4100;
508 	v86.edx = dskp->drive;
509 	v86.ebx = 0x55aa;
510 	v86int();
511 	if (V86_CY(v86.efl) ||  /* carry set */
512 	    (v86.ebx & 0xffff) != 0xaa55 || /* signature */
513 	    (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
514 		return (size);
515 
516 	tmp = drvsize(dskp);
517 	if (tmp > size)
518 		size = tmp;
519 
520 	return (size);
521 }
522 
523 /*
524  * The "layered" ioctl to read disk/partition size. Unfortunately
525  * the zfsboot case is hardest, because we do not have full software
526  * stack available, so we need to do some manual work here.
527  */
528 uint64_t
529 ldi_get_size(void *priv)
530 {
531 	struct dsk *dskp = priv;
532 	uint64_t size = dskp->size;
533 
534 	if (dskp->start == 0)
535 		size = drvsize_ext(dskp);
536 
537 	return (size * DEV_BSIZE);
538 }
539 
540 static void
541 probe_drive(struct dsk *dsk)
542 {
543 #ifdef GPT
544     struct gpt_hdr hdr;
545     struct gpt_ent *ent;
546     unsigned part, entries_per_sec;
547     daddr_t slba;
548 #endif
549 #if defined(GPT) || defined(LOADER_GELI_SUPPORT)
550     daddr_t elba;
551 #endif
552 
553     struct dos_partition *dp;
554     char *sec;
555     unsigned i;
556 
557     /*
558      * If we find a vdev on the whole disk, stop here.
559      */
560     if (vdev_probe(vdev_read, dsk, NULL) == 0)
561 	return;
562 
563 #ifdef LOADER_GELI_SUPPORT
564     /*
565      * Taste the disk, if it is GELI encrypted, decrypt it and check to see if
566      * it is a usable vdev then. Otherwise dig
567      * out the partition table and probe each slice/partition
568      * in turn for a vdev or GELI encrypted vdev.
569      */
570     elba = drvsize_ext(dsk);
571     if (elba > 0) {
572 	elba--;
573     }
574     if (geli_taste(vdev_read, dsk, elba) == 0) {
575 	if (geli_havekey(dsk) == 0 || geli_passphrase(&gelipw, dsk->unit,
576 	  ':', 0, dsk) == 0) {
577 	    if (vdev_probe(vdev_read, dsk, NULL) == 0) {
578 		return;
579 	    }
580 	}
581     }
582 #endif /* LOADER_GELI_SUPPORT */
583 
584     sec = dmadat->secbuf;
585     dsk->start = 0;
586 
587 #ifdef GPT
588     /*
589      * First check for GPT.
590      */
591     if (drvread(dsk, sec, 1, 1)) {
592 	return;
593     }
594     memcpy(&hdr, sec, sizeof(hdr));
595     if (memcmp(hdr.hdr_sig, GPT_HDR_SIG, sizeof(hdr.hdr_sig)) != 0 ||
596 	hdr.hdr_lba_self != 1 || hdr.hdr_revision < 0x00010000 ||
597 	hdr.hdr_entsz < sizeof(*ent) || DEV_BSIZE % hdr.hdr_entsz != 0) {
598 	goto trymbr;
599     }
600 
601     /*
602      * Probe all GPT partitions for the presence of ZFS pools. We
603      * return the spa_t for the first we find (if requested). This
604      * will have the effect of booting from the first pool on the
605      * disk.
606      *
607      * If no vdev is found, GELI decrypting the device and try again
608      */
609     entries_per_sec = DEV_BSIZE / hdr.hdr_entsz;
610     slba = hdr.hdr_lba_table;
611     elba = slba + hdr.hdr_entries / entries_per_sec;
612     while (slba < elba) {
613 	dsk->start = 0;
614 	if (drvread(dsk, sec, slba, 1))
615 	    return;
616 	for (part = 0; part < entries_per_sec; part++) {
617 	    ent = (struct gpt_ent *)(sec + part * hdr.hdr_entsz);
618 	    if (memcmp(&ent->ent_type, &freebsd_zfs_uuid,
619 		     sizeof(uuid_t)) == 0) {
620 		dsk->start = ent->ent_lba_start;
621 		dsk->size = ent->ent_lba_end - ent->ent_lba_start + 1;
622 		dsk->slice = part + 1;
623 		dsk->part = 255;
624 		if (vdev_probe(vdev_read, dsk, NULL) == 0) {
625 		    /*
626 		     * This slice had a vdev. We need a new dsk
627 		     * structure now since the vdev now owns this one.
628 		     */
629 		    dsk = copy_dsk(dsk);
630 		}
631 #ifdef LOADER_GELI_SUPPORT
632 		else if (geli_taste(vdev_read, dsk, ent->ent_lba_end -
633 			 ent->ent_lba_start) == 0) {
634 		    if (geli_havekey(dsk) == 0 || geli_passphrase(&gelipw,
635 		      dsk->unit, 'p', dsk->slice, dsk) == 0) {
636 			/*
637 			 * This slice has GELI, check it for ZFS.
638 			 */
639 			if (vdev_probe(vdev_read, dsk, NULL) == 0) {
640 			    /*
641 			     * This slice had a vdev. We need a new dsk
642 			     * structure now since the vdev now owns this one.
643 			     */
644 			    dsk = copy_dsk(dsk);
645 			}
646 			break;
647 		    }
648 		}
649 #endif /* LOADER_GELI_SUPPORT */
650 	    }
651 	}
652 	slba++;
653     }
654     return;
655 trymbr:
656 #endif /* GPT */
657 
658     if (drvread(dsk, sec, DOSBBSECTOR, 1))
659 	return;
660     dp = (void *)(sec + DOSPARTOFF);
661 
662     for (i = 0; i < NDOSPART; i++) {
663 	if (!dp[i].dp_typ)
664 	    continue;
665 	dsk->start = dp[i].dp_start;
666 	dsk->size = dp[i].dp_size;
667 	dsk->slice = i + 1;
668 	if (vdev_probe(vdev_read, dsk, NULL) == 0) {
669 	    dsk = copy_dsk(dsk);
670 	}
671 #ifdef LOADER_GELI_SUPPORT
672 	else if (geli_taste(vdev_read, dsk, dp[i].dp_size -
673 		 dp[i].dp_start) == 0) {
674 	    if (geli_havekey(dsk) == 0 || geli_passphrase(&gelipw, dsk->unit,
675 	      's', i, dsk) == 0) {
676 		/*
677 		 * This slice has GELI, check it for ZFS.
678 		 */
679 		if (vdev_probe(vdev_read, dsk, NULL) == 0) {
680 		    /*
681 		     * This slice had a vdev. We need a new dsk
682 		     * structure now since the vdev now owns this one.
683 		     */
684 		    dsk = copy_dsk(dsk);
685 		}
686 		break;
687 	    }
688 	}
689 #endif /* LOADER_GELI_SUPPORT */
690     }
691 }
692 
693 int
694 main(void)
695 {
696     dnode_phys_t dn;
697     off_t off;
698     struct dsk *dsk;
699     int autoboot, i;
700     int nextboot;
701     int rc;
702 
703     dmadat = (void *)(roundup2(__base + (int32_t)&_end, 0x10000) - __base);
704 
705     bios_getmem();
706 
707     if (high_heap_size > 0) {
708 	heap_end = PTOV(high_heap_base + high_heap_size);
709 	heap_next = PTOV(high_heap_base);
710     } else {
711 	heap_next = (char *)dmadat + sizeof(*dmadat);
712 	heap_end = (char *)PTOV(bios_basemem);
713     }
714 
715     dsk = malloc(sizeof(struct dsk));
716     dsk->drive = *(uint8_t *)PTOV(ARGS);
717     dsk->type = dsk->drive & DRV_HARD ? TYPE_AD : TYPE_FD;
718     dsk->unit = dsk->drive & DRV_MASK;
719     dsk->slice = *(uint8_t *)PTOV(ARGS + 1) + 1;
720     dsk->part = 0;
721     dsk->start = 0;
722     dsk->size = 0;
723 
724     bootinfo.bi_version = BOOTINFO_VERSION;
725     bootinfo.bi_size = sizeof(bootinfo);
726     bootinfo.bi_basemem = bios_basemem / 1024;
727     bootinfo.bi_extmem = bios_extmem / 1024;
728     bootinfo.bi_memsizes_valid++;
729     bootinfo.bi_bios_dev = dsk->drive;
730 
731     bootdev = MAKEBOOTDEV(dev_maj[dsk->type],
732 			  dsk->slice, dsk->unit, dsk->part);
733 
734     /* Process configuration file */
735 
736     autoboot = 1;
737 
738 #ifdef LOADER_GELI_SUPPORT
739     geli_init();
740 #endif
741     zfs_init();
742 
743     /*
744      * Probe the boot drive first - we will try to boot from whatever
745      * pool we find on that drive.
746      */
747     probe_drive(dsk);
748 
749     /*
750      * Probe the rest of the drives that the bios knows about. This
751      * will find any other available pools and it may fill in missing
752      * vdevs for the boot pool.
753      */
754 #ifndef VIRTUALBOX
755     for (i = 0; i < *(unsigned char *)PTOV(BIOS_NUMDRIVES); i++)
756 #else
757     for (i = 0; i < MAXBDDEV; i++)
758 #endif
759     {
760 	if ((i | DRV_HARD) == *(uint8_t *)PTOV(ARGS))
761 	    continue;
762 
763 	if (!int13probe(i | DRV_HARD))
764 	    break;
765 
766 	dsk = malloc(sizeof(struct dsk));
767 	dsk->drive = i | DRV_HARD;
768 	dsk->type = dsk->drive & TYPE_AD;
769 	dsk->unit = i;
770 	dsk->slice = 0;
771 	dsk->part = 0;
772 	dsk->start = 0;
773 	dsk->size = 0;
774 	probe_drive(dsk);
775     }
776 
777     /*
778      * The first discovered pool, if any, is the pool.
779      */
780     spa = spa_get_primary();
781     if (!spa) {
782 	printf("%s: No ZFS pools located, can't boot\n", BOOTPROG);
783 	for (;;)
784 	    ;
785     }
786 
787     primary_spa = spa;
788     primary_vdev = spa_get_primary_vdev(spa);
789 
790     nextboot = 0;
791     rc  = vdev_read_pad2(primary_vdev, cmd, sizeof(cmd));
792     if (vdev_clear_pad2(primary_vdev))
793 	printf("failed to clear pad2 area of primary vdev\n");
794     if (rc == 0) {
795 	if (*cmd) {
796 	    /*
797 	     * We could find an old-style ZFS Boot Block header here.
798 	     * Simply ignore it.
799 	     */
800 	    if (*(uint64_t *)cmd != 0x2f5b007b10c) {
801 		/*
802 		 * Note that parse() is destructive to cmd[] and we also want
803 		 * to honor RBX_QUIET option that could be present in cmd[].
804 		 */
805 		nextboot = 1;
806 		memcpy(cmddup, cmd, sizeof(cmd));
807 		if (parse_cmd()) {
808 		    printf("failed to parse pad2 area of primary vdev\n");
809 		    reboot();
810 		}
811 		if (!OPT_CHECK(RBX_QUIET))
812 		    printf("zfs nextboot: %s\n", cmddup);
813 	    }
814 	    /* Do not process this command twice */
815 	    *cmd = 0;
816 	}
817     } else
818 	printf("failed to read pad2 area of primary vdev\n");
819 
820     /* Mount ZFS only if it's not already mounted via nextboot parsing. */
821     if (zfsmount.spa == NULL &&
822 	(zfs_spa_init(spa) != 0 || zfs_mount(spa, 0, &zfsmount) != 0)) {
823 	printf("%s: failed to mount default pool %s\n",
824 	    BOOTPROG, spa->spa_name);
825 	autoboot = 0;
826     } else if (zfs_lookup(&zfsmount, PATH_CONFIG, &dn) == 0 ||
827         zfs_lookup(&zfsmount, PATH_DOTCONFIG, &dn) == 0) {
828 	off = 0;
829 	zfs_read(spa, &dn, &off, cmd, sizeof(cmd));
830     }
831 
832     if (*cmd) {
833 	/*
834 	 * Note that parse_cmd() is destructive to cmd[] and we also want
835 	 * to honor RBX_QUIET option that could be present in cmd[].
836 	 */
837 	memcpy(cmddup, cmd, sizeof(cmd));
838 	if (parse_cmd())
839 	    autoboot = 0;
840 	if (!OPT_CHECK(RBX_QUIET))
841 	    printf("%s: %s\n", PATH_CONFIG, cmddup);
842 	/* Do not process this command twice */
843 	*cmd = 0;
844     }
845 
846     /* Do not risk waiting at the prompt forever. */
847     if (nextboot && !autoboot)
848 	reboot();
849 
850     /*
851      * Try to exec /boot/loader. If interrupted by a keypress,
852      * or in case of failure, try to load a kernel directly instead.
853      */
854 
855     if (autoboot && !*kname) {
856 	memcpy(kname, PATH_LOADER_ZFS, sizeof(PATH_LOADER_ZFS));
857 	if (!keyhit(3)) {
858 	    load();
859 	    memcpy(kname, PATH_KERNEL, sizeof(PATH_KERNEL));
860 	}
861     }
862 
863     /* Present the user with the boot2 prompt. */
864 
865     for (;;) {
866 	if (!autoboot || !OPT_CHECK(RBX_QUIET)) {
867 	    printf("\nFreeBSD/x86 boot\n");
868 	    if (zfs_rlookup(spa, zfsmount.rootobj, rootname) != 0)
869 		printf("Default: %s/<0x%llx>:%s\n"
870 		       "boot: ",
871 		       spa->spa_name, zfsmount.rootobj, kname);
872 	    else if (rootname[0] != '\0')
873 		printf("Default: %s/%s:%s\n"
874 		       "boot: ",
875 		       spa->spa_name, rootname, kname);
876 	    else
877 		printf("Default: %s:%s\n"
878 		       "boot: ",
879 		       spa->spa_name, kname);
880 	}
881 	if (ioctrl & IO_SERIAL)
882 	    sio_flush();
883 	if (!autoboot || keyhit(5))
884 	    getstr(cmd, sizeof(cmd));
885 	else if (!autoboot || !OPT_CHECK(RBX_QUIET))
886 	    putchar('\n');
887 	autoboot = 0;
888 	if (parse_cmd())
889 	    putchar('\a');
890 	else
891 	    load();
892     }
893 }
894 
895 /* XXX - Needed for btxld to link the boot2 binary; do not remove. */
896 void
897 exit(int x)
898 {
899     __exit(x);
900 }
901 
902 void
903 reboot(void)
904 {
905     __exit(0);
906 }
907 
908 static void
909 load(void)
910 {
911     union {
912 	struct exec ex;
913 	Elf32_Ehdr eh;
914     } hdr;
915     static Elf32_Phdr ep[2];
916     static Elf32_Shdr es[2];
917     caddr_t p;
918     dnode_phys_t dn;
919     off_t off;
920     uint32_t addr, x;
921     int fmt, i, j;
922 
923     if (zfs_lookup(&zfsmount, kname, &dn)) {
924 	printf("\nCan't find %s\n", kname);
925 	return;
926     }
927     off = 0;
928     if (xfsread(&dn, &off, &hdr, sizeof(hdr)))
929 	return;
930     if (N_GETMAGIC(hdr.ex) == ZMAGIC)
931 	fmt = 0;
932     else if (IS_ELF(hdr.eh))
933 	fmt = 1;
934     else {
935 	printf("Invalid %s\n", "format");
936 	return;
937     }
938     if (fmt == 0) {
939 	addr = hdr.ex.a_entry & 0xffffff;
940 	p = PTOV(addr);
941 	off = PAGE_SIZE;
942 	if (xfsread(&dn, &off, p, hdr.ex.a_text))
943 	    return;
944 	p += roundup2(hdr.ex.a_text, PAGE_SIZE);
945 	if (xfsread(&dn, &off, p, hdr.ex.a_data))
946 	    return;
947 	p += hdr.ex.a_data + roundup2(hdr.ex.a_bss, PAGE_SIZE);
948 	bootinfo.bi_symtab = VTOP(p);
949 	memcpy(p, &hdr.ex.a_syms, sizeof(hdr.ex.a_syms));
950 	p += sizeof(hdr.ex.a_syms);
951 	if (hdr.ex.a_syms) {
952 	    if (xfsread(&dn, &off, p, hdr.ex.a_syms))
953 		return;
954 	    p += hdr.ex.a_syms;
955 	    if (xfsread(&dn, &off, p, sizeof(int)))
956 		return;
957 	    x = *(uint32_t *)p;
958 	    p += sizeof(int);
959 	    x -= sizeof(int);
960 	    if (xfsread(&dn, &off, p, x))
961 		return;
962 	    p += x;
963 	}
964     } else {
965 	off = hdr.eh.e_phoff;
966 	for (j = i = 0; i < hdr.eh.e_phnum && j < 2; i++) {
967 	    if (xfsread(&dn, &off, ep + j, sizeof(ep[0])))
968 		return;
969 	    if (ep[j].p_type == PT_LOAD)
970 		j++;
971 	}
972 	for (i = 0; i < 2; i++) {
973 	    p = PTOV(ep[i].p_paddr & 0xffffff);
974 	    off = ep[i].p_offset;
975 	    if (xfsread(&dn, &off, p, ep[i].p_filesz))
976 		return;
977 	}
978 	p += roundup2(ep[1].p_memsz, PAGE_SIZE);
979 	bootinfo.bi_symtab = VTOP(p);
980 	if (hdr.eh.e_shnum == hdr.eh.e_shstrndx + 3) {
981 	    off = hdr.eh.e_shoff + sizeof(es[0]) *
982 		(hdr.eh.e_shstrndx + 1);
983 	    if (xfsread(&dn, &off, &es, sizeof(es)))
984 		return;
985 	    for (i = 0; i < 2; i++) {
986 		memcpy(p, &es[i].sh_size, sizeof(es[i].sh_size));
987 		p += sizeof(es[i].sh_size);
988 		off = es[i].sh_offset;
989 		if (xfsread(&dn, &off, p, es[i].sh_size))
990 		    return;
991 		p += es[i].sh_size;
992 	    }
993 	}
994 	addr = hdr.eh.e_entry & 0xffffff;
995     }
996     bootinfo.bi_esymtab = VTOP(p);
997     bootinfo.bi_kernelname = VTOP(kname);
998     zfsargs.size = sizeof(zfsargs);
999     zfsargs.pool = zfsmount.spa->spa_guid;
1000     zfsargs.root = zfsmount.rootobj;
1001     zfsargs.primary_pool = primary_spa->spa_guid;
1002 #ifdef LOADER_GELI_SUPPORT
1003     explicit_bzero(gelipw, sizeof(gelipw));
1004     gelibuf = malloc(sizeof(struct keybuf) + (GELI_MAX_KEYS * sizeof(struct keybuf_ent)));
1005     geli_fill_keybuf(gelibuf);
1006     zfsargs.notapw = '\0';
1007     zfsargs.keybuf_sentinel = KEYBUF_SENTINEL;
1008     zfsargs.keybuf = gelibuf;
1009 #else
1010     zfsargs.gelipw[0] = '\0';
1011 #endif
1012     if (primary_vdev != NULL)
1013 	zfsargs.primary_vdev = primary_vdev->v_guid;
1014     else
1015 	printf("failed to detect primary vdev\n");
1016     __exec((caddr_t)addr, RB_BOOTINFO | (opts & RBX_MASK),
1017 	   bootdev,
1018 	   KARGS_FLAGS_ZFS | KARGS_FLAGS_EXTARG,
1019 	   (uint32_t) spa->spa_guid,
1020 	   (uint32_t) (spa->spa_guid >> 32),
1021 	   VTOP(&bootinfo),
1022 	   zfsargs);
1023 }
1024 
1025 static int
1026 zfs_mount_ds(char *dsname)
1027 {
1028     uint64_t newroot;
1029     spa_t *newspa;
1030     char *q;
1031 
1032     q = strchr(dsname, '/');
1033     if (q)
1034 	*q++ = '\0';
1035     newspa = spa_find_by_name(dsname);
1036     if (newspa == NULL) {
1037 	printf("\nCan't find ZFS pool %s\n", dsname);
1038 	return -1;
1039     }
1040 
1041     if (zfs_spa_init(newspa))
1042 	return -1;
1043 
1044     newroot = 0;
1045     if (q) {
1046 	if (zfs_lookup_dataset(newspa, q, &newroot)) {
1047 	    printf("\nCan't find dataset %s in ZFS pool %s\n",
1048 		    q, newspa->spa_name);
1049 	    return -1;
1050 	}
1051     }
1052     if (zfs_mount(newspa, newroot, &zfsmount)) {
1053 	printf("\nCan't mount ZFS dataset\n");
1054 	return -1;
1055     }
1056     spa = newspa;
1057     return (0);
1058 }
1059 
1060 static int
1061 parse_cmd(void)
1062 {
1063     char *arg = cmd;
1064     char *ep, *p, *q;
1065     const char *cp;
1066     int c, i, j;
1067 
1068     while ((c = *arg++)) {
1069 	if (c == ' ' || c == '\t' || c == '\n')
1070 	    continue;
1071 	for (p = arg; *p && *p != '\n' && *p != ' ' && *p != '\t'; p++);
1072 	ep = p;
1073 	if (*p)
1074 	    *p++ = 0;
1075 	if (c == '-') {
1076 	    while ((c = *arg++)) {
1077 		if (c == 'P') {
1078 		    if (*(uint8_t *)PTOV(0x496) & 0x10) {
1079 			cp = "yes";
1080 		    } else {
1081 			opts |= OPT_SET(RBX_DUAL) | OPT_SET(RBX_SERIAL);
1082 			cp = "no";
1083 		    }
1084 		    printf("Keyboard: %s\n", cp);
1085 		    continue;
1086 		} else if (c == 'S') {
1087 		    j = 0;
1088 		    while ((unsigned int)(i = *arg++ - '0') <= 9)
1089 			j = j * 10 + i;
1090 		    if (j > 0 && i == -'0') {
1091 			comspeed = j;
1092 			break;
1093 		    }
1094 		    /* Fall through to error below ('S' not in optstr[]). */
1095 		}
1096 		for (i = 0; c != optstr[i]; i++)
1097 		    if (i == NOPT - 1)
1098 			return -1;
1099 		opts ^= OPT_SET(flags[i]);
1100 	    }
1101 	    ioctrl = OPT_CHECK(RBX_DUAL) ? (IO_SERIAL|IO_KEYBOARD) :
1102 		     OPT_CHECK(RBX_SERIAL) ? IO_SERIAL : IO_KEYBOARD;
1103 	    if (ioctrl & IO_SERIAL) {
1104 	        if (sio_init(115200 / comspeed) != 0)
1105 		    ioctrl &= ~IO_SERIAL;
1106 	    }
1107 	} if (c == '?') {
1108 	    dnode_phys_t dn;
1109 
1110 	    if (zfs_lookup(&zfsmount, arg, &dn) == 0) {
1111 		zap_list(spa, &dn);
1112 	    }
1113 	    return -1;
1114 	} else {
1115 	    arg--;
1116 
1117 	    /*
1118 	     * Report pool status if the comment is 'status'. Lets
1119 	     * hope no-one wants to load /status as a kernel.
1120 	     */
1121 	    if (!strcmp(arg, "status")) {
1122 		spa_all_status();
1123 		return -1;
1124 	    }
1125 
1126 	    /*
1127 	     * If there is "zfs:" prefix simply ignore it.
1128 	     */
1129 	    if (strncmp(arg, "zfs:", 4) == 0)
1130 		arg += 4;
1131 
1132 	    /*
1133 	     * If there is a colon, switch pools.
1134 	     */
1135 	    q = strchr(arg, ':');
1136 	    if (q) {
1137 		*q++ = '\0';
1138 		if (zfs_mount_ds(arg) != 0)
1139 		    return -1;
1140 		arg = q;
1141 	    }
1142 	    if ((i = ep - arg)) {
1143 		if ((size_t)i >= sizeof(kname))
1144 		    return -1;
1145 		memcpy(kname, arg, i + 1);
1146 	    }
1147 	}
1148 	arg = p;
1149     }
1150     return 0;
1151 }
1152