zfsboot.c revision 344399
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: stable/11/stand/i386/zfsboot/zfsboot.c 344399 2019-02-20 23:55:35Z kevans $");
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
66extern uint32_t _end;
67
68#ifdef GPT
69static const uuid_t freebsd_zfs_uuid = GPT_ENT_TYPE_FREEBSD_ZFS;
70#endif
71static const char optstr[NOPT] = "DhaCcdgmnpqrsv"; /* Also 'P', 'S' */
72static 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};
88uint32_t opts;
89
90static const unsigned char dev_maj[NDEV] = {30, 4, 2};
91
92static char cmd[512];
93static char cmddup[512];
94static char kname[1024];
95static char rootname[256];
96static int comspeed = SIOSPD;
97static struct bootinfo bootinfo;
98static uint32_t bootdev;
99static struct zfs_boot_args zfsargs;
100
101vm_offset_t	high_heap_base;
102uint32_t	bios_basemem, bios_extmem, high_heap_size;
103
104static 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
111static char *heap_next;
112static char *heap_end;
113
114/* Buffers that must not span a 64k boundary. */
115#define READ_BUF_SIZE		8192
116struct dmadat {
117	char rdbuf[READ_BUF_SIZE];	/* for reading large things */
118	char secbuf[READ_BUF_SIZE];	/* for MBR/disklabel */
119};
120static struct dmadat *dmadat;
121
122void exit(int);
123void reboot(void);
124static void load(void);
125static int parse_cmd(void);
126static void bios_getmem(void);
127int main(void);
128
129#ifdef LOADER_GELI_SUPPORT
130#include "geliboot.h"
131static char gelipw[GELI_PW_MAXLEN];
132#endif
133
134struct zfsdsk {
135	struct dsk       dsk;
136#ifdef LOADER_GELI_SUPPORT
137	struct geli_dev *gdev;
138#endif
139};
140
141#include "zfsimpl.c"
142
143/*
144 * Read from a dnode (which must be from a ZPL filesystem).
145 */
146static int
147zfs_read(spa_t *spa, const dnode_phys_t *dnode, off_t *offp, void *start, size_t size)
148{
149	const znode_phys_t *zp = (const znode_phys_t *) dnode->dn_bonus;
150	size_t n;
151	int rc;
152
153	n = size;
154	if (*offp + n > zp->zp_size)
155		n = zp->zp_size - *offp;
156
157	rc = dnode_read(spa, dnode, *offp, start, n);
158	if (rc)
159		return (-1);
160	*offp += n;
161
162	return (n);
163}
164
165/*
166 * Current ZFS pool
167 */
168static spa_t *spa;
169static spa_t *primary_spa;
170static vdev_t *primary_vdev;
171
172/*
173 * A wrapper for dskread that doesn't have to worry about whether the
174 * buffer pointer crosses a 64k boundary.
175 */
176static int
177vdev_read(void *xvdev, void *priv, off_t off, void *buf, size_t bytes)
178{
179	char *p;
180	daddr_t lba, alignlba;
181	off_t diff;
182	unsigned int nb, alignnb;
183	struct zfsdsk *zdsk = (struct zfsdsk *) priv;
184
185	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
186		return -1;
187
188	p = buf;
189	lba = off / DEV_BSIZE;
190	lba += zdsk->dsk.start;
191	/*
192	 * Align reads to 4k else 4k sector GELIs will not decrypt.
193	 * Round LBA down to nearest multiple of DEV_GELIBOOT_BSIZE bytes.
194	 */
195	alignlba = rounddown2(off, DEV_GELIBOOT_BSIZE) / DEV_BSIZE;
196	/*
197	 * The read must be aligned to DEV_GELIBOOT_BSIZE bytes relative to the
198	 * start of the GELI partition, not the start of the actual disk.
199	 */
200	alignlba += zdsk->dsk.start;
201	diff = (lba - alignlba) * DEV_BSIZE;
202
203	while (bytes > 0) {
204		nb = bytes / DEV_BSIZE;
205		/*
206		 * Ensure that the read size plus the leading offset does not
207		 * exceed the size of the read buffer.
208		 */
209		if (nb > (READ_BUF_SIZE - diff) / DEV_BSIZE)
210			nb = (READ_BUF_SIZE - diff) / DEV_BSIZE;
211		/*
212		 * Round the number of blocks to read up to the nearest multiple
213		 * of DEV_GELIBOOT_BSIZE.
214		 */
215		alignnb = roundup2(nb * DEV_BSIZE + diff, DEV_GELIBOOT_BSIZE)
216		    / DEV_BSIZE;
217
218		if (zdsk->dsk.size > 0 && alignlba + alignnb >
219		    zdsk->dsk.size + zdsk->dsk.start) {
220			printf("Shortening read at %lld from %d to %lld\n",
221			    alignlba, alignnb,
222			    (zdsk->dsk.size + zdsk->dsk.start) - alignlba);
223			alignnb = (zdsk->dsk.size + zdsk->dsk.start) - alignlba;
224		}
225
226		if (drvread(&zdsk->dsk, dmadat->rdbuf, alignlba, alignnb))
227			return -1;
228#ifdef LOADER_GELI_SUPPORT
229		/* decrypt */
230		if (zdsk->gdev != NULL) {
231			if (geli_read(zdsk->gdev, ((alignlba - zdsk->dsk.start) *
232			    DEV_BSIZE), dmadat->rdbuf, alignnb * DEV_BSIZE))
233				return (-1);
234		}
235#endif
236		memcpy(p, dmadat->rdbuf + diff, nb * DEV_BSIZE);
237		p += nb * DEV_BSIZE;
238		lba += nb;
239		alignlba += alignnb;
240		bytes -= nb * DEV_BSIZE;
241		/* Don't need the leading offset after the first block. */
242		diff = 0;
243	}
244
245	return 0;
246}
247/* Match the signature exactly due to signature madness */
248static int
249vdev_read2(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
250{
251	return vdev_read(vdev, priv, off, buf, bytes);
252}
253
254
255static int
256vdev_write(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
257{
258	char *p;
259	daddr_t lba;
260	unsigned int nb;
261	struct zfsdsk *zdsk = (struct zfsdsk *) priv;
262
263	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
264		return -1;
265
266	p = buf;
267	lba = off / DEV_BSIZE;
268	lba += zdsk->dsk.start;
269	while (bytes > 0) {
270		nb = bytes / DEV_BSIZE;
271		if (nb > READ_BUF_SIZE / DEV_BSIZE)
272			nb = READ_BUF_SIZE / DEV_BSIZE;
273		memcpy(dmadat->rdbuf, p, nb * DEV_BSIZE);
274		if (drvwrite(&zdsk->dsk, dmadat->rdbuf, lba, nb))
275			return -1;
276		p += nb * DEV_BSIZE;
277		lba += nb;
278		bytes -= nb * DEV_BSIZE;
279	}
280
281	return 0;
282}
283
284static int
285xfsread(const dnode_phys_t *dnode, off_t *offp, void *buf, size_t nbyte)
286{
287    if ((size_t)zfs_read(spa, dnode, offp, buf, nbyte) != nbyte) {
288	printf("Invalid format\n");
289	return -1;
290    }
291    return 0;
292}
293
294/*
295 * Read Pad2 (formerly "Boot Block Header") area of the first
296 * vdev label of the given vdev.
297 */
298static int
299vdev_read_pad2(vdev_t *vdev, char *buf, size_t size)
300{
301	blkptr_t bp;
302	char *tmp = zap_scratch;
303	off_t off = offsetof(vdev_label_t, vl_pad2);
304
305	if (size > VDEV_PAD_SIZE)
306		size = VDEV_PAD_SIZE;
307
308	BP_ZERO(&bp);
309	BP_SET_LSIZE(&bp, VDEV_PAD_SIZE);
310	BP_SET_PSIZE(&bp, VDEV_PAD_SIZE);
311	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
312	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
313	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
314	if (vdev_read_phys(vdev, &bp, tmp, off, 0))
315		return (EIO);
316	memcpy(buf, tmp, size);
317	return (0);
318}
319
320static int
321vdev_clear_pad2(vdev_t *vdev)
322{
323	char *zeroes = zap_scratch;
324	uint64_t *end;
325	off_t off = offsetof(vdev_label_t, vl_pad2);
326
327	memset(zeroes, 0, VDEV_PAD_SIZE);
328	end = (uint64_t *)(zeroes + VDEV_PAD_SIZE);
329	/* ZIO_CHECKSUM_LABEL magic and pre-calcualted checksum for all zeros */
330	end[-5] = 0x0210da7ab10c7a11;
331	end[-4] = 0x97f48f807f6e2a3f;
332	end[-3] = 0xaf909f1658aacefc;
333	end[-2] = 0xcbd1ea57ff6db48b;
334	end[-1] = 0x6ec692db0d465fab;
335	if (vdev_write(vdev, vdev->v_read_priv, off, zeroes, VDEV_PAD_SIZE))
336		return (EIO);
337	return (0);
338}
339
340static void
341bios_getmem(void)
342{
343    uint64_t size;
344
345    /* Parse system memory map */
346    v86.ebx = 0;
347    do {
348	v86.ctl = V86_FLAGS;
349	v86.addr = 0x15;		/* int 0x15 function 0xe820*/
350	v86.eax = 0xe820;
351	v86.ecx = sizeof(struct bios_smap);
352	v86.edx = SMAP_SIG;
353	v86.es = VTOPSEG(&smap);
354	v86.edi = VTOPOFF(&smap);
355	v86int();
356	if (V86_CY(v86.efl) || (v86.eax != SMAP_SIG))
357	    break;
358	/* look for a low-memory segment that's large enough */
359	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0) &&
360	    (smap.length >= (512 * 1024)))
361	    bios_basemem = smap.length;
362	/* look for the first segment in 'extended' memory */
363	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0x100000)) {
364	    bios_extmem = smap.length;
365	}
366
367	/*
368	 * Look for the largest segment in 'extended' memory beyond
369	 * 1MB but below 4GB.
370	 */
371	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base > 0x100000) &&
372	    (smap.base < 0x100000000ull)) {
373	    size = smap.length;
374
375	    /*
376	     * If this segment crosses the 4GB boundary, truncate it.
377	     */
378	    if (smap.base + size > 0x100000000ull)
379		size = 0x100000000ull - smap.base;
380
381	    if (size > high_heap_size) {
382		high_heap_size = size;
383		high_heap_base = smap.base;
384	    }
385	}
386    } while (v86.ebx != 0);
387
388    /* Fall back to the old compatibility function for base memory */
389    if (bios_basemem == 0) {
390	v86.ctl = 0;
391	v86.addr = 0x12;		/* int 0x12 */
392	v86int();
393
394	bios_basemem = (v86.eax & 0xffff) * 1024;
395    }
396
397    /* Fall back through several compatibility functions for extended memory */
398    if (bios_extmem == 0) {
399	v86.ctl = V86_FLAGS;
400	v86.addr = 0x15;		/* int 0x15 function 0xe801*/
401	v86.eax = 0xe801;
402	v86int();
403	if (!V86_CY(v86.efl)) {
404	    bios_extmem = ((v86.ecx & 0xffff) + ((v86.edx & 0xffff) * 64)) * 1024;
405	}
406    }
407    if (bios_extmem == 0) {
408	v86.ctl = 0;
409	v86.addr = 0x15;		/* int 0x15 function 0x88*/
410	v86.eax = 0x8800;
411	v86int();
412	bios_extmem = (v86.eax & 0xffff) * 1024;
413    }
414
415    /*
416     * If we have extended memory and did not find a suitable heap
417     * region in the SMAP, use the last 3MB of 'extended' memory as a
418     * high heap candidate.
419     */
420    if (bios_extmem >= HEAP_MIN && high_heap_size < HEAP_MIN) {
421	high_heap_size = HEAP_MIN;
422	high_heap_base = bios_extmem + 0x100000 - HEAP_MIN;
423    }
424}
425
426/*
427 * Try to detect a device supported by the legacy int13 BIOS
428 */
429static int
430int13probe(int drive)
431{
432    v86.ctl = V86_FLAGS;
433    v86.addr = 0x13;
434    v86.eax = 0x800;
435    v86.edx = drive;
436    v86int();
437
438    if (!V86_CY(v86.efl) &&				/* carry clear */
439	((v86.edx & 0xff) != (drive & DRV_MASK))) {	/* unit # OK */
440	if ((v86.ecx & 0x3f) == 0) {			/* absurd sector size */
441		return(0);				/* skip device */
442	}
443	return (1);
444    }
445    return(0);
446}
447
448/*
449 * We call this when we find a ZFS vdev - ZFS consumes the dsk
450 * structure so we must make a new one.
451 */
452static struct zfsdsk *
453copy_dsk(struct zfsdsk *zdsk)
454{
455    struct zfsdsk *newdsk;
456
457    newdsk = malloc(sizeof(struct zfsdsk));
458    *newdsk = *zdsk;
459    return (newdsk);
460}
461
462/*
463 * Get disk size from eax=0x800 and 0x4800. We need to probe both
464 * because 0x4800 may not be available and we would like to get more
465 * or less correct disk size - if it is possible at all.
466 * Note we do not really want to touch drv.c because that code is shared
467 * with boot2 and we can not afford to grow that code.
468 */
469static uint64_t
470drvsize_ext(struct zfsdsk *zdsk)
471{
472	struct dsk *dskp;
473	uint64_t size, tmp;
474	int cyl, hds, sec;
475
476	dskp = &zdsk->dsk;
477
478	v86.ctl = V86_FLAGS;
479	v86.addr = 0x13;
480	v86.eax = 0x800;
481	v86.edx = dskp->drive;
482	v86int();
483
484	/* Don't error out if we get bad sector number, try EDD as well */
485	if (V86_CY(v86.efl) ||	/* carry set */
486	    (v86.edx & 0xff) <= (unsigned)(dskp->drive & 0x7f)) /* unit # bad */
487		return (0);
488	cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
489	/* Convert max head # -> # of heads */
490	hds = ((v86.edx & 0xff00) >> 8) + 1;
491	sec = v86.ecx & 0x3f;
492
493	size = (uint64_t)cyl * hds * sec;
494
495	/* Determine if we can use EDD with this device. */
496	v86.ctl = V86_FLAGS;
497	v86.addr = 0x13;
498	v86.eax = 0x4100;
499	v86.edx = dskp->drive;
500	v86.ebx = 0x55aa;
501	v86int();
502	if (V86_CY(v86.efl) ||  /* carry set */
503	    (v86.ebx & 0xffff) != 0xaa55 || /* signature */
504	    (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
505		return (size);
506
507	tmp = drvsize(dskp);
508	if (tmp > size)
509		size = tmp;
510
511	return (size);
512}
513
514/*
515 * The "layered" ioctl to read disk/partition size. Unfortunately
516 * the zfsboot case is hardest, because we do not have full software
517 * stack available, so we need to do some manual work here.
518 */
519uint64_t
520ldi_get_size(void *priv)
521{
522	struct zfsdsk *zdsk = priv;
523	uint64_t size = zdsk->dsk.size;
524
525	if (zdsk->dsk.start == 0)
526		size = drvsize_ext(zdsk);
527
528	return (size * DEV_BSIZE);
529}
530
531static void
532probe_drive(struct zfsdsk *zdsk)
533{
534#ifdef GPT
535    struct gpt_hdr hdr;
536    struct gpt_ent *ent;
537    unsigned part, entries_per_sec;
538    daddr_t slba;
539#endif
540#if defined(GPT) || defined(LOADER_GELI_SUPPORT)
541    daddr_t elba;
542#endif
543
544    struct dos_partition *dp;
545    char *sec;
546    unsigned i;
547
548    /*
549     * If we find a vdev on the whole disk, stop here.
550     */
551    if (vdev_probe(vdev_read2, zdsk, NULL) == 0)
552	return;
553
554#ifdef LOADER_GELI_SUPPORT
555    /*
556     * Taste the disk, if it is GELI encrypted, decrypt it and check to see if
557     * it is a usable vdev then. Otherwise dig
558     * out the partition table and probe each slice/partition
559     * in turn for a vdev or GELI encrypted vdev.
560     */
561    elba = drvsize_ext(zdsk);
562    if (elba > 0) {
563	elba--;
564    }
565    zdsk->gdev = geli_taste(vdev_read, zdsk, elba, "disk%u:0:");
566    if (zdsk->gdev != NULL) {
567	if (geli_havekey(zdsk->gdev) == 0 ||
568	    geli_passphrase(zdsk->gdev, gelipw) == 0) {
569	    if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
570		return;
571	    }
572	}
573    }
574#endif /* LOADER_GELI_SUPPORT */
575
576    sec = dmadat->secbuf;
577    zdsk->dsk.start = 0;
578
579#ifdef GPT
580    /*
581     * First check for GPT.
582     */
583    if (drvread(&zdsk->dsk, sec, 1, 1)) {
584	return;
585    }
586    memcpy(&hdr, sec, sizeof(hdr));
587    if (memcmp(hdr.hdr_sig, GPT_HDR_SIG, sizeof(hdr.hdr_sig)) != 0 ||
588	hdr.hdr_lba_self != 1 || hdr.hdr_revision < 0x00010000 ||
589	hdr.hdr_entsz < sizeof(*ent) || DEV_BSIZE % hdr.hdr_entsz != 0) {
590	goto trymbr;
591    }
592
593    /*
594     * Probe all GPT partitions for the presence of ZFS pools. We
595     * return the spa_t for the first we find (if requested). This
596     * will have the effect of booting from the first pool on the
597     * disk.
598     *
599     * If no vdev is found, GELI decrypting the device and try again
600     */
601    entries_per_sec = DEV_BSIZE / hdr.hdr_entsz;
602    slba = hdr.hdr_lba_table;
603    elba = slba + hdr.hdr_entries / entries_per_sec;
604    while (slba < elba) {
605	zdsk->dsk.start = 0;
606	if (drvread(&zdsk->dsk, sec, slba, 1))
607	    return;
608	for (part = 0; part < entries_per_sec; part++) {
609	    ent = (struct gpt_ent *)(sec + part * hdr.hdr_entsz);
610	    if (memcmp(&ent->ent_type, &freebsd_zfs_uuid,
611		     sizeof(uuid_t)) == 0) {
612		zdsk->dsk.start = ent->ent_lba_start;
613		zdsk->dsk.size = ent->ent_lba_end - ent->ent_lba_start + 1;
614		zdsk->dsk.slice = part + 1;
615		zdsk->dsk.part = 255;
616		if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
617		    /*
618		     * This slice had a vdev. We need a new dsk
619		     * structure now since the vdev now owns this one.
620		     */
621		    zdsk = copy_dsk(zdsk);
622		}
623#ifdef LOADER_GELI_SUPPORT
624		else if ((zdsk->gdev = geli_taste(vdev_read, zdsk,
625		    ent->ent_lba_end - ent->ent_lba_start, "disk%up%u:",
626		    zdsk->dsk.unit, zdsk->dsk.slice)) != NULL) {
627		    if (geli_havekey(zdsk->gdev) == 0 ||
628			geli_passphrase(zdsk->gdev, gelipw) == 0) {
629			/*
630			 * This slice has GELI, check it for ZFS.
631			 */
632			if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
633			    /*
634			     * This slice had a vdev. We need a new dsk
635			     * structure now since the vdev now owns this one.
636			     */
637			    zdsk = copy_dsk(zdsk);
638			}
639			break;
640		    }
641		}
642#endif /* LOADER_GELI_SUPPORT */
643	    }
644	}
645	slba++;
646    }
647    return;
648trymbr:
649#endif /* GPT */
650
651    if (drvread(&zdsk->dsk, sec, DOSBBSECTOR, 1))
652	return;
653    dp = (void *)(sec + DOSPARTOFF);
654
655    for (i = 0; i < NDOSPART; i++) {
656	if (!dp[i].dp_typ)
657	    continue;
658	zdsk->dsk.start = dp[i].dp_start;
659	zdsk->dsk.size = dp[i].dp_size;
660	zdsk->dsk.slice = i + 1;
661	if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
662	    zdsk = copy_dsk(zdsk);
663	}
664#ifdef LOADER_GELI_SUPPORT
665	else if ((zdsk->gdev = geli_taste(vdev_read, zdsk, dp[i].dp_size -
666		 dp[i].dp_start, "disk%us%u:")) != NULL) {
667	    if (geli_havekey(zdsk->gdev) == 0 ||
668		geli_passphrase(zdsk->gdev, gelipw) == 0) {
669		/*
670		 * This slice has GELI, check it for ZFS.
671		 */
672		if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
673		    /*
674		     * This slice had a vdev. We need a new dsk
675		     * structure now since the vdev now owns this one.
676		     */
677		    zdsk = copy_dsk(zdsk);
678		}
679		break;
680	    }
681	}
682#endif /* LOADER_GELI_SUPPORT */
683    }
684}
685
686int
687main(void)
688{
689    dnode_phys_t dn;
690    off_t off;
691    struct zfsdsk *zdsk;
692    int autoboot, i;
693    int nextboot;
694    int rc;
695
696    dmadat = (void *)(roundup2(__base + (int32_t)&_end, 0x10000) - __base);
697
698    bios_getmem();
699
700    if (high_heap_size > 0) {
701	heap_end = PTOV(high_heap_base + high_heap_size);
702	heap_next = PTOV(high_heap_base);
703    } else {
704	heap_next = (char *)dmadat + sizeof(*dmadat);
705	heap_end = (char *)PTOV(bios_basemem);
706    }
707    setheap(heap_next, heap_end);
708
709    zdsk = calloc(1, sizeof(struct zfsdsk));
710    zdsk->dsk.drive = *(uint8_t *)PTOV(ARGS);
711    zdsk->dsk.type = zdsk->dsk.drive & DRV_HARD ? TYPE_AD : TYPE_FD;
712    zdsk->dsk.unit = zdsk->dsk.drive & DRV_MASK;
713    zdsk->dsk.slice = *(uint8_t *)PTOV(ARGS + 1) + 1;
714    zdsk->dsk.part = 0;
715    zdsk->dsk.start = 0;
716    zdsk->dsk.size = drvsize_ext(zdsk);
717
718    bootinfo.bi_version = BOOTINFO_VERSION;
719    bootinfo.bi_size = sizeof(bootinfo);
720    bootinfo.bi_basemem = bios_basemem / 1024;
721    bootinfo.bi_extmem = bios_extmem / 1024;
722    bootinfo.bi_memsizes_valid++;
723    bootinfo.bi_bios_dev = zdsk->dsk.drive;
724
725    bootdev = MAKEBOOTDEV(dev_maj[zdsk->dsk.type],
726			  zdsk->dsk.slice, zdsk->dsk.unit, zdsk->dsk.part);
727
728    /* Process configuration file */
729
730    autoboot = 1;
731
732    zfs_init();
733
734    /*
735     * Probe the boot drive first - we will try to boot from whatever
736     * pool we find on that drive.
737     */
738    probe_drive(zdsk);
739
740    /*
741     * Probe the rest of the drives that the bios knows about. This
742     * will find any other available pools and it may fill in missing
743     * vdevs for the boot pool.
744     */
745#ifndef VIRTUALBOX
746    for (i = 0; i < *(unsigned char *)PTOV(BIOS_NUMDRIVES); i++)
747#else
748    for (i = 0; i < MAXBDDEV; i++)
749#endif
750    {
751	if ((i | DRV_HARD) == *(uint8_t *)PTOV(ARGS))
752	    continue;
753
754	if (!int13probe(i | DRV_HARD))
755	    break;
756
757	zdsk = calloc(1, sizeof(struct zfsdsk));
758	zdsk->dsk.drive = i | DRV_HARD;
759	zdsk->dsk.type = zdsk->dsk.drive & TYPE_AD;
760	zdsk->dsk.unit = i;
761	zdsk->dsk.slice = 0;
762	zdsk->dsk.part = 0;
763	zdsk->dsk.start = 0;
764	zdsk->dsk.size = drvsize_ext(zdsk);
765	probe_drive(zdsk);
766    }
767
768    /*
769     * The first discovered pool, if any, is the pool.
770     */
771    spa = spa_get_primary();
772    if (!spa) {
773	printf("%s: No ZFS pools located, can't boot\n", BOOTPROG);
774	for (;;)
775	    ;
776    }
777
778    primary_spa = spa;
779    primary_vdev = spa_get_primary_vdev(spa);
780
781    nextboot = 0;
782    rc  = vdev_read_pad2(primary_vdev, cmd, sizeof(cmd));
783    if (vdev_clear_pad2(primary_vdev))
784	printf("failed to clear pad2 area of primary vdev\n");
785    if (rc == 0) {
786	if (*cmd) {
787	    /*
788	     * We could find an old-style ZFS Boot Block header here.
789	     * Simply ignore it.
790	     */
791	    if (*(uint64_t *)cmd != 0x2f5b007b10c) {
792		/*
793		 * Note that parse() is destructive to cmd[] and we also want
794		 * to honor RBX_QUIET option that could be present in cmd[].
795		 */
796		nextboot = 1;
797		memcpy(cmddup, cmd, sizeof(cmd));
798		if (parse_cmd()) {
799		    printf("failed to parse pad2 area of primary vdev\n");
800		    reboot();
801		}
802		if (!OPT_CHECK(RBX_QUIET))
803		    printf("zfs nextboot: %s\n", cmddup);
804	    }
805	    /* Do not process this command twice */
806	    *cmd = 0;
807	}
808    } else
809	printf("failed to read pad2 area of primary vdev\n");
810
811    /* Mount ZFS only if it's not already mounted via nextboot parsing. */
812    if (zfsmount.spa == NULL &&
813	(zfs_spa_init(spa) != 0 || zfs_mount(spa, 0, &zfsmount) != 0)) {
814	printf("%s: failed to mount default pool %s\n",
815	    BOOTPROG, spa->spa_name);
816	autoboot = 0;
817    } else if (zfs_lookup(&zfsmount, PATH_CONFIG, &dn) == 0 ||
818        zfs_lookup(&zfsmount, PATH_DOTCONFIG, &dn) == 0) {
819	off = 0;
820	zfs_read(spa, &dn, &off, cmd, sizeof(cmd));
821    }
822
823    if (*cmd) {
824	/*
825	 * Note that parse_cmd() is destructive to cmd[] and we also want
826	 * to honor RBX_QUIET option that could be present in cmd[].
827	 */
828	memcpy(cmddup, cmd, sizeof(cmd));
829	if (parse_cmd())
830	    autoboot = 0;
831	if (!OPT_CHECK(RBX_QUIET))
832	    printf("%s: %s\n", PATH_CONFIG, cmddup);
833	/* Do not process this command twice */
834	*cmd = 0;
835    }
836
837    /* Do not risk waiting at the prompt forever. */
838    if (nextboot && !autoboot)
839	reboot();
840
841    /*
842     * Try to exec /boot/loader. If interrupted by a keypress,
843     * or in case of failure, try to load a kernel directly instead.
844     */
845
846    if (autoboot && !*kname) {
847	memcpy(kname, PATH_LOADER, sizeof(PATH_LOADER));
848	if (!keyhit(3)) {
849	    load();
850	    memcpy(kname, PATH_KERNEL, sizeof(PATH_KERNEL));
851	}
852    }
853
854    /* Present the user with the boot2 prompt. */
855
856    for (;;) {
857	if (!autoboot || !OPT_CHECK(RBX_QUIET)) {
858	    printf("\nFreeBSD/x86 boot\n");
859	    if (zfs_rlookup(spa, zfsmount.rootobj, rootname) != 0)
860		printf("Default: %s/<0x%llx>:%s\n"
861		       "boot: ",
862		       spa->spa_name, zfsmount.rootobj, kname);
863	    else if (rootname[0] != '\0')
864		printf("Default: %s/%s:%s\n"
865		       "boot: ",
866		       spa->spa_name, rootname, kname);
867	    else
868		printf("Default: %s:%s\n"
869		       "boot: ",
870		       spa->spa_name, kname);
871	}
872	if (ioctrl & IO_SERIAL)
873	    sio_flush();
874	if (!autoboot || keyhit(5))
875	    getstr(cmd, sizeof(cmd));
876	else if (!autoboot || !OPT_CHECK(RBX_QUIET))
877	    putchar('\n');
878	autoboot = 0;
879	if (parse_cmd())
880	    putchar('\a');
881	else
882	    load();
883    }
884}
885
886/* XXX - Needed for btxld to link the boot2 binary; do not remove. */
887void
888exit(int x)
889{
890    __exit(x);
891}
892
893void
894reboot(void)
895{
896    __exit(0);
897}
898
899static void
900load(void)
901{
902    union {
903	struct exec ex;
904	Elf32_Ehdr eh;
905    } hdr;
906    static Elf32_Phdr ep[2];
907    static Elf32_Shdr es[2];
908    caddr_t p;
909    dnode_phys_t dn;
910    off_t off;
911    uint32_t addr, x;
912    int fmt, i, j;
913
914    if (zfs_lookup(&zfsmount, kname, &dn)) {
915	printf("\nCan't find %s\n", kname);
916	return;
917    }
918    off = 0;
919    if (xfsread(&dn, &off, &hdr, sizeof(hdr)))
920	return;
921    if (N_GETMAGIC(hdr.ex) == ZMAGIC)
922	fmt = 0;
923    else if (IS_ELF(hdr.eh))
924	fmt = 1;
925    else {
926	printf("Invalid %s\n", "format");
927	return;
928    }
929    if (fmt == 0) {
930	addr = hdr.ex.a_entry & 0xffffff;
931	p = PTOV(addr);
932	off = PAGE_SIZE;
933	if (xfsread(&dn, &off, p, hdr.ex.a_text))
934	    return;
935	p += roundup2(hdr.ex.a_text, PAGE_SIZE);
936	if (xfsread(&dn, &off, p, hdr.ex.a_data))
937	    return;
938	p += hdr.ex.a_data + roundup2(hdr.ex.a_bss, PAGE_SIZE);
939	bootinfo.bi_symtab = VTOP(p);
940	memcpy(p, &hdr.ex.a_syms, sizeof(hdr.ex.a_syms));
941	p += sizeof(hdr.ex.a_syms);
942	if (hdr.ex.a_syms) {
943	    if (xfsread(&dn, &off, p, hdr.ex.a_syms))
944		return;
945	    p += hdr.ex.a_syms;
946	    if (xfsread(&dn, &off, p, sizeof(int)))
947		return;
948	    x = *(uint32_t *)p;
949	    p += sizeof(int);
950	    x -= sizeof(int);
951	    if (xfsread(&dn, &off, p, x))
952		return;
953	    p += x;
954	}
955    } else {
956	off = hdr.eh.e_phoff;
957	for (j = i = 0; i < hdr.eh.e_phnum && j < 2; i++) {
958	    if (xfsread(&dn, &off, ep + j, sizeof(ep[0])))
959		return;
960	    if (ep[j].p_type == PT_LOAD)
961		j++;
962	}
963	for (i = 0; i < 2; i++) {
964	    p = PTOV(ep[i].p_paddr & 0xffffff);
965	    off = ep[i].p_offset;
966	    if (xfsread(&dn, &off, p, ep[i].p_filesz))
967		return;
968	}
969	p += roundup2(ep[1].p_memsz, PAGE_SIZE);
970	bootinfo.bi_symtab = VTOP(p);
971	if (hdr.eh.e_shnum == hdr.eh.e_shstrndx + 3) {
972	    off = hdr.eh.e_shoff + sizeof(es[0]) *
973		(hdr.eh.e_shstrndx + 1);
974	    if (xfsread(&dn, &off, &es, sizeof(es)))
975		return;
976	    for (i = 0; i < 2; i++) {
977		memcpy(p, &es[i].sh_size, sizeof(es[i].sh_size));
978		p += sizeof(es[i].sh_size);
979		off = es[i].sh_offset;
980		if (xfsread(&dn, &off, p, es[i].sh_size))
981		    return;
982		p += es[i].sh_size;
983	    }
984	}
985	addr = hdr.eh.e_entry & 0xffffff;
986    }
987    bootinfo.bi_esymtab = VTOP(p);
988    bootinfo.bi_kernelname = VTOP(kname);
989    zfsargs.size = sizeof(zfsargs);
990    zfsargs.pool = zfsmount.spa->spa_guid;
991    zfsargs.root = zfsmount.rootobj;
992    zfsargs.primary_pool = primary_spa->spa_guid;
993#ifdef LOADER_GELI_SUPPORT
994    explicit_bzero(gelipw, sizeof(gelipw));
995    export_geli_boot_data(&zfsargs.gelidata);
996#endif
997    if (primary_vdev != NULL)
998	zfsargs.primary_vdev = primary_vdev->v_guid;
999    else
1000	printf("failed to detect primary vdev\n");
1001    /*
1002     * Note that the zfsargs struct is passed by value, not by pointer.  Code in
1003     * btxldr.S copies the values from the entry stack to a fixed location
1004     * within loader(8) at startup due to the presence of KARGS_FLAGS_EXTARG.
1005     */
1006    __exec((caddr_t)addr, RB_BOOTINFO | (opts & RBX_MASK),
1007	   bootdev,
1008	   KARGS_FLAGS_ZFS | KARGS_FLAGS_EXTARG,
1009	   (uint32_t) spa->spa_guid,
1010	   (uint32_t) (spa->spa_guid >> 32),
1011	   VTOP(&bootinfo),
1012	   zfsargs);
1013}
1014
1015static int
1016zfs_mount_ds(char *dsname)
1017{
1018    uint64_t newroot;
1019    spa_t *newspa;
1020    char *q;
1021
1022    q = strchr(dsname, '/');
1023    if (q)
1024	*q++ = '\0';
1025    newspa = spa_find_by_name(dsname);
1026    if (newspa == NULL) {
1027	printf("\nCan't find ZFS pool %s\n", dsname);
1028	return -1;
1029    }
1030
1031    if (zfs_spa_init(newspa))
1032	return -1;
1033
1034    newroot = 0;
1035    if (q) {
1036	if (zfs_lookup_dataset(newspa, q, &newroot)) {
1037	    printf("\nCan't find dataset %s in ZFS pool %s\n",
1038		    q, newspa->spa_name);
1039	    return -1;
1040	}
1041    }
1042    if (zfs_mount(newspa, newroot, &zfsmount)) {
1043	printf("\nCan't mount ZFS dataset\n");
1044	return -1;
1045    }
1046    spa = newspa;
1047    return (0);
1048}
1049
1050static int
1051parse_cmd(void)
1052{
1053    char *arg = cmd;
1054    char *ep, *p, *q;
1055    const char *cp;
1056    int c, i, j;
1057
1058    while ((c = *arg++)) {
1059	if (c == ' ' || c == '\t' || c == '\n')
1060	    continue;
1061	for (p = arg; *p && *p != '\n' && *p != ' ' && *p != '\t'; p++);
1062	ep = p;
1063	if (*p)
1064	    *p++ = 0;
1065	if (c == '-') {
1066	    while ((c = *arg++)) {
1067		if (c == 'P') {
1068		    if (*(uint8_t *)PTOV(0x496) & 0x10) {
1069			cp = "yes";
1070		    } else {
1071			opts |= OPT_SET(RBX_DUAL) | OPT_SET(RBX_SERIAL);
1072			cp = "no";
1073		    }
1074		    printf("Keyboard: %s\n", cp);
1075		    continue;
1076		} else if (c == 'S') {
1077		    j = 0;
1078		    while ((unsigned int)(i = *arg++ - '0') <= 9)
1079			j = j * 10 + i;
1080		    if (j > 0 && i == -'0') {
1081			comspeed = j;
1082			break;
1083		    }
1084		    /* Fall through to error below ('S' not in optstr[]). */
1085		}
1086		for (i = 0; c != optstr[i]; i++)
1087		    if (i == NOPT - 1)
1088			return -1;
1089		opts ^= OPT_SET(flags[i]);
1090	    }
1091	    ioctrl = OPT_CHECK(RBX_DUAL) ? (IO_SERIAL|IO_KEYBOARD) :
1092		     OPT_CHECK(RBX_SERIAL) ? IO_SERIAL : IO_KEYBOARD;
1093	    if (ioctrl & IO_SERIAL) {
1094	        if (sio_init(115200 / comspeed) != 0)
1095		    ioctrl &= ~IO_SERIAL;
1096	    }
1097	} if (c == '?') {
1098	    dnode_phys_t dn;
1099
1100	    if (zfs_lookup(&zfsmount, arg, &dn) == 0) {
1101		zap_list(spa, &dn);
1102	    }
1103	    return -1;
1104	} else {
1105	    arg--;
1106
1107	    /*
1108	     * Report pool status if the comment is 'status'. Lets
1109	     * hope no-one wants to load /status as a kernel.
1110	     */
1111	    if (!strcmp(arg, "status")) {
1112		spa_all_status();
1113		return -1;
1114	    }
1115
1116	    /*
1117	     * If there is "zfs:" prefix simply ignore it.
1118	     */
1119	    if (strncmp(arg, "zfs:", 4) == 0)
1120		arg += 4;
1121
1122	    /*
1123	     * If there is a colon, switch pools.
1124	     */
1125	    q = strchr(arg, ':');
1126	    if (q) {
1127		*q++ = '\0';
1128		if (zfs_mount_ds(arg) != 0)
1129		    return -1;
1130		arg = q;
1131	    }
1132	    if ((i = ep - arg)) {
1133		if ((size_t)i >= sizeof(kname))
1134		    return -1;
1135		memcpy(kname, arg, i + 1);
1136	    }
1137	}
1138	arg = p;
1139    }
1140    return 0;
1141}
1142