zfsboot.c revision 299660 
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: head/sys/boot/i386/zfsboot/zfsboot.c 299660 2016-05-13 11:20:21Z ngie $");
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 "util.h"
43#include "cons.h"
44#include "bootargs.h"
45#include "paths.h"
46
47#include "libzfs.h"
48
49#define ARGS			0x900
50#define NOPT			14
51#define NDEV			3
52
53#define BIOS_NUMDRIVES		0x475
54#define DRV_HARD		0x80
55#define DRV_MASK		0x7f
56
57#define TYPE_AD			0
58#define TYPE_DA			1
59#define TYPE_MAXHARD		TYPE_DA
60#define TYPE_FD			2
61
62#define DEV_GELIBOOT_BSIZE	4096
63
64extern uint32_t _end;
65
66#ifdef GPT
67static const uuid_t freebsd_zfs_uuid = GPT_ENT_TYPE_FREEBSD_ZFS;
68#endif
69static const char optstr[NOPT] = "DhaCcdgmnpqrsv"; /* Also 'P', 'S' */
70static const unsigned char flags[NOPT] = {
71    RBX_DUAL,
72    RBX_SERIAL,
73    RBX_ASKNAME,
74    RBX_CDROM,
75    RBX_CONFIG,
76    RBX_KDB,
77    RBX_GDB,
78    RBX_MUTE,
79    RBX_NOINTR,
80    RBX_PAUSE,
81    RBX_QUIET,
82    RBX_DFLTROOT,
83    RBX_SINGLE,
84    RBX_VERBOSE
85};
86uint32_t opts;
87
88static const unsigned char dev_maj[NDEV] = {30, 4, 2};
89
90static char cmd[512];
91static char cmddup[512];
92static char kname[1024];
93static char rootname[256];
94static int comspeed = SIOSPD;
95static struct bootinfo bootinfo;
96static uint32_t bootdev;
97static struct zfs_boot_args zfsargs;
98static struct zfsmount zfsmount;
99
100vm_offset_t	high_heap_base;
101uint32_t	bios_basemem, bios_extmem, high_heap_size;
102
103static struct bios_smap smap;
104
105/*
106 * The minimum amount of memory to reserve in bios_extmem for the heap.
107 */
108#define	HEAP_MIN		(3 * 1024 * 1024)
109
110static char *heap_next;
111static char *heap_end;
112
113/* Buffers that must not span a 64k boundary. */
114#define READ_BUF_SIZE		8192
115struct dmadat {
116	char rdbuf[READ_BUF_SIZE];	/* for reading large things */
117	char secbuf[READ_BUF_SIZE];	/* for MBR/disklabel */
118};
119static struct dmadat *dmadat;
120
121void exit(int);
122static void load(void);
123static int parse(void);
124static void bios_getmem(void);
125void *malloc(size_t n);
126void free(void *ptr);
127
128void *
129malloc(size_t n)
130{
131	char *p = heap_next;
132	if (p + n > heap_end) {
133		printf("malloc failure\n");
134		for (;;)
135		    ;
136		/* NOTREACHED */
137		return (0);
138	}
139	heap_next += n;
140	return (p);
141}
142
143void
144free(void *ptr)
145{
146
147	return;
148}
149
150static char *
151strdup(const char *s)
152{
153	char *p = malloc(strlen(s) + 1);
154	strcpy(p, s);
155	return (p);
156}
157
158#ifdef LOADER_GELI_SUPPORT
159#include "geliboot.c"
160static char gelipw[GELI_PW_MAXLEN];
161#endif
162
163#include "zfsimpl.c"
164
165/*
166 * Read from a dnode (which must be from a ZPL filesystem).
167 */
168static int
169zfs_read(spa_t *spa, const dnode_phys_t *dnode, off_t *offp, void *start, size_t size)
170{
171	const znode_phys_t *zp = (const znode_phys_t *) dnode->dn_bonus;
172	size_t n;
173	int rc;
174
175	n = size;
176	if (*offp + n > zp->zp_size)
177		n = zp->zp_size - *offp;
178
179	rc = dnode_read(spa, dnode, *offp, start, n);
180	if (rc)
181		return (-1);
182	*offp += n;
183
184	return (n);
185}
186
187/*
188 * Current ZFS pool
189 */
190static spa_t *spa;
191static spa_t *primary_spa;
192static vdev_t *primary_vdev;
193
194/*
195 * A wrapper for dskread that doesn't have to worry about whether the
196 * buffer pointer crosses a 64k boundary.
197 */
198static int
199vdev_read(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
200{
201	char *p;
202	daddr_t lba, alignlba;
203	off_t alignoff, diff;
204	unsigned int nb, alignnb;
205	struct dsk *dsk = (struct dsk *) priv;
206
207	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
208		return -1;
209
210	p = buf;
211	lba = off / DEV_BSIZE;
212	lba += dsk->start;
213	/* Align reads to 4k else 4k sector GELIs will not decrypt. */
214	alignoff = off & ~ (off_t)(DEV_GELIBOOT_BSIZE - 1);
215	/* Round LBA down to nearest multiple of DEV_GELIBOOT_BSIZE bytes. */
216	alignlba = alignoff / DEV_BSIZE;
217	/*
218	 * The read must be aligned to DEV_GELIBOOT_BSIZE bytes relative to the
219	 * start of the GELI partition, not the start of the actual disk.
220	 */
221	alignlba += dsk->start;
222	diff = (lba - alignlba) * DEV_BSIZE;
223
224	while (bytes > 0) {
225		nb = bytes / DEV_BSIZE;
226		/*
227		 * Ensure that the read size plus the leading offset does not
228		 * exceed the size of the read buffer.
229		 */
230		if (nb > (READ_BUF_SIZE - diff) / DEV_BSIZE)
231			nb = (READ_BUF_SIZE - diff) / DEV_BSIZE;
232		/*
233		 * Round the number of blocks to read up to the nearest multiple
234		 * of DEV_GELIBOOT_BSIZE.
235		 */
236		alignnb = roundup2(nb * DEV_BSIZE + diff, DEV_GELIBOOT_BSIZE)
237		    / DEV_BSIZE;
238
239		if (drvread(dsk, dmadat->rdbuf, alignlba, alignnb))
240			return -1;
241#ifdef LOADER_GELI_SUPPORT
242		/* decrypt */
243		if (is_geli(dsk) == 0) {
244			if (geli_read(dsk, ((alignlba - dsk->start) *
245			    DEV_BSIZE), dmadat->rdbuf, alignnb * DEV_BSIZE))
246				return (-1);
247		}
248#endif
249		memcpy(p, dmadat->rdbuf + diff, nb * DEV_BSIZE);
250		p += nb * DEV_BSIZE;
251		lba += nb;
252		alignlba += alignnb;
253		bytes -= nb * DEV_BSIZE;
254		/* Don't need the leading offset after the first block. */
255		diff = 0;
256	}
257
258	return 0;
259}
260
261static int
262xfsread(const dnode_phys_t *dnode, off_t *offp, void *buf, size_t nbyte)
263{
264    if ((size_t)zfs_read(spa, dnode, offp, buf, nbyte) != nbyte) {
265	printf("Invalid format\n");
266	return -1;
267    }
268    return 0;
269}
270
271static void
272bios_getmem(void)
273{
274    uint64_t size;
275
276    /* Parse system memory map */
277    v86.ebx = 0;
278    do {
279	v86.ctl = V86_FLAGS;
280	v86.addr = 0x15;		/* int 0x15 function 0xe820*/
281	v86.eax = 0xe820;
282	v86.ecx = sizeof(struct bios_smap);
283	v86.edx = SMAP_SIG;
284	v86.es = VTOPSEG(&smap);
285	v86.edi = VTOPOFF(&smap);
286	v86int();
287	if (V86_CY(v86.efl) || (v86.eax != SMAP_SIG))
288	    break;
289	/* look for a low-memory segment that's large enough */
290	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0) &&
291	    (smap.length >= (512 * 1024)))
292	    bios_basemem = smap.length;
293	/* look for the first segment in 'extended' memory */
294	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0x100000)) {
295	    bios_extmem = smap.length;
296	}
297
298	/*
299	 * Look for the largest segment in 'extended' memory beyond
300	 * 1MB but below 4GB.
301	 */
302	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base > 0x100000) &&
303	    (smap.base < 0x100000000ull)) {
304	    size = smap.length;
305
306	    /*
307	     * If this segment crosses the 4GB boundary, truncate it.
308	     */
309	    if (smap.base + size > 0x100000000ull)
310		size = 0x100000000ull - smap.base;
311
312	    if (size > high_heap_size) {
313		high_heap_size = size;
314		high_heap_base = smap.base;
315	    }
316	}
317    } while (v86.ebx != 0);
318
319    /* Fall back to the old compatibility function for base memory */
320    if (bios_basemem == 0) {
321	v86.ctl = 0;
322	v86.addr = 0x12;		/* int 0x12 */
323	v86int();
324
325	bios_basemem = (v86.eax & 0xffff) * 1024;
326    }
327
328    /* Fall back through several compatibility functions for extended memory */
329    if (bios_extmem == 0) {
330	v86.ctl = V86_FLAGS;
331	v86.addr = 0x15;		/* int 0x15 function 0xe801*/
332	v86.eax = 0xe801;
333	v86int();
334	if (!V86_CY(v86.efl)) {
335	    bios_extmem = ((v86.ecx & 0xffff) + ((v86.edx & 0xffff) * 64)) * 1024;
336	}
337    }
338    if (bios_extmem == 0) {
339	v86.ctl = 0;
340	v86.addr = 0x15;		/* int 0x15 function 0x88*/
341	v86.eax = 0x8800;
342	v86int();
343	bios_extmem = (v86.eax & 0xffff) * 1024;
344    }
345
346    /*
347     * If we have extended memory and did not find a suitable heap
348     * region in the SMAP, use the last 3MB of 'extended' memory as a
349     * high heap candidate.
350     */
351    if (bios_extmem >= HEAP_MIN && high_heap_size < HEAP_MIN) {
352	high_heap_size = HEAP_MIN;
353	high_heap_base = bios_extmem + 0x100000 - HEAP_MIN;
354    }
355}
356
357/*
358 * Try to detect a device supported by the legacy int13 BIOS
359 */
360static int
361int13probe(int drive)
362{
363    v86.ctl = V86_FLAGS;
364    v86.addr = 0x13;
365    v86.eax = 0x800;
366    v86.edx = drive;
367    v86int();
368
369    if (!V86_CY(v86.efl) &&				/* carry clear */
370	((v86.edx & 0xff) != (drive & DRV_MASK))) {	/* unit # OK */
371	if ((v86.ecx & 0x3f) == 0) {			/* absurd sector size */
372		return(0);				/* skip device */
373	}
374	return (1);
375    }
376    return(0);
377}
378
379/*
380 * We call this when we find a ZFS vdev - ZFS consumes the dsk
381 * structure so we must make a new one.
382 */
383static struct dsk *
384copy_dsk(struct dsk *dsk)
385{
386    struct dsk *newdsk;
387
388    newdsk = malloc(sizeof(struct dsk));
389    *newdsk = *dsk;
390    return (newdsk);
391}
392
393static void
394probe_drive(struct dsk *dsk)
395{
396#ifdef GPT
397    struct gpt_hdr hdr;
398    struct gpt_ent *ent;
399    unsigned part, entries_per_sec;
400#endif
401#ifdef LOADER_GELI_SUPPORT
402    daddr_t slba, elba;
403#endif
404    struct dos_partition *dp;
405    char *sec;
406    unsigned i;
407
408    /*
409     * If we find a vdev on the whole disk, stop here.
410     */
411    if (vdev_probe(vdev_read, dsk, NULL) == 0)
412	return;
413
414#ifdef LOADER_GELI_SUPPORT
415    /*
416     * Taste the disk, if it is GELI encrypted, decrypt it and check to see if
417     * it is a usable vdev then. Otherwise dig
418     * out the partition table and probe each slice/partition
419     * in turn for a vdev or GELI encrypted vdev.
420     */
421    elba = drvsize(dsk);
422    if (elba > 0) {
423	elba--;
424    }
425    if (geli_taste(vdev_read, dsk, elba) == 0) {
426	if (geli_passphrase(&gelipw, dsk->unit, ':', 0, dsk) == 0) {
427	    if (vdev_probe(vdev_read, dsk, NULL) == 0) {
428		return;
429	    }
430	}
431    }
432#endif /* LOADER_GELI_SUPPORT */
433
434    sec = dmadat->secbuf;
435    dsk->start = 0;
436
437#ifdef GPT
438    /*
439     * First check for GPT.
440     */
441    if (drvread(dsk, sec, 1, 1)) {
442	return;
443    }
444    memcpy(&hdr, sec, sizeof(hdr));
445    if (memcmp(hdr.hdr_sig, GPT_HDR_SIG, sizeof(hdr.hdr_sig)) != 0 ||
446	hdr.hdr_lba_self != 1 || hdr.hdr_revision < 0x00010000 ||
447	hdr.hdr_entsz < sizeof(*ent) || DEV_BSIZE % hdr.hdr_entsz != 0) {
448	goto trymbr;
449    }
450
451    /*
452     * Probe all GPT partitions for the presence of ZFS pools. We
453     * return the spa_t for the first we find (if requested). This
454     * will have the effect of booting from the first pool on the
455     * disk.
456     *
457     * If no vdev is found, GELI decrypting the device and try again
458     */
459    entries_per_sec = DEV_BSIZE / hdr.hdr_entsz;
460    slba = hdr.hdr_lba_table;
461    elba = slba + hdr.hdr_entries / entries_per_sec;
462    while (slba < elba) {
463	dsk->start = 0;
464	if (drvread(dsk, sec, slba, 1))
465	    return;
466	for (part = 0; part < entries_per_sec; part++) {
467	    ent = (struct gpt_ent *)(sec + part * hdr.hdr_entsz);
468	    if (memcmp(&ent->ent_type, &freebsd_zfs_uuid,
469		     sizeof(uuid_t)) == 0) {
470		dsk->start = ent->ent_lba_start;
471		dsk->slice = part + 1;
472		dsk->part = 255;
473		if (vdev_probe(vdev_read, dsk, NULL) == 0) {
474		    /*
475		     * This slice had a vdev. We need a new dsk
476		     * structure now since the vdev now owns this one.
477		     */
478		    dsk = copy_dsk(dsk);
479		}
480#ifdef LOADER_GELI_SUPPORT
481		else if (geli_taste(vdev_read, dsk, ent->ent_lba_end -
482			 ent->ent_lba_start) == 0) {
483		    if (geli_passphrase(&gelipw, dsk->unit, 'p', dsk->slice, dsk) == 0) {
484			/*
485			 * This slice has GELI, check it for ZFS.
486			 */
487			if (vdev_probe(vdev_read, dsk, NULL) == 0) {
488			    /*
489			     * This slice had a vdev. We need a new dsk
490			     * structure now since the vdev now owns this one.
491			     */
492			    dsk = copy_dsk(dsk);
493			}
494			break;
495		    }
496		}
497#endif /* LOADER_GELI_SUPPORT */
498	    }
499	}
500	slba++;
501    }
502    return;
503trymbr:
504#endif /* GPT */
505
506    if (drvread(dsk, sec, DOSBBSECTOR, 1))
507	return;
508    dp = (void *)(sec + DOSPARTOFF);
509
510    for (i = 0; i < NDOSPART; i++) {
511	if (!dp[i].dp_typ)
512	    continue;
513	dsk->start = dp[i].dp_start;
514	dsk->slice = i + 1;
515	if (vdev_probe(vdev_read, dsk, NULL) == 0) {
516	    dsk = copy_dsk(dsk);
517	}
518#ifdef LOADER_GELI_SUPPORT
519	else if (geli_taste(vdev_read, dsk, dp[i].dp_size -
520		 dp[i].dp_start) == 0) {
521	    if (geli_passphrase(&gelipw, dsk->unit, 's', i, dsk) == 0) {
522		/*
523		 * This slice has GELI, check it for ZFS.
524		 */
525		if (vdev_probe(vdev_read, dsk, NULL) == 0) {
526		    /*
527		     * This slice had a vdev. We need a new dsk
528		     * structure now since the vdev now owns this one.
529		     */
530		    dsk = copy_dsk(dsk);
531		}
532		break;
533	    }
534	}
535#endif /* LOADER_GELI_SUPPORT */
536    }
537}
538
539int
540main(void)
541{
542    int autoboot, i;
543    dnode_phys_t dn;
544    off_t off;
545    struct dsk *dsk;
546
547    dmadat = (void *)(roundup2(__base + (int32_t)&_end, 0x10000) - __base);
548
549    bios_getmem();
550
551    if (high_heap_size > 0) {
552	heap_end = PTOV(high_heap_base + high_heap_size);
553	heap_next = PTOV(high_heap_base);
554    } else {
555	heap_next = (char *)dmadat + sizeof(*dmadat);
556	heap_end = (char *)PTOV(bios_basemem);
557    }
558
559    dsk = malloc(sizeof(struct dsk));
560    dsk->drive = *(uint8_t *)PTOV(ARGS);
561    dsk->type = dsk->drive & DRV_HARD ? TYPE_AD : TYPE_FD;
562    dsk->unit = dsk->drive & DRV_MASK;
563    dsk->slice = *(uint8_t *)PTOV(ARGS + 1) + 1;
564    dsk->part = 0;
565    dsk->start = 0;
566    dsk->init = 0;
567
568    bootinfo.bi_version = BOOTINFO_VERSION;
569    bootinfo.bi_size = sizeof(bootinfo);
570    bootinfo.bi_basemem = bios_basemem / 1024;
571    bootinfo.bi_extmem = bios_extmem / 1024;
572    bootinfo.bi_memsizes_valid++;
573    bootinfo.bi_bios_dev = dsk->drive;
574
575    bootdev = MAKEBOOTDEV(dev_maj[dsk->type],
576			  dsk->slice, dsk->unit, dsk->part),
577
578    /* Process configuration file */
579
580    autoboot = 1;
581
582#ifdef LOADER_GELI_SUPPORT
583    geli_init();
584#endif
585    zfs_init();
586
587    /*
588     * Probe the boot drive first - we will try to boot from whatever
589     * pool we find on that drive.
590     */
591    probe_drive(dsk);
592
593    /*
594     * Probe the rest of the drives that the bios knows about. This
595     * will find any other available pools and it may fill in missing
596     * vdevs for the boot pool.
597     */
598#ifndef VIRTUALBOX
599    for (i = 0; i < *(unsigned char *)PTOV(BIOS_NUMDRIVES); i++)
600#else
601    for (i = 0; i < MAXBDDEV; i++)
602#endif
603    {
604	if ((i | DRV_HARD) == *(uint8_t *)PTOV(ARGS))
605	    continue;
606
607	if (!int13probe(i | DRV_HARD))
608	    break;
609
610	dsk = malloc(sizeof(struct dsk));
611	dsk->drive = i | DRV_HARD;
612	dsk->type = dsk->drive & TYPE_AD;
613	dsk->unit = i;
614	dsk->slice = 0;
615	dsk->part = 0;
616	dsk->start = 0;
617	dsk->init = 0;
618	probe_drive(dsk);
619    }
620
621    /*
622     * The first discovered pool, if any, is the pool.
623     */
624    spa = spa_get_primary();
625    if (!spa) {
626	printf("%s: No ZFS pools located, can't boot\n", BOOTPROG);
627	for (;;)
628	    ;
629    }
630
631    primary_spa = spa;
632    primary_vdev = spa_get_primary_vdev(spa);
633
634    if (zfs_spa_init(spa) != 0 || zfs_mount(spa, 0, &zfsmount) != 0) {
635	printf("%s: failed to mount default pool %s\n",
636	    BOOTPROG, spa->spa_name);
637	autoboot = 0;
638    } else if (zfs_lookup(&zfsmount, PATH_CONFIG, &dn) == 0 ||
639        zfs_lookup(&zfsmount, PATH_DOTCONFIG, &dn) == 0) {
640	off = 0;
641	zfs_read(spa, &dn, &off, cmd, sizeof(cmd));
642    }
643
644    if (*cmd) {
645	/*
646	 * Note that parse() is destructive to cmd[] and we also want
647	 * to honor RBX_QUIET option that could be present in cmd[].
648	 */
649	memcpy(cmddup, cmd, sizeof(cmd));
650	if (parse())
651	    autoboot = 0;
652	if (!OPT_CHECK(RBX_QUIET))
653	    printf("%s: %s\n", PATH_CONFIG, cmddup);
654	/* Do not process this command twice */
655	*cmd = 0;
656    }
657
658    /*
659     * Try to exec /boot/loader. If interrupted by a keypress,
660     * or in case of failure, try to load a kernel directly instead.
661     */
662
663    if (autoboot && !*kname) {
664	memcpy(kname, PATH_LOADER_ZFS, sizeof(PATH_LOADER_ZFS));
665	if (!keyhit(3)) {
666	    load();
667	    memcpy(kname, PATH_KERNEL, sizeof(PATH_KERNEL));
668	}
669    }
670
671    /* Present the user with the boot2 prompt. */
672
673    for (;;) {
674	if (!autoboot || !OPT_CHECK(RBX_QUIET)) {
675	    printf("\nFreeBSD/x86 boot\n");
676	    if (zfs_rlookup(spa, zfsmount.rootobj, rootname) != 0)
677		printf("Default: %s/<0x%llx>:%s\n"
678		       "boot: ",
679		       spa->spa_name, zfsmount.rootobj, kname);
680	    else if (rootname[0] != '\0')
681		printf("Default: %s/%s:%s\n"
682		       "boot: ",
683		       spa->spa_name, rootname, kname);
684	    else
685		printf("Default: %s:%s\n"
686		       "boot: ",
687		       spa->spa_name, kname);
688	}
689	if (ioctrl & IO_SERIAL)
690	    sio_flush();
691	if (!autoboot || keyhit(5))
692	    getstr(cmd, sizeof(cmd));
693	else if (!autoboot || !OPT_CHECK(RBX_QUIET))
694	    putchar('\n');
695	autoboot = 0;
696	if (parse())
697	    putchar('\a');
698	else
699	    load();
700    }
701}
702
703/* XXX - Needed for btxld to link the boot2 binary; do not remove. */
704void
705exit(int x)
706{
707}
708
709static void
710load(void)
711{
712    union {
713	struct exec ex;
714	Elf32_Ehdr eh;
715    } hdr;
716    static Elf32_Phdr ep[2];
717    static Elf32_Shdr es[2];
718    caddr_t p;
719    dnode_phys_t dn;
720    off_t off;
721    uint32_t addr, x;
722    int fmt, i, j;
723
724    if (zfs_lookup(&zfsmount, kname, &dn)) {
725	printf("\nCan't find %s\n", kname);
726	return;
727    }
728    off = 0;
729    if (xfsread(&dn, &off, &hdr, sizeof(hdr)))
730	return;
731    if (N_GETMAGIC(hdr.ex) == ZMAGIC)
732	fmt = 0;
733    else if (IS_ELF(hdr.eh))
734	fmt = 1;
735    else {
736	printf("Invalid %s\n", "format");
737	return;
738    }
739    if (fmt == 0) {
740	addr = hdr.ex.a_entry & 0xffffff;
741	p = PTOV(addr);
742	off = PAGE_SIZE;
743	if (xfsread(&dn, &off, p, hdr.ex.a_text))
744	    return;
745	p += roundup2(hdr.ex.a_text, PAGE_SIZE);
746	if (xfsread(&dn, &off, p, hdr.ex.a_data))
747	    return;
748	p += hdr.ex.a_data + roundup2(hdr.ex.a_bss, PAGE_SIZE);
749	bootinfo.bi_symtab = VTOP(p);
750	memcpy(p, &hdr.ex.a_syms, sizeof(hdr.ex.a_syms));
751	p += sizeof(hdr.ex.a_syms);
752	if (hdr.ex.a_syms) {
753	    if (xfsread(&dn, &off, p, hdr.ex.a_syms))
754		return;
755	    p += hdr.ex.a_syms;
756	    if (xfsread(&dn, &off, p, sizeof(int)))
757		return;
758	    x = *(uint32_t *)p;
759	    p += sizeof(int);
760	    x -= sizeof(int);
761	    if (xfsread(&dn, &off, p, x))
762		return;
763	    p += x;
764	}
765    } else {
766	off = hdr.eh.e_phoff;
767	for (j = i = 0; i < hdr.eh.e_phnum && j < 2; i++) {
768	    if (xfsread(&dn, &off, ep + j, sizeof(ep[0])))
769		return;
770	    if (ep[j].p_type == PT_LOAD)
771		j++;
772	}
773	for (i = 0; i < 2; i++) {
774	    p = PTOV(ep[i].p_paddr & 0xffffff);
775	    off = ep[i].p_offset;
776	    if (xfsread(&dn, &off, p, ep[i].p_filesz))
777		return;
778	}
779	p += roundup2(ep[1].p_memsz, PAGE_SIZE);
780	bootinfo.bi_symtab = VTOP(p);
781	if (hdr.eh.e_shnum == hdr.eh.e_shstrndx + 3) {
782	    off = hdr.eh.e_shoff + sizeof(es[0]) *
783		(hdr.eh.e_shstrndx + 1);
784	    if (xfsread(&dn, &off, &es, sizeof(es)))
785		return;
786	    for (i = 0; i < 2; i++) {
787		memcpy(p, &es[i].sh_size, sizeof(es[i].sh_size));
788		p += sizeof(es[i].sh_size);
789		off = es[i].sh_offset;
790		if (xfsread(&dn, &off, p, es[i].sh_size))
791		    return;
792		p += es[i].sh_size;
793	    }
794	}
795	addr = hdr.eh.e_entry & 0xffffff;
796    }
797    bootinfo.bi_esymtab = VTOP(p);
798    bootinfo.bi_kernelname = VTOP(kname);
799    zfsargs.size = sizeof(zfsargs);
800    zfsargs.pool = zfsmount.spa->spa_guid;
801    zfsargs.root = zfsmount.rootobj;
802    zfsargs.primary_pool = primary_spa->spa_guid;
803#ifdef LOADER_GELI_SUPPORT
804    bcopy(gelipw, zfsargs.gelipw, sizeof(zfsargs.gelipw));
805    bzero(gelipw, sizeof(gelipw));
806#else
807    zfsargs.gelipw[0] = '\0';
808#endif
809    if (primary_vdev != NULL)
810	zfsargs.primary_vdev = primary_vdev->v_guid;
811    else
812	printf("failed to detect primary vdev\n");
813    __exec((caddr_t)addr, RB_BOOTINFO | (opts & RBX_MASK),
814	   bootdev,
815	   KARGS_FLAGS_ZFS | KARGS_FLAGS_EXTARG,
816	   (uint32_t) spa->spa_guid,
817	   (uint32_t) (spa->spa_guid >> 32),
818	   VTOP(&bootinfo),
819	   zfsargs);
820}
821
822static int
823zfs_mount_ds(char *dsname)
824{
825    uint64_t newroot;
826    spa_t *newspa;
827    char *q;
828
829    q = strchr(dsname, '/');
830    if (q)
831	*q++ = '\0';
832    newspa = spa_find_by_name(dsname);
833    if (newspa == NULL) {
834	printf("\nCan't find ZFS pool %s\n", dsname);
835	return -1;
836    }
837
838    if (zfs_spa_init(newspa))
839	return -1;
840
841    newroot = 0;
842    if (q) {
843	if (zfs_lookup_dataset(newspa, q, &newroot)) {
844	    printf("\nCan't find dataset %s in ZFS pool %s\n",
845		    q, newspa->spa_name);
846	    return -1;
847	}
848    }
849    if (zfs_mount(newspa, newroot, &zfsmount)) {
850	printf("\nCan't mount ZFS dataset\n");
851	return -1;
852    }
853    spa = newspa;
854    return (0);
855}
856
857static int
858parse(void)
859{
860    char *arg = cmd;
861    char *ep, *p, *q;
862    const char *cp;
863    int c, i, j;
864
865    while ((c = *arg++)) {
866	if (c == ' ' || c == '\t' || c == '\n')
867	    continue;
868	for (p = arg; *p && *p != '\n' && *p != ' ' && *p != '\t'; p++);
869	ep = p;
870	if (*p)
871	    *p++ = 0;
872	if (c == '-') {
873	    while ((c = *arg++)) {
874		if (c == 'P') {
875		    if (*(uint8_t *)PTOV(0x496) & 0x10) {
876			cp = "yes";
877		    } else {
878			opts |= OPT_SET(RBX_DUAL) | OPT_SET(RBX_SERIAL);
879			cp = "no";
880		    }
881		    printf("Keyboard: %s\n", cp);
882		    continue;
883		} else if (c == 'S') {
884		    j = 0;
885		    while ((unsigned int)(i = *arg++ - '0') <= 9)
886			j = j * 10 + i;
887		    if (j > 0 && i == -'0') {
888			comspeed = j;
889			break;
890		    }
891		    /* Fall through to error below ('S' not in optstr[]). */
892		}
893		for (i = 0; c != optstr[i]; i++)
894		    if (i == NOPT - 1)
895			return -1;
896		opts ^= OPT_SET(flags[i]);
897	    }
898	    ioctrl = OPT_CHECK(RBX_DUAL) ? (IO_SERIAL|IO_KEYBOARD) :
899		     OPT_CHECK(RBX_SERIAL) ? IO_SERIAL : IO_KEYBOARD;
900	    if (ioctrl & IO_SERIAL) {
901	        if (sio_init(115200 / comspeed) != 0)
902		    ioctrl &= ~IO_SERIAL;
903	    }
904	} if (c == '?') {
905	    dnode_phys_t dn;
906
907	    if (zfs_lookup(&zfsmount, arg, &dn) == 0) {
908		zap_list(spa, &dn);
909	    }
910	    return -1;
911	} else {
912	    arg--;
913
914	    /*
915	     * Report pool status if the comment is 'status'. Lets
916	     * hope no-one wants to load /status as a kernel.
917	     */
918	    if (!strcmp(arg, "status")) {
919		spa_all_status();
920		return -1;
921	    }
922
923	    /*
924	     * If there is "zfs:" prefix simply ignore it.
925	     */
926	    if (strncmp(arg, "zfs:", 4) == 0)
927		arg += 4;
928
929	    /*
930	     * If there is a colon, switch pools.
931	     */
932	    q = strchr(arg, ':');
933	    if (q) {
934		*q++ = '\0';
935		if (zfs_mount_ds(arg) != 0)
936		    return -1;
937		arg = q;
938	    }
939	    if ((i = ep - arg)) {
940		if ((size_t)i >= sizeof(kname))
941		    return -1;
942		memcpy(kname, arg, i + 1);
943	    }
944	}
945	arg = p;
946    }
947    return 0;
948}
949