xref: /freebsd/stand/i386/zfsboot/zfsboot.c (revision 780fb4a2fa9a9aee5ac48a60b790f567c0dc13e9)
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 "stand.h"
20 
21 #include <sys/param.h>
22 #include <sys/errno.h>
23 #include <sys/diskmbr.h>
24 #ifdef GPT
25 #include <sys/gpt.h>
26 #endif
27 #include <sys/reboot.h>
28 #include <sys/queue.h>
29 
30 #include <machine/bootinfo.h>
31 #include <machine/elf.h>
32 #include <machine/pc/bios.h>
33 
34 #include <stdarg.h>
35 #include <stddef.h>
36 
37 #include <a.out.h>
38 
39 #include <btxv86.h>
40 
41 #include "lib.h"
42 #include "rbx.h"
43 #include "drv.h"
44 #include "edd.h"
45 #include "cons.h"
46 #include "bootargs.h"
47 #include "paths.h"
48 
49 #include "libzfs.h"
50 
51 #define ARGS			0x900
52 #define NOPT			14
53 #define NDEV			3
54 
55 #define BIOS_NUMDRIVES		0x475
56 #define DRV_HARD		0x80
57 #define DRV_MASK		0x7f
58 
59 #define TYPE_AD			0
60 #define TYPE_DA			1
61 #define TYPE_MAXHARD		TYPE_DA
62 #define TYPE_FD			2
63 
64 #define DEV_GELIBOOT_BSIZE	4096
65 
66 extern uint32_t _end;
67 
68 #ifdef GPT
69 static const uuid_t freebsd_zfs_uuid = GPT_ENT_TYPE_FREEBSD_ZFS;
70 #endif
71 static const char optstr[NOPT] = "DhaCcdgmnpqrsv"; /* Also 'P', 'S' */
72 static const unsigned char flags[NOPT] = {
73     RBX_DUAL,
74     RBX_SERIAL,
75     RBX_ASKNAME,
76     RBX_CDROM,
77     RBX_CONFIG,
78     RBX_KDB,
79     RBX_GDB,
80     RBX_MUTE,
81     RBX_NOINTR,
82     RBX_PAUSE,
83     RBX_QUIET,
84     RBX_DFLTROOT,
85     RBX_SINGLE,
86     RBX_VERBOSE
87 };
88 uint32_t opts;
89 
90 static const unsigned char dev_maj[NDEV] = {30, 4, 2};
91 
92 static char cmd[512];
93 static char cmddup[512];
94 static char kname[1024];
95 static char rootname[256];
96 static int comspeed = SIOSPD;
97 static struct bootinfo bootinfo;
98 static uint32_t bootdev;
99 static struct zfs_boot_args zfsargs;
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 int main(void);
128 
129 #ifdef LOADER_GELI_SUPPORT
130 #include "geliboot.c"
131 static char gelipw[GELI_PW_MAXLEN];
132 static struct keybuf *gelibuf;
133 #endif
134 
135 #include "zfsimpl.c"
136 
137 /*
138  * Read from a dnode (which must be from a ZPL filesystem).
139  */
140 static int
141 zfs_read(spa_t *spa, const dnode_phys_t *dnode, off_t *offp, void *start, size_t size)
142 {
143 	const znode_phys_t *zp = (const znode_phys_t *) dnode->dn_bonus;
144 	size_t n;
145 	int rc;
146 
147 	n = size;
148 	if (*offp + n > zp->zp_size)
149 		n = zp->zp_size - *offp;
150 
151 	rc = dnode_read(spa, dnode, *offp, start, n);
152 	if (rc)
153 		return (-1);
154 	*offp += n;
155 
156 	return (n);
157 }
158 
159 /*
160  * Current ZFS pool
161  */
162 static spa_t *spa;
163 static spa_t *primary_spa;
164 static vdev_t *primary_vdev;
165 
166 /*
167  * A wrapper for dskread that doesn't have to worry about whether the
168  * buffer pointer crosses a 64k boundary.
169  */
170 static int
171 vdev_read(void *xvdev, void *priv, off_t off, void *buf, size_t bytes)
172 {
173 	char *p;
174 	daddr_t lba, alignlba;
175 	off_t diff;
176 	unsigned int nb, alignnb;
177 	struct dsk *dsk = (struct dsk *) priv;
178 
179 	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
180 		return -1;
181 
182 	p = buf;
183 	lba = off / DEV_BSIZE;
184 	lba += dsk->start;
185 	/*
186 	 * Align reads to 4k else 4k sector GELIs will not decrypt.
187 	 * Round LBA down to nearest multiple of DEV_GELIBOOT_BSIZE bytes.
188 	 */
189 	alignlba = rounddown2(off, DEV_GELIBOOT_BSIZE) / DEV_BSIZE;
190 	/*
191 	 * The read must be aligned to DEV_GELIBOOT_BSIZE bytes relative to the
192 	 * start of the GELI partition, not the start of the actual disk.
193 	 */
194 	alignlba += dsk->start;
195 	diff = (lba - alignlba) * DEV_BSIZE;
196 
197 	while (bytes > 0) {
198 		nb = bytes / DEV_BSIZE;
199 		/*
200 		 * Ensure that the read size plus the leading offset does not
201 		 * exceed the size of the read buffer.
202 		 */
203 		if (nb > (READ_BUF_SIZE - diff) / DEV_BSIZE)
204 			nb = (READ_BUF_SIZE - diff) / DEV_BSIZE;
205 		/*
206 		 * Round the number of blocks to read up to the nearest multiple
207 		 * of DEV_GELIBOOT_BSIZE.
208 		 */
209 		alignnb = roundup2(nb * DEV_BSIZE + diff, DEV_GELIBOOT_BSIZE)
210 		    / DEV_BSIZE;
211 
212 		if (dsk->size > 0 && alignlba + alignnb > dsk->size + dsk->start) {
213 			printf("Shortening read at %lld from %d to %lld\n", alignlba,
214 			    alignnb, (dsk->size + dsk->start) - alignlba);
215 			alignnb = (dsk->size + dsk->start) - alignlba;
216 		}
217 
218 		if (drvread(dsk, dmadat->rdbuf, alignlba, alignnb))
219 			return -1;
220 #ifdef LOADER_GELI_SUPPORT
221 		/* decrypt */
222 		if (is_geli(dsk) == 0) {
223 			if (geli_read(dsk, ((alignlba - dsk->start) *
224 			    DEV_BSIZE), dmadat->rdbuf, alignnb * DEV_BSIZE))
225 				return (-1);
226 		}
227 #endif
228 		memcpy(p, dmadat->rdbuf + diff, nb * DEV_BSIZE);
229 		p += nb * DEV_BSIZE;
230 		lba += nb;
231 		alignlba += alignnb;
232 		bytes -= nb * DEV_BSIZE;
233 		/* Don't need the leading offset after the first block. */
234 		diff = 0;
235 	}
236 
237 	return 0;
238 }
239 /* Match the signature exactly due to signature madness */
240 static int
241 vdev_read2(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
242 {
243 	return vdev_read(vdev, priv, off, buf, bytes);
244 }
245 
246 
247 static int
248 vdev_write(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
249 {
250 	char *p;
251 	daddr_t lba;
252 	unsigned int nb;
253 	struct dsk *dsk = (struct dsk *) priv;
254 
255 	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
256 		return -1;
257 
258 	p = buf;
259 	lba = off / DEV_BSIZE;
260 	lba += dsk->start;
261 	while (bytes > 0) {
262 		nb = bytes / DEV_BSIZE;
263 		if (nb > READ_BUF_SIZE / DEV_BSIZE)
264 			nb = READ_BUF_SIZE / DEV_BSIZE;
265 		memcpy(dmadat->rdbuf, p, nb * DEV_BSIZE);
266 		if (drvwrite(dsk, dmadat->rdbuf, lba, nb))
267 			return -1;
268 		p += nb * DEV_BSIZE;
269 		lba += nb;
270 		bytes -= nb * DEV_BSIZE;
271 	}
272 
273 	return 0;
274 }
275 
276 static int
277 xfsread(const dnode_phys_t *dnode, off_t *offp, void *buf, size_t nbyte)
278 {
279     if ((size_t)zfs_read(spa, dnode, offp, buf, nbyte) != nbyte) {
280 	printf("Invalid format\n");
281 	return -1;
282     }
283     return 0;
284 }
285 
286 /*
287  * Read Pad2 (formerly "Boot Block Header") area of the first
288  * vdev label of the given vdev.
289  */
290 static int
291 vdev_read_pad2(vdev_t *vdev, char *buf, size_t size)
292 {
293 	blkptr_t bp;
294 	char *tmp = zap_scratch;
295 	off_t off = offsetof(vdev_label_t, vl_pad2);
296 
297 	if (size > VDEV_PAD_SIZE)
298 		size = VDEV_PAD_SIZE;
299 
300 	BP_ZERO(&bp);
301 	BP_SET_LSIZE(&bp, VDEV_PAD_SIZE);
302 	BP_SET_PSIZE(&bp, VDEV_PAD_SIZE);
303 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
304 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
305 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
306 	if (vdev_read_phys(vdev, &bp, tmp, off, 0))
307 		return (EIO);
308 	memcpy(buf, tmp, size);
309 	return (0);
310 }
311 
312 static int
313 vdev_clear_pad2(vdev_t *vdev)
314 {
315 	char *zeroes = zap_scratch;
316 	uint64_t *end;
317 	off_t off = offsetof(vdev_label_t, vl_pad2);
318 
319 	memset(zeroes, 0, VDEV_PAD_SIZE);
320 	end = (uint64_t *)(zeroes + VDEV_PAD_SIZE);
321 	/* ZIO_CHECKSUM_LABEL magic and pre-calcualted checksum for all zeros */
322 	end[-5] = 0x0210da7ab10c7a11;
323 	end[-4] = 0x97f48f807f6e2a3f;
324 	end[-3] = 0xaf909f1658aacefc;
325 	end[-2] = 0xcbd1ea57ff6db48b;
326 	end[-1] = 0x6ec692db0d465fab;
327 	if (vdev_write(vdev, vdev->v_read_priv, off, zeroes, VDEV_PAD_SIZE))
328 		return (EIO);
329 	return (0);
330 }
331 
332 static void
333 bios_getmem(void)
334 {
335     uint64_t size;
336 
337     /* Parse system memory map */
338     v86.ebx = 0;
339     do {
340 	v86.ctl = V86_FLAGS;
341 	v86.addr = 0x15;		/* int 0x15 function 0xe820*/
342 	v86.eax = 0xe820;
343 	v86.ecx = sizeof(struct bios_smap);
344 	v86.edx = SMAP_SIG;
345 	v86.es = VTOPSEG(&smap);
346 	v86.edi = VTOPOFF(&smap);
347 	v86int();
348 	if (V86_CY(v86.efl) || (v86.eax != SMAP_SIG))
349 	    break;
350 	/* look for a low-memory segment that's large enough */
351 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0) &&
352 	    (smap.length >= (512 * 1024)))
353 	    bios_basemem = smap.length;
354 	/* look for the first segment in 'extended' memory */
355 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0x100000)) {
356 	    bios_extmem = smap.length;
357 	}
358 
359 	/*
360 	 * Look for the largest segment in 'extended' memory beyond
361 	 * 1MB but below 4GB.
362 	 */
363 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base > 0x100000) &&
364 	    (smap.base < 0x100000000ull)) {
365 	    size = smap.length;
366 
367 	    /*
368 	     * If this segment crosses the 4GB boundary, truncate it.
369 	     */
370 	    if (smap.base + size > 0x100000000ull)
371 		size = 0x100000000ull - smap.base;
372 
373 	    if (size > high_heap_size) {
374 		high_heap_size = size;
375 		high_heap_base = smap.base;
376 	    }
377 	}
378     } while (v86.ebx != 0);
379 
380     /* Fall back to the old compatibility function for base memory */
381     if (bios_basemem == 0) {
382 	v86.ctl = 0;
383 	v86.addr = 0x12;		/* int 0x12 */
384 	v86int();
385 
386 	bios_basemem = (v86.eax & 0xffff) * 1024;
387     }
388 
389     /* Fall back through several compatibility functions for extended memory */
390     if (bios_extmem == 0) {
391 	v86.ctl = V86_FLAGS;
392 	v86.addr = 0x15;		/* int 0x15 function 0xe801*/
393 	v86.eax = 0xe801;
394 	v86int();
395 	if (!V86_CY(v86.efl)) {
396 	    bios_extmem = ((v86.ecx & 0xffff) + ((v86.edx & 0xffff) * 64)) * 1024;
397 	}
398     }
399     if (bios_extmem == 0) {
400 	v86.ctl = 0;
401 	v86.addr = 0x15;		/* int 0x15 function 0x88*/
402 	v86.eax = 0x8800;
403 	v86int();
404 	bios_extmem = (v86.eax & 0xffff) * 1024;
405     }
406 
407     /*
408      * If we have extended memory and did not find a suitable heap
409      * region in the SMAP, use the last 3MB of 'extended' memory as a
410      * high heap candidate.
411      */
412     if (bios_extmem >= HEAP_MIN && high_heap_size < HEAP_MIN) {
413 	high_heap_size = HEAP_MIN;
414 	high_heap_base = bios_extmem + 0x100000 - HEAP_MIN;
415     }
416 }
417 
418 /*
419  * Try to detect a device supported by the legacy int13 BIOS
420  */
421 static int
422 int13probe(int drive)
423 {
424     v86.ctl = V86_FLAGS;
425     v86.addr = 0x13;
426     v86.eax = 0x800;
427     v86.edx = drive;
428     v86int();
429 
430     if (!V86_CY(v86.efl) &&				/* carry clear */
431 	((v86.edx & 0xff) != (drive & DRV_MASK))) {	/* unit # OK */
432 	if ((v86.ecx & 0x3f) == 0) {			/* absurd sector size */
433 		return(0);				/* skip device */
434 	}
435 	return (1);
436     }
437     return(0);
438 }
439 
440 /*
441  * We call this when we find a ZFS vdev - ZFS consumes the dsk
442  * structure so we must make a new one.
443  */
444 static struct dsk *
445 copy_dsk(struct dsk *dsk)
446 {
447     struct dsk *newdsk;
448 
449     newdsk = malloc(sizeof(struct dsk));
450     *newdsk = *dsk;
451     return (newdsk);
452 }
453 
454 /*
455  * Get disk size from eax=0x800 and 0x4800. We need to probe both
456  * because 0x4800 may not be available and we would like to get more
457  * or less correct disk size - if it is possible at all.
458  * Note we do not really want to touch drv.c because that code is shared
459  * with boot2 and we can not afford to grow that code.
460  */
461 static uint64_t
462 drvsize_ext(struct dsk *dskp)
463 {
464 	uint64_t size, tmp;
465 	int cyl, hds, sec;
466 
467 	v86.ctl = V86_FLAGS;
468 	v86.addr = 0x13;
469 	v86.eax = 0x800;
470 	v86.edx = dskp->drive;
471 	v86int();
472 
473 	/* Don't error out if we get bad sector number, try EDD as well */
474 	if (V86_CY(v86.efl) ||	/* carry set */
475 	    (v86.edx & 0xff) <= (unsigned)(dskp->drive & 0x7f)) /* unit # bad */
476 		return (0);
477 	cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
478 	/* Convert max head # -> # of heads */
479 	hds = ((v86.edx & 0xff00) >> 8) + 1;
480 	sec = v86.ecx & 0x3f;
481 
482 	size = (uint64_t)cyl * hds * sec;
483 
484 	/* Determine if we can use EDD with this device. */
485 	v86.ctl = V86_FLAGS;
486 	v86.addr = 0x13;
487 	v86.eax = 0x4100;
488 	v86.edx = dskp->drive;
489 	v86.ebx = 0x55aa;
490 	v86int();
491 	if (V86_CY(v86.efl) ||  /* carry set */
492 	    (v86.ebx & 0xffff) != 0xaa55 || /* signature */
493 	    (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
494 		return (size);
495 
496 	tmp = drvsize(dskp);
497 	if (tmp > size)
498 		size = tmp;
499 
500 	return (size);
501 }
502 
503 /*
504  * The "layered" ioctl to read disk/partition size. Unfortunately
505  * the zfsboot case is hardest, because we do not have full software
506  * stack available, so we need to do some manual work here.
507  */
508 uint64_t
509 ldi_get_size(void *priv)
510 {
511 	struct dsk *dskp = priv;
512 	uint64_t size = dskp->size;
513 
514 	if (dskp->start == 0)
515 		size = drvsize_ext(dskp);
516 
517 	return (size * DEV_BSIZE);
518 }
519 
520 static void
521 probe_drive(struct dsk *dsk)
522 {
523 #ifdef GPT
524     struct gpt_hdr hdr;
525     struct gpt_ent *ent;
526     unsigned part, entries_per_sec;
527     daddr_t slba;
528 #endif
529 #if defined(GPT) || defined(LOADER_GELI_SUPPORT)
530     daddr_t elba;
531 #endif
532 
533     struct dos_partition *dp;
534     char *sec;
535     unsigned i;
536 
537     /*
538      * If we find a vdev on the whole disk, stop here.
539      */
540     if (vdev_probe(vdev_read2, dsk, NULL) == 0)
541 	return;
542 
543 #ifdef LOADER_GELI_SUPPORT
544     /*
545      * Taste the disk, if it is GELI encrypted, decrypt it and check to see if
546      * it is a usable vdev then. Otherwise dig
547      * out the partition table and probe each slice/partition
548      * in turn for a vdev or GELI encrypted vdev.
549      */
550     elba = drvsize_ext(dsk);
551     if (elba > 0) {
552 	elba--;
553     }
554     if (geli_taste(vdev_read, dsk, elba) == 0) {
555 	if (geli_havekey(dsk) == 0 || geli_passphrase(gelipw, dsk->unit,
556 	  ':', 0, dsk) == 0) {
557 	    if (vdev_probe(vdev_read2, dsk, NULL) == 0) {
558 		return;
559 	    }
560 	}
561     }
562 #endif /* LOADER_GELI_SUPPORT */
563 
564     sec = dmadat->secbuf;
565     dsk->start = 0;
566 
567 #ifdef GPT
568     /*
569      * First check for GPT.
570      */
571     if (drvread(dsk, sec, 1, 1)) {
572 	return;
573     }
574     memcpy(&hdr, sec, sizeof(hdr));
575     if (memcmp(hdr.hdr_sig, GPT_HDR_SIG, sizeof(hdr.hdr_sig)) != 0 ||
576 	hdr.hdr_lba_self != 1 || hdr.hdr_revision < 0x00010000 ||
577 	hdr.hdr_entsz < sizeof(*ent) || DEV_BSIZE % hdr.hdr_entsz != 0) {
578 	goto trymbr;
579     }
580 
581     /*
582      * Probe all GPT partitions for the presence of ZFS pools. We
583      * return the spa_t for the first we find (if requested). This
584      * will have the effect of booting from the first pool on the
585      * disk.
586      *
587      * If no vdev is found, GELI decrypting the device and try again
588      */
589     entries_per_sec = DEV_BSIZE / hdr.hdr_entsz;
590     slba = hdr.hdr_lba_table;
591     elba = slba + hdr.hdr_entries / entries_per_sec;
592     while (slba < elba) {
593 	dsk->start = 0;
594 	if (drvread(dsk, sec, slba, 1))
595 	    return;
596 	for (part = 0; part < entries_per_sec; part++) {
597 	    ent = (struct gpt_ent *)(sec + part * hdr.hdr_entsz);
598 	    if (memcmp(&ent->ent_type, &freebsd_zfs_uuid,
599 		     sizeof(uuid_t)) == 0) {
600 		dsk->start = ent->ent_lba_start;
601 		dsk->size = ent->ent_lba_end - ent->ent_lba_start + 1;
602 		dsk->slice = part + 1;
603 		dsk->part = 255;
604 		if (vdev_probe(vdev_read2, dsk, NULL) == 0) {
605 		    /*
606 		     * This slice had a vdev. We need a new dsk
607 		     * structure now since the vdev now owns this one.
608 		     */
609 		    dsk = copy_dsk(dsk);
610 		}
611 #ifdef LOADER_GELI_SUPPORT
612 		else if (geli_taste(vdev_read, dsk, ent->ent_lba_end -
613 			 ent->ent_lba_start) == 0) {
614 		    if (geli_havekey(dsk) == 0 || geli_passphrase(gelipw,
615 		      dsk->unit, 'p', dsk->slice, dsk) == 0) {
616 			/*
617 			 * This slice has GELI, check it for ZFS.
618 			 */
619 			if (vdev_probe(vdev_read2, dsk, NULL) == 0) {
620 			    /*
621 			     * This slice had a vdev. We need a new dsk
622 			     * structure now since the vdev now owns this one.
623 			     */
624 			    dsk = copy_dsk(dsk);
625 			}
626 			break;
627 		    }
628 		}
629 #endif /* LOADER_GELI_SUPPORT */
630 	    }
631 	}
632 	slba++;
633     }
634     return;
635 trymbr:
636 #endif /* GPT */
637 
638     if (drvread(dsk, sec, DOSBBSECTOR, 1))
639 	return;
640     dp = (void *)(sec + DOSPARTOFF);
641 
642     for (i = 0; i < NDOSPART; i++) {
643 	if (!dp[i].dp_typ)
644 	    continue;
645 	dsk->start = dp[i].dp_start;
646 	dsk->size = dp[i].dp_size;
647 	dsk->slice = i + 1;
648 	if (vdev_probe(vdev_read2, dsk, NULL) == 0) {
649 	    dsk = copy_dsk(dsk);
650 	}
651 #ifdef LOADER_GELI_SUPPORT
652 	else if (geli_taste(vdev_read, dsk, dp[i].dp_size -
653 		 dp[i].dp_start) == 0) {
654 	    if (geli_havekey(dsk) == 0 || geli_passphrase(gelipw, dsk->unit,
655 	      's', i, dsk) == 0) {
656 		/*
657 		 * This slice has GELI, check it for ZFS.
658 		 */
659 		if (vdev_probe(vdev_read2, dsk, NULL) == 0) {
660 		    /*
661 		     * This slice had a vdev. We need a new dsk
662 		     * structure now since the vdev now owns this one.
663 		     */
664 		    dsk = copy_dsk(dsk);
665 		}
666 		break;
667 	    }
668 	}
669 #endif /* LOADER_GELI_SUPPORT */
670     }
671 }
672 
673 int
674 main(void)
675 {
676     dnode_phys_t dn;
677     off_t off;
678     struct dsk *dsk;
679     int autoboot, i;
680     int nextboot;
681     int rc;
682 
683     dmadat = (void *)(roundup2(__base + (int32_t)&_end, 0x10000) - __base);
684 
685     bios_getmem();
686 
687     if (high_heap_size > 0) {
688 	heap_end = PTOV(high_heap_base + high_heap_size);
689 	heap_next = PTOV(high_heap_base);
690     } else {
691 	heap_next = (char *)dmadat + sizeof(*dmadat);
692 	heap_end = (char *)PTOV(bios_basemem);
693     }
694     setheap(heap_next, heap_end);
695 
696     dsk = malloc(sizeof(struct dsk));
697     dsk->drive = *(uint8_t *)PTOV(ARGS);
698     dsk->type = dsk->drive & DRV_HARD ? TYPE_AD : TYPE_FD;
699     dsk->unit = dsk->drive & DRV_MASK;
700     dsk->slice = *(uint8_t *)PTOV(ARGS + 1) + 1;
701     dsk->part = 0;
702     dsk->start = 0;
703     dsk->size = drvsize_ext(dsk);
704 
705     bootinfo.bi_version = BOOTINFO_VERSION;
706     bootinfo.bi_size = sizeof(bootinfo);
707     bootinfo.bi_basemem = bios_basemem / 1024;
708     bootinfo.bi_extmem = bios_extmem / 1024;
709     bootinfo.bi_memsizes_valid++;
710     bootinfo.bi_bios_dev = dsk->drive;
711 
712     bootdev = MAKEBOOTDEV(dev_maj[dsk->type],
713 			  dsk->slice, dsk->unit, dsk->part);
714 
715     /* Process configuration file */
716 
717     autoboot = 1;
718 
719 #ifdef LOADER_GELI_SUPPORT
720     geli_init();
721 #endif
722     zfs_init();
723 
724     /*
725      * Probe the boot drive first - we will try to boot from whatever
726      * pool we find on that drive.
727      */
728     probe_drive(dsk);
729 
730     /*
731      * Probe the rest of the drives that the bios knows about. This
732      * will find any other available pools and it may fill in missing
733      * vdevs for the boot pool.
734      */
735 #ifndef VIRTUALBOX
736     for (i = 0; i < *(unsigned char *)PTOV(BIOS_NUMDRIVES); i++)
737 #else
738     for (i = 0; i < MAXBDDEV; i++)
739 #endif
740     {
741 	if ((i | DRV_HARD) == *(uint8_t *)PTOV(ARGS))
742 	    continue;
743 
744 	if (!int13probe(i | DRV_HARD))
745 	    break;
746 
747 	dsk = malloc(sizeof(struct dsk));
748 	dsk->drive = i | DRV_HARD;
749 	dsk->type = dsk->drive & TYPE_AD;
750 	dsk->unit = i;
751 	dsk->slice = 0;
752 	dsk->part = 0;
753 	dsk->start = 0;
754 	dsk->size = drvsize_ext(dsk);
755 	probe_drive(dsk);
756     }
757 
758     /*
759      * The first discovered pool, if any, is the pool.
760      */
761     spa = spa_get_primary();
762     if (!spa) {
763 	printf("%s: No ZFS pools located, can't boot\n", BOOTPROG);
764 	for (;;)
765 	    ;
766     }
767 
768     primary_spa = spa;
769     primary_vdev = spa_get_primary_vdev(spa);
770 
771     nextboot = 0;
772     rc  = vdev_read_pad2(primary_vdev, cmd, sizeof(cmd));
773     if (vdev_clear_pad2(primary_vdev))
774 	printf("failed to clear pad2 area of primary vdev\n");
775     if (rc == 0) {
776 	if (*cmd) {
777 	    /*
778 	     * We could find an old-style ZFS Boot Block header here.
779 	     * Simply ignore it.
780 	     */
781 	    if (*(uint64_t *)cmd != 0x2f5b007b10c) {
782 		/*
783 		 * Note that parse() is destructive to cmd[] and we also want
784 		 * to honor RBX_QUIET option that could be present in cmd[].
785 		 */
786 		nextboot = 1;
787 		memcpy(cmddup, cmd, sizeof(cmd));
788 		if (parse_cmd()) {
789 		    printf("failed to parse pad2 area of primary vdev\n");
790 		    reboot();
791 		}
792 		if (!OPT_CHECK(RBX_QUIET))
793 		    printf("zfs nextboot: %s\n", cmddup);
794 	    }
795 	    /* Do not process this command twice */
796 	    *cmd = 0;
797 	}
798     } else
799 	printf("failed to read pad2 area of primary vdev\n");
800 
801     /* Mount ZFS only if it's not already mounted via nextboot parsing. */
802     if (zfsmount.spa == NULL &&
803 	(zfs_spa_init(spa) != 0 || zfs_mount(spa, 0, &zfsmount) != 0)) {
804 	printf("%s: failed to mount default pool %s\n",
805 	    BOOTPROG, spa->spa_name);
806 	autoboot = 0;
807     } else if (zfs_lookup(&zfsmount, PATH_CONFIG, &dn) == 0 ||
808         zfs_lookup(&zfsmount, PATH_DOTCONFIG, &dn) == 0) {
809 	off = 0;
810 	zfs_read(spa, &dn, &off, cmd, sizeof(cmd));
811     }
812 
813     if (*cmd) {
814 	/*
815 	 * Note that parse_cmd() is destructive to cmd[] and we also want
816 	 * to honor RBX_QUIET option that could be present in cmd[].
817 	 */
818 	memcpy(cmddup, cmd, sizeof(cmd));
819 	if (parse_cmd())
820 	    autoboot = 0;
821 	if (!OPT_CHECK(RBX_QUIET))
822 	    printf("%s: %s\n", PATH_CONFIG, cmddup);
823 	/* Do not process this command twice */
824 	*cmd = 0;
825     }
826 
827     /* Do not risk waiting at the prompt forever. */
828     if (nextboot && !autoboot)
829 	reboot();
830 
831     /*
832      * Try to exec /boot/loader. If interrupted by a keypress,
833      * or in case of failure, try to load a kernel directly instead.
834      */
835 
836     if (autoboot && !*kname) {
837 	memcpy(kname, PATH_LOADER_ZFS, sizeof(PATH_LOADER_ZFS));
838 	if (!keyhit(3)) {
839 	    load();
840 	    memcpy(kname, PATH_KERNEL, sizeof(PATH_KERNEL));
841 	}
842     }
843 
844     /* Present the user with the boot2 prompt. */
845 
846     for (;;) {
847 	if (!autoboot || !OPT_CHECK(RBX_QUIET)) {
848 	    printf("\nFreeBSD/x86 boot\n");
849 	    if (zfs_rlookup(spa, zfsmount.rootobj, rootname) != 0)
850 		printf("Default: %s/<0x%llx>:%s\n"
851 		       "boot: ",
852 		       spa->spa_name, zfsmount.rootobj, kname);
853 	    else if (rootname[0] != '\0')
854 		printf("Default: %s/%s:%s\n"
855 		       "boot: ",
856 		       spa->spa_name, rootname, kname);
857 	    else
858 		printf("Default: %s:%s\n"
859 		       "boot: ",
860 		       spa->spa_name, kname);
861 	}
862 	if (ioctrl & IO_SERIAL)
863 	    sio_flush();
864 	if (!autoboot || keyhit(5))
865 	    getstr(cmd, sizeof(cmd));
866 	else if (!autoboot || !OPT_CHECK(RBX_QUIET))
867 	    putchar('\n');
868 	autoboot = 0;
869 	if (parse_cmd())
870 	    putchar('\a');
871 	else
872 	    load();
873     }
874 }
875 
876 /* XXX - Needed for btxld to link the boot2 binary; do not remove. */
877 void
878 exit(int x)
879 {
880     __exit(x);
881 }
882 
883 void
884 reboot(void)
885 {
886     __exit(0);
887 }
888 
889 static void
890 load(void)
891 {
892     union {
893 	struct exec ex;
894 	Elf32_Ehdr eh;
895     } hdr;
896     static Elf32_Phdr ep[2];
897     static Elf32_Shdr es[2];
898     caddr_t p;
899     dnode_phys_t dn;
900     off_t off;
901     uint32_t addr, x;
902     int fmt, i, j;
903 
904     if (zfs_lookup(&zfsmount, kname, &dn)) {
905 	printf("\nCan't find %s\n", kname);
906 	return;
907     }
908     off = 0;
909     if (xfsread(&dn, &off, &hdr, sizeof(hdr)))
910 	return;
911     if (N_GETMAGIC(hdr.ex) == ZMAGIC)
912 	fmt = 0;
913     else if (IS_ELF(hdr.eh))
914 	fmt = 1;
915     else {
916 	printf("Invalid %s\n", "format");
917 	return;
918     }
919     if (fmt == 0) {
920 	addr = hdr.ex.a_entry & 0xffffff;
921 	p = PTOV(addr);
922 	off = PAGE_SIZE;
923 	if (xfsread(&dn, &off, p, hdr.ex.a_text))
924 	    return;
925 	p += roundup2(hdr.ex.a_text, PAGE_SIZE);
926 	if (xfsread(&dn, &off, p, hdr.ex.a_data))
927 	    return;
928 	p += hdr.ex.a_data + roundup2(hdr.ex.a_bss, PAGE_SIZE);
929 	bootinfo.bi_symtab = VTOP(p);
930 	memcpy(p, &hdr.ex.a_syms, sizeof(hdr.ex.a_syms));
931 	p += sizeof(hdr.ex.a_syms);
932 	if (hdr.ex.a_syms) {
933 	    if (xfsread(&dn, &off, p, hdr.ex.a_syms))
934 		return;
935 	    p += hdr.ex.a_syms;
936 	    if (xfsread(&dn, &off, p, sizeof(int)))
937 		return;
938 	    x = *(uint32_t *)p;
939 	    p += sizeof(int);
940 	    x -= sizeof(int);
941 	    if (xfsread(&dn, &off, p, x))
942 		return;
943 	    p += x;
944 	}
945     } else {
946 	off = hdr.eh.e_phoff;
947 	for (j = i = 0; i < hdr.eh.e_phnum && j < 2; i++) {
948 	    if (xfsread(&dn, &off, ep + j, sizeof(ep[0])))
949 		return;
950 	    if (ep[j].p_type == PT_LOAD)
951 		j++;
952 	}
953 	for (i = 0; i < 2; i++) {
954 	    p = PTOV(ep[i].p_paddr & 0xffffff);
955 	    off = ep[i].p_offset;
956 	    if (xfsread(&dn, &off, p, ep[i].p_filesz))
957 		return;
958 	}
959 	p += roundup2(ep[1].p_memsz, PAGE_SIZE);
960 	bootinfo.bi_symtab = VTOP(p);
961 	if (hdr.eh.e_shnum == hdr.eh.e_shstrndx + 3) {
962 	    off = hdr.eh.e_shoff + sizeof(es[0]) *
963 		(hdr.eh.e_shstrndx + 1);
964 	    if (xfsread(&dn, &off, &es, sizeof(es)))
965 		return;
966 	    for (i = 0; i < 2; i++) {
967 		memcpy(p, &es[i].sh_size, sizeof(es[i].sh_size));
968 		p += sizeof(es[i].sh_size);
969 		off = es[i].sh_offset;
970 		if (xfsread(&dn, &off, p, es[i].sh_size))
971 		    return;
972 		p += es[i].sh_size;
973 	    }
974 	}
975 	addr = hdr.eh.e_entry & 0xffffff;
976     }
977     bootinfo.bi_esymtab = VTOP(p);
978     bootinfo.bi_kernelname = VTOP(kname);
979     zfsargs.size = sizeof(zfsargs);
980     zfsargs.pool = zfsmount.spa->spa_guid;
981     zfsargs.root = zfsmount.rootobj;
982     zfsargs.primary_pool = primary_spa->spa_guid;
983 #ifdef LOADER_GELI_SUPPORT
984     explicit_bzero(gelipw, sizeof(gelipw));
985     gelibuf = malloc(sizeof(struct keybuf) + (GELI_MAX_KEYS * sizeof(struct keybuf_ent)));
986     geli_fill_keybuf(gelibuf);
987     zfsargs.notapw = '\0';
988     zfsargs.keybuf_sentinel = KEYBUF_SENTINEL;
989     zfsargs.keybuf = gelibuf;
990 #else
991     zfsargs.gelipw[0] = '\0';
992 #endif
993     if (primary_vdev != NULL)
994 	zfsargs.primary_vdev = primary_vdev->v_guid;
995     else
996 	printf("failed to detect primary vdev\n");
997     __exec((caddr_t)addr, RB_BOOTINFO | (opts & RBX_MASK),
998 	   bootdev,
999 	   KARGS_FLAGS_ZFS | KARGS_FLAGS_EXTARG,
1000 	   (uint32_t) spa->spa_guid,
1001 	   (uint32_t) (spa->spa_guid >> 32),
1002 	   VTOP(&bootinfo),
1003 	   zfsargs);
1004 }
1005 
1006 static int
1007 zfs_mount_ds(char *dsname)
1008 {
1009     uint64_t newroot;
1010     spa_t *newspa;
1011     char *q;
1012 
1013     q = strchr(dsname, '/');
1014     if (q)
1015 	*q++ = '\0';
1016     newspa = spa_find_by_name(dsname);
1017     if (newspa == NULL) {
1018 	printf("\nCan't find ZFS pool %s\n", dsname);
1019 	return -1;
1020     }
1021 
1022     if (zfs_spa_init(newspa))
1023 	return -1;
1024 
1025     newroot = 0;
1026     if (q) {
1027 	if (zfs_lookup_dataset(newspa, q, &newroot)) {
1028 	    printf("\nCan't find dataset %s in ZFS pool %s\n",
1029 		    q, newspa->spa_name);
1030 	    return -1;
1031 	}
1032     }
1033     if (zfs_mount(newspa, newroot, &zfsmount)) {
1034 	printf("\nCan't mount ZFS dataset\n");
1035 	return -1;
1036     }
1037     spa = newspa;
1038     return (0);
1039 }
1040 
1041 static int
1042 parse_cmd(void)
1043 {
1044     char *arg = cmd;
1045     char *ep, *p, *q;
1046     const char *cp;
1047     int c, i, j;
1048 
1049     while ((c = *arg++)) {
1050 	if (c == ' ' || c == '\t' || c == '\n')
1051 	    continue;
1052 	for (p = arg; *p && *p != '\n' && *p != ' ' && *p != '\t'; p++);
1053 	ep = p;
1054 	if (*p)
1055 	    *p++ = 0;
1056 	if (c == '-') {
1057 	    while ((c = *arg++)) {
1058 		if (c == 'P') {
1059 		    if (*(uint8_t *)PTOV(0x496) & 0x10) {
1060 			cp = "yes";
1061 		    } else {
1062 			opts |= OPT_SET(RBX_DUAL) | OPT_SET(RBX_SERIAL);
1063 			cp = "no";
1064 		    }
1065 		    printf("Keyboard: %s\n", cp);
1066 		    continue;
1067 		} else if (c == 'S') {
1068 		    j = 0;
1069 		    while ((unsigned int)(i = *arg++ - '0') <= 9)
1070 			j = j * 10 + i;
1071 		    if (j > 0 && i == -'0') {
1072 			comspeed = j;
1073 			break;
1074 		    }
1075 		    /* Fall through to error below ('S' not in optstr[]). */
1076 		}
1077 		for (i = 0; c != optstr[i]; i++)
1078 		    if (i == NOPT - 1)
1079 			return -1;
1080 		opts ^= OPT_SET(flags[i]);
1081 	    }
1082 	    ioctrl = OPT_CHECK(RBX_DUAL) ? (IO_SERIAL|IO_KEYBOARD) :
1083 		     OPT_CHECK(RBX_SERIAL) ? IO_SERIAL : IO_KEYBOARD;
1084 	    if (ioctrl & IO_SERIAL) {
1085 	        if (sio_init(115200 / comspeed) != 0)
1086 		    ioctrl &= ~IO_SERIAL;
1087 	    }
1088 	} if (c == '?') {
1089 	    dnode_phys_t dn;
1090 
1091 	    if (zfs_lookup(&zfsmount, arg, &dn) == 0) {
1092 		zap_list(spa, &dn);
1093 	    }
1094 	    return -1;
1095 	} else {
1096 	    arg--;
1097 
1098 	    /*
1099 	     * Report pool status if the comment is 'status'. Lets
1100 	     * hope no-one wants to load /status as a kernel.
1101 	     */
1102 	    if (!strcmp(arg, "status")) {
1103 		spa_all_status();
1104 		return -1;
1105 	    }
1106 
1107 	    /*
1108 	     * If there is "zfs:" prefix simply ignore it.
1109 	     */
1110 	    if (strncmp(arg, "zfs:", 4) == 0)
1111 		arg += 4;
1112 
1113 	    /*
1114 	     * If there is a colon, switch pools.
1115 	     */
1116 	    q = strchr(arg, ':');
1117 	    if (q) {
1118 		*q++ = '\0';
1119 		if (zfs_mount_ds(arg) != 0)
1120 		    return -1;
1121 		arg = q;
1122 	    }
1123 	    if ((i = ep - arg)) {
1124 		if ((size_t)i >= sizeof(kname))
1125 		    return -1;
1126 		memcpy(kname, arg, i + 1);
1127 	    }
1128 	}
1129 	arg = p;
1130     }
1131     return 0;
1132 }
1133